{"row_id": "simsat_gedaref_grain_silos_no_material_change", "split": "eval", "images": ["images/baseline/simsat_gedaref_grain_silos_no_material_change_baseline.png", "images/current/simsat_gedaref_grain_silos_no_material_change_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Gedaref, Sudan. Event/context: Gedaref grain silo no-material-change control (2023-2024). Date window: 2024-01-09T08:16:09Z to 2024-06-07T08:16:19Z. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The grain silos, conveyors, and adjacent storage apron remain materially stable between images.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\"], \"uncertainty_factors\": [\"Food-storage hard negative; no visible damage in paired Sentinel-2 imagery.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.95, \"short_rationale\": \"No visible civilian disruption is defensible; discard rather than infer from regional risk.\"}"}], "source_dataset": "Blackline Atlas Sentinel-2 paired capture", "source_event": "Gedaref grain silo no-material-change control (2023-2024)", "event_id": "simsat:gedaref_grain_silos_no_material_change", "location_name": "Gedaref", "country": "Sudan", "modality": "optical-to-optical", "recommended_action": "discard", "source_license": "Copernicus Sentinel data terms / internal derived capture; verify for downstream redistribution"}
{"row_id": "simsat_khan_younis_training_centre_shelling", "split": "eval", "images": ["images/baseline/simsat_khan_younis_training_centre_shelling_baseline.png", "images/current/simsat_khan_younis_training_centre_shelling_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Khan Younis, Gaza Strip, Palestine. Event/context: Khan Younis training centre shelling/damage aftermath (2024). Date window: 2024-01-20T08:31:31Z to 2024-03-05T08:31:32Z. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The current image shows heavy scarring and structural reduction across a civilian shelter/training campus versus baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"debris_field\", \"broad_urban_destruction\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Dense urban context makes individual structure attribution uncertain.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.84, \"short_rationale\": \"Visible macro-scale disruption affects a humanitarian shelter/campus and warrants immediate analyst review.\"}"}], "source_dataset": "Blackline Atlas Sentinel-2 paired capture", "source_event": "Khan Younis training centre shelling/damage aftermath (2024)", "event_id": "simsat:khan_younis_training_centre_shelling", "location_name": "Khan Younis, Gaza Strip", "country": "Palestine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "Copernicus Sentinel data terms / internal derived capture; verify for downstream redistribution"}
{"row_id": "simsat_kramatorsk_filtration_station_signal_soft", "split": "eval", "images": ["images/baseline/simsat_kramatorsk_filtration_station_signal_soft_baseline.png", "images/current/simsat_kramatorsk_filtration_station_signal_soft_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Kramatorsk, Donetsk Oblast, Ukraine. Event/context: Kramatorsk filtration station soft-signal control (2023-2024). Date window: 2023-09-20T08:36:48Z to 2024-04-22T08:36:45Z. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The filtration-station parcel appears materially stable; no clear roof loss, basin breach, or burn scar is visible.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\", \"low_visibility_or_subtle_signal\"], \"uncertainty_factors\": [\"Any signal is too subtle for a confident disruption report at this resolution.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.91, \"short_rationale\": \"The site is civilian water infrastructure, but visible evidence is insufficient for escalation.\"}"}], "source_dataset": "Blackline Atlas Sentinel-2 paired capture", "source_event": "Kramatorsk filtration station soft-signal control (2023-2024)", "event_id": "simsat:kramatorsk_filtration_station_signal_soft", "location_name": "Kramatorsk, Donetsk Oblast", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "discard", "source_license": "Copernicus Sentinel data terms / internal derived capture; verify for downstream redistribution"}
{"row_id": "simsat_mosul_medical_city_broad_damage_ambiguity", "split": "eval", "images": ["images/baseline/simsat_mosul_medical_city_broad_damage_ambiguity_baseline.png", "images/current/simsat_mosul_medical_city_broad_damage_ambiguity_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Mosul, Iraq. Event/context: Mosul medical city broad damage ambiguity check. Date window: 2017-02-11T07:55:46Z to 2017-02-21T07:58:34Z. Modality: optical-to-optical-cloudy.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The image pair suggests broader urban disruption near a medical area, but the exact civilian facility signal is ambiguous.\", \"civilian_disruption_evidence\": [\"broad_urban_destruction\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Damage is broad and ambiguous; individual medical-campus attribution is weak.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.62, \"short_rationale\": \"Visible change may be relevant but is not localized or certain enough for immediate downlink; defer for analyst review.\"}"}], "source_dataset": "Blackline Atlas Sentinel-2 paired capture", "source_event": "Mosul medical city broad damage ambiguity check", "event_id": "simsat:mosul_medical_city_broad_damage_ambiguity", "location_name": "Mosul", "country": "Iraq", "modality": "optical-to-optical-cloudy", "recommended_action": "defer", "source_license": "Copernicus Sentinel data terms / internal derived capture; verify for downstream redistribution"}
{"row_id": "simsat_okhmatdyt_hospital_strike", "split": "eval", "images": ["images/baseline/simsat_okhmatdyt_hospital_strike_baseline.png", "images/current/simsat_okhmatdyt_hospital_strike_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Kyiv, Ukraine. Event/context: Okhmatdyt children's hospital strike aftermath (2024). Date window: 2024-06-30T09:06:31Z to 2024-07-13T09:16:26Z. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The current image shows roof/scar changes on a medical campus compared with a cleaner baseline.\", \"civilian_disruption_evidence\": [\"roof_loss\", \"debris_field\", \"damaged_water_or_power_facility\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Medical-campus damage evidence is subtle at Sentinel-2 resolution.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.78, \"short_rationale\": \"The site is civilian medical infrastructure and the paired images show disruption signals that merit immediate review.\"}"}], "source_dataset": "Blackline Atlas Sentinel-2 paired capture", "source_event": "Okhmatdyt children's hospital strike aftermath (2024)", "event_id": "simsat:okhmatdyt_hospital_strike", "location_name": "Kyiv", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "Copernicus Sentinel data terms / internal derived capture; verify for downstream redistribution"}
{"row_id": "bright_eval_bright_beirut-explosion_0000", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0000_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0000_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows blast_or_crater_scarring around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"blast_or_crater_scarring\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.802, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: blast_or_crater_scarring. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0001", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0001_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0001_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building, damaged_warehouse_block, missing_building_footprint around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"damaged_warehouse_block\", \"missing_building_footprint\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.978, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: collapsed_building, damaged_warehouse_block, missing_building_footprint. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0002", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0002_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0002_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building, damaged_warehouse_block, large_rubble_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"damaged_warehouse_block\", \"large_rubble_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.966, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: collapsed_building, damaged_warehouse_block, large_rubble_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0003", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0003_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0003_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block, damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\", \"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.855, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: damaged_warehouse_block, damaged_port_or_logistics_apron. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0004", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0004_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0004_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows large_rubble_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"large_rubble_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.979, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: large_rubble_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0005", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0005_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0005_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows large_rubble_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"large_rubble_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.83, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: large_rubble_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0006", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0006_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0006_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows large_rubble_field, damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"large_rubble_field\", \"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.922, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: large_rubble_field, damaged_port_or_logistics_apron. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0007", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0007_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0007_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building, damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.972, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: collapsed_building, damaged_port_or_logistics_apron. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0008", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0008_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0008_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows debris_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.804, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: debris_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0009", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0009_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0009_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.873, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: damaged_port_or_logistics_apron. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0010", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0010_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0010_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows missing_building_footprint, damaged_warehouse_block, debris_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"missing_building_footprint\", \"damaged_warehouse_block\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.849, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: missing_building_footprint, damaged_warehouse_block, debris_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0011", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0011_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0011_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.838, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: damaged_warehouse_block. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0012", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0012_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0012_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building, missing_building_footprint, damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.861, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: collapsed_building, missing_building_footprint, damaged_warehouse_block. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0013", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0013_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0013_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows debris_field, collapsed_building around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"debris_field\", \"collapsed_building\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.821, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: debris_field, collapsed_building. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0014", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0014_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0014_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows blast_or_crater_scarring, debris_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"blast_or_crater_scarring\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.83, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: blast_or_crater_scarring, debris_field. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0015", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0015_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0015_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows blast_or_crater_scarring around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"blast_or_crater_scarring\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.956, \"short_rationale\": \"Imagery quality is excellent with clear detail. Visible evidence of disruption: blast_or_crater_scarring. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0016", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0016_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0016_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows blast_or_crater_scarring around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"blast_or_crater_scarring\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.819, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: blast_or_crater_scarring. Evidence strength is strong. Civilian infrastructure type: port_logistics_apron. Action = downlink_now: strong visual evidence of civilian infrastructure disruption with good visibility and tight localization.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "downlink_now", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0019", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0019_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0019_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows missing_building_footprint around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"missing_building_footprint\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.756, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: missing_building_footprint. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0029", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0029_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0029_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.743, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: damaged_port_or_logistics_apron. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0030", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0030_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0030_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows missing_building_footprint around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"missing_building_footprint\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.76, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: missing_building_footprint. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0032", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0032_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0032_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.585, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: collapsed_building. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0036", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0036_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0036_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.669, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: damaged_warehouse_block. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0041", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0041_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0041_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows large_rubble_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"large_rubble_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.717, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: large_rubble_field. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0043", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0043_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0043_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.554, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: damaged_port_or_logistics_apron. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0046", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0046_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0046_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows large_rubble_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"large_rubble_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.622, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: large_rubble_field. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0047", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0047_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0047_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.751, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: damaged_warehouse_block. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0051", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0051_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0051_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.679, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: damaged_warehouse_block. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0052", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0052_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0052_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.691, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: damaged_port_or_logistics_apron. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0058", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0058_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0058_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows missing_building_footprint around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"missing_building_footprint\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.56, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: missing_building_footprint. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0060", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0060_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0060_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows debris_field around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is moderate; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.781, \"short_rationale\": \"Imagery quality is good; features are distinguishable. Visible evidence of disruption: debris_field. Evidence strength is moderate. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0061", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0061_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0061_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows missing_building_footprint around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"missing_building_footprint\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.387, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: missing_building_footprint. Evidence strength is weak. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0063", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0063_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0063_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows collapsed_building around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"collapsed_building\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.516, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: collapsed_building. Evidence strength is weak. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0064", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0064_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0064_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows debris_field, missing_building_footprint, damaged_port_or_logistics_apron around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"debris_field\", \"missing_building_footprint\", \"damaged_port_or_logistics_apron\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.491, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: debris_field, missing_building_footprint, damaged_port_or_logistics_apron. Evidence strength is weak. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_0065", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_0065_baseline.png", "images/current/bright_eval_bright_beirut-explosion_0065_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Current imagery shows damaged_warehouse_block around port_logistics_apron compared with the baseline.\", \"civilian_disruption_evidence\": [\"damaged_warehouse_block\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.409, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Visible evidence of disruption: damaged_warehouse_block. Evidence strength is weak. Civilian infrastructure type: port_logistics_apron. Action = defer: ambiguous, weak, or partial evidence; visibility limitations or modality ambiguity prevent confident triage.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "defer", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0000", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0000_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0000_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.178, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0001", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0001_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0001_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.08, \"short_rationale\": \"Imagery quality is good; features are distinguishable. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0002", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0002_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0002_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.141, \"short_rationale\": \"Imagery quality is good; features are distinguishable. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0003", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0003_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0003_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\", \"unchanged_civilian_site\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.156, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: unchanged_civilian_site. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0004", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0004_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0004_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\", \"unchanged_civilian_site\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.077, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: unchanged_civilian_site. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0005", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0005_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0005_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"sar_speckle_artifact\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.177, \"short_rationale\": \"Imagery quality is fair; some detail is lost. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: sar_speckle_artifact. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0006", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0006_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0006_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"sar_speckle_artifact\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.08, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: sar_speckle_artifact. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0007", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0007_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0007_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"sar_speckle_artifact\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.217, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: sar_speckle_artifact. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0008", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0008_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0008_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.077, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0009", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0009_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0009_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.078, \"short_rationale\": \"Imagery quality is good; features are distinguishable. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0010", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0010_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0010_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"near_conflict_no_damage\", \"no_visible_change\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.209, \"short_rationale\": \"Imagery quality is good; features are distinguishable. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: near_conflict_no_damage. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0011", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0011_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0011_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"low_visibility\", \"modality_mismatch\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.162, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: modality_mismatch. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0012", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0012_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0012_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"low_visibility\", \"modality_mismatch\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.212, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. Bounding box covers most of the tile; localization is weak. Hard negative type: modality_mismatch. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0013", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0013_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0013_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\", \"unchanged_civilian_site\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.233, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: unchanged_civilian_site. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0014", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0014_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0014_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"low_visibility\", \"modality_mismatch\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.206, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: modality_mismatch. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0015", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0015_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0015_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"sar_speckle_artifact\", \"sar_speckle_or_modality_artifact\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Evidence strength is weak; analyst confirmation is recommended.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.119, \"short_rationale\": \"Imagery quality is poor; limited useful detail visible. Apparent changes are likely SAR speckle or modality artifacts, not physical damage. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: sar_speckle_artifact. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "bright_eval_bright_beirut-explosion_neg_0016", "split": "eval", "images": ["images/baseline/bright_eval_bright_beirut-explosion_neg_0016_baseline.png", "images/current/bright_eval_bright_beirut-explosion_neg_0016_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Beirut, Lebanon. Event/context: Beirut port ammonium nitrate explosion (2020-08-04). Date window: pre-event to post-event. Modality: optical-to-SAR.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"No defensible macro-visible civilian disruption is visible between baseline and current imagery.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\", \"unchanged_civilian_site\"], \"uncertainty_factors\": [\"Cross-modality optical-to-SAR comparison can introduce SAR speckle or modality artifacts.\", \"Localization quality is weak or unavailable; reporter output should not be used as precise localization.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.062, \"short_rationale\": \"Imagery quality is fair; some detail is lost. No visible change between baseline and current imagery. Civilian infrastructure type: port_logistics_apron. No defensible bounding box could be assigned. Hard negative type: unchanged_civilian_site. Action = discard: no visible disruption, low-quality imagery, or likely modality/noise artifact.\"}"}], "source_dataset": "GabeT29/BRIGHT-XView2Format via ChrisRPL/satellite-disruption-triage-aux-v2-2", "source_event": "Beirut port ammonium nitrate explosion (2020-08-04)", "event_id": "bright:Beirut port ammonium nitrate explosion (2020-08-04)", "location_name": "Beirut", "country": "Lebanon", "modality": "optical-to-SAR", "recommended_action": "discard", "source_license": "CC-BY-NC-4.0"}
{"row_id": "xbd_ukraine_rubizhne_0000", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0000_baseline.png", "images/current/xbd_ukraine_rubizhne_0000_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'minor-damage': 1, 'destroyed': 5}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0002", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0002_baseline.png", "images/current/xbd_ukraine_rubizhne_0002_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 24, 'minor-damage': 4, 'destroyed': 1}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0003", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0003_baseline.png", "images/current/xbd_ukraine_rubizhne_0003_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'minor-damage': 3, 'destroyed': 3, 'no-damage': 75, 'major-damage': 4}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0004", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0004_baseline.png", "images/current/xbd_ukraine_rubizhne_0004_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 53, 'minor-damage': 6, 'major-damage': 2, 'destroyed': 1}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0005", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0005_baseline.png", "images/current/xbd_ukraine_rubizhne_0005_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains limited building-damage labels, but the tile-level disruption signal is moderate or localized.\", \"civilian_disruption_evidence\": [\"roof_loss\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.72, \"short_rationale\": \"Damage labels are present but not broad enough for a confident immediate action; defer for analyst confirmation. Building label counts: {'no-damage': 42, 'minor-damage': 4}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "defer", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0006", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0006_baseline.png", "images/current/xbd_ukraine_rubizhne_0006_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.86, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 15, 'major-damage': 5}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0009", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0009_baseline.png", "images/current/xbd_ukraine_rubizhne_0009_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'minor-damage': 21, 'no-damage': 195, 'destroyed': 34, 'major-damage': 24}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0010", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0010_baseline.png", "images/current/xbd_ukraine_rubizhne_0010_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 299, 'minor-damage': 22, 'major-damage': 16, 'destroyed': 11}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0011", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0011_baseline.png", "images/current/xbd_ukraine_rubizhne_0011_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 380, 'minor-damage': 43, 'major-damage': 14, 'destroyed': 15}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0012", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0012_baseline.png", "images/current/xbd_ukraine_rubizhne_0012_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.86, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 118, 'minor-damage': 24, 'major-damage': 5}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0013", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0013_baseline.png", "images/current/xbd_ukraine_rubizhne_0013_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains limited building-damage labels, but the tile-level disruption signal is moderate or localized.\", \"civilian_disruption_evidence\": [\"roof_loss\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.72, \"short_rationale\": \"Damage labels are present but not broad enough for a confident immediate action; defer for analyst confirmation. Building label counts: {'no-damage': 79, 'minor-damage': 13, 'major-damage': 1}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "defer", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0014", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0014_baseline.png", "images/current/xbd_ukraine_rubizhne_0014_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 81, 'minor-damage': 12, 'major-damage': 5, 'destroyed': 1}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0015", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0015_baseline.png", "images/current/xbd_ukraine_rubizhne_0015_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'major-damage': 7, 'minor-damage': 23, 'no-damage': 197, 'destroyed': 11}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0016", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0016_baseline.png", "images/current/xbd_ukraine_rubizhne_0016_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 59, 'minor-damage': 7, 'destroyed': 3, 'major-damage': 3}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0017", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0017_baseline.png", "images/current/xbd_ukraine_rubizhne_0017_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"The Rubizhne image pair has no building-damage labels sufficient to support a visible civilian disruption report.\", \"civilian_disruption_evidence\": [], \"negative_evidence\": [\"no_visible_change\"], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"none\", \"recommended_action\": \"discard\", \"confidence\": 0.9, \"short_rationale\": \"No visible building-damage evidence is recorded for this tile; discard to control false positives. Building label counts: {'no-damage': 6}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "discard", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0019", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0019_baseline.png", "images/current/xbd_ukraine_rubizhne_0019_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 128, 'minor-damage': 21, 'destroyed': 8, 'major-damage': 7}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0020", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0020_baseline.png", "images/current/xbd_ukraine_rubizhne_0020_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'major-damage': 17, 'no-damage': 215, 'destroyed': 18, 'minor-damage': 38}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0021", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0021_baseline.png", "images/current/xbd_ukraine_rubizhne_0021_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 48, 'minor-damage': 42, 'destroyed': 45, 'major-damage': 29}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0022", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0022_baseline.png", "images/current/xbd_ukraine_rubizhne_0022_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'no-damage': 91, 'minor-damage': 49, 'major-damage': 28, 'destroyed': 73}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0023", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0023_baseline.png", "images/current/xbd_ukraine_rubizhne_0023_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'destroyed': 110, 'major-damage': 42, 'minor-damage': 46, 'no-damage': 87}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0024", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0024_baseline.png", "images/current/xbd_ukraine_rubizhne_0024_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'destroyed': 189, 'no-damage': 513, 'minor-damage': 32, 'major-damage': 59}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0025", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0025_baseline.png", "images/current/xbd_ukraine_rubizhne_0025_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains multiple major/destroyed building labels consistent with visible civilian building disruption.\", \"civilian_disruption_evidence\": [\"collapsed_building\", \"missing_building_footprint\", \"debris_field\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"high\", \"recommended_action\": \"downlink_now\", \"confidence\": 0.9, \"short_rationale\": \"Aggregated xBD-style labels indicate strong building damage in a civilian urban tile; recommend immediate analyst review. Building label counts: {'destroyed': 70, 'minor-damage': 40, 'no-damage': 243, 'major-damage': 24}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "downlink_now", "source_license": "MIT"}
{"row_id": "xbd_ukraine_rubizhne_0026", "split": "eval", "images": ["images/baseline/xbd_ukraine_rubizhne_0026_baseline.png", "images/current/xbd_ukraine_rubizhne_0026_current.png"], "messages": [{"role": "system", "content": "You are a paired-image civilian disruption reporter for satellite imagery. Compare only the baseline and current images plus provided context. Report macro-visible civilian disruption from conflict, bombardment, explosion, shelling, or related human-caused disruption. Do not identify military assets, troops, weapons, bases, routes, or tactical opportunities. Return exactly one strict JSON object with keys: visible_change_summary, civilian_disruption_evidence, negative_evidence, uncertainty_factors, severity_hint, recommended_action, confidence, short_rationale. recommended_action must be one of discard, defer, downlink_now and must be derived from the evidence."}, {"role": "user", "content": "Task: compare Image 1 (baseline/pre-event) and Image 2 (current/post-event).\nLocation: Rubizhne, Ukraine. Event/context: Russian invasion of Ukraine — Rubizhne civilian building damage. Date window: pre-2022 conflict baseline to post-2022 conflict damage imagery. Modality: optical-to-optical.\nFocus on visible civilian disruption evidence: damaged civilian buildings, ports/logistics, bridges/roads, water/power/medical/aid infrastructure, burn scars, rubble/debris, shelling or explosion effects.\nIf evidence is weak, obscured, cross-modality, seasonal, or not clearly on the civilian site, use defer or discard rather than over-calling.\nReturn strict JSON only."}, {"role": "assistant", "content": "{\"visible_change_summary\": \"Post-event imagery for Rubizhne contains limited building-damage labels, but the tile-level disruption signal is moderate or localized.\", \"civilian_disruption_evidence\": [\"collapsed_building\"], \"negative_evidence\": [], \"uncertainty_factors\": [\"xBD-Ukraine labels are building-level annotations aggregated to a tile-level reporter target.\", \"This row should not be used for tactical localization or individual-building targeting.\"], \"severity_hint\": \"medium\", \"recommended_action\": \"defer\", \"confidence\": 0.72, \"short_rationale\": \"Damage labels are present but not broad enough for a confident immediate action; defer for analyst confirmation. Building label counts: {'major-damage': 1, 'no-damage': 21}.\"}"}], "source_dataset": "xBD-Ukraine local materialization from sda-kr/xbd-ukraine lineage", "source_event": "Russian invasion of Ukraine — Rubizhne civilian building damage", "event_id": "xbd-ukraine:rubizhne", "location_name": "Rubizhne", "country": "Ukraine", "modality": "optical-to-optical", "recommended_action": "defer", "source_license": "MIT"}