Submission 011: workload variant, scheduled_fraction 0.046

.gitignore

@@ -33,3 +33,4 @@ submission_artifacts/*
 catboost_info/
 submission.csv
 !submission_artifacts/weekly_planner.joblib
+!submission_artifacts/workload_planner.joblib

README.md

@@ -1,66 +1,69 @@
-# BatterySwapAI 2026 — temperature-gated identity candidate
+---
+license: mit
+library_name: joblib
+tags:
+- batteryswapai-2026
+- predictive-maintenance
+- survival-analysis
+- scheduling
+---
 
-Local candidate only. It has not been submitted to the evaluator. The publicly proven V05 artifact
-remains preserved in the Hugging Face repository and its history (`1541.3141`, Hub revision
-`dfa8acbc`, artifact SHA-256 `63d71091d7f46e8fb11798b7c2be936cc7e0022954824a4eb98c07dc1bf9b0b3`).
+# BatterySwapAI 2026 — MnesisLab
 
-## Current result
+Causal battery-swap planning. Current submission **011**: hierarchical Wiener
+first-passage reranking plus operational capacity policy, over the byte-preserved
+public V05 base.
 
-The primary candidate combines the V06 full-lookback planner with two independently fitted identity-preserving
-rank signals (censored long-term first-passage AFT and K=8 cross-building trajectory similarity),
-then applies a seasonal temperature forecast only to batteries whose forecast minimum smoothed
-voltage is at most **2.50 V**. The 2.50 V gate is fixed in production and preserves the planner's
-risk multiset within each freshness stratum.
+## Deployed artifact
 
-Leakage-clean outer replay on all 48 official train scenarios:
+`submission_artifacts/workload_planner.joblib` — SHA-256
+`336713ecfec29f070e2209b71906d1016755363d9fd570a80e40040e7ec2db13`
 
-| candidate | mean total | delta vs clean operational reference | first / second half | paired t |
-|---|---:|---:|---:|---:|
-| **temperature <=2.50 V (primary)** | **1534.8431** | **-78.4134** | **-110.9689 / -45.8578** | **-2.3034** |
-| identity-only fallback | 1569.8205 | -43.4359 | -75.3522 / -11.5196 | -1.1667 |
-| clean V05 predictions + V06/V07 policy reference | 1613.2564 | — | — | — |
+| setting | value |
+|---|---|
+| FPT rank residual weight | 0.15 |
+| capacity lookback | 42 days |
+| emergency operational scale | 0.5 |
+| scheduled fraction | 0.046 |
+| capacity weekly limit fraction | 0.95 (public V05 default) |
+| embedded public V05 base | `63d71091d7f46e8fb11798b7c2be936cc7e0022954824a4eb98c07dc1bf9b0b3` |
 
-The primary improved 23 scenarios, tied 12, and worsened 13. It passes the predeclared gate:
-delta <= -40, neither chronological half worse, and paired t <= -2. The unrestricted temperature
-rerank and the tighter 2.40 V gate were both weaker; no post-result grid search is used.
+## Submission history
 
-These local absolute totals are not directly comparable with the public leaderboard. The evidence
-supports the candidate by paired delta against the same clean reference, not by claiming a public
-score. The public V05 result remains the only submitted anchor.
+| # | public score | note |
+|---|---:|---|
+| V05 | 1541.31 | causal anchor, revision `dfa8acbc` |
+| 007 | 1563.95 | identity + temperature rerank |
+| **008** | **1517.22** | FPT residual + emergency repricing |
+| 009 | 1528.06 | FPT residual 0.25, emergency buffer 3 |
+| 010 | pending | weekly margin 0.95 -> 0.98 on the 008 stack |
+| 011 | pending | workload variant, scheduled_fraction 0.046 |
 
 ## Causal and official contract
 
 - Every scenario uses only readings with `end_time <= scenario.start_time`.
-- EOL is reconstructed exactly as the organizer evaluator does: strict `10 < temperature < 30`,
-  daily median, days with fewer than five readings masked, seven-calendar-day rolling median with
-  `min_periods=3`, and first smoothed voltage `<= 2.40 V`.
-- The reconstruction matches all 82 observed train EOL devices exactly; censored devices remain
-  censored.
-- Each outer validation fold rebuilds trajectory, EOL, and model state from outer-train batteries
-  only. Assertions require zero held-out battery, building, trajectory, and EOL overlap.
-- Production feature replay matches all 19,890 frozen causal rows; see
-  `artifacts/identity_feature_parity.json` after running the release checks.
-- Runtime uses `batteryswap_public==0.3.4`, CPU only, no network, and emits every live battery once
-  with a valid plan date.
+- EOL is reconstructed exactly as the evaluator defines it: strict `10 < T < 30`,
+  daily median, days with fewer than five readings masked, seven-calendar-day
+  rolling median with `min_periods=3`, first smoothed voltage `<= 2.40 V`.
+  The reconstruction matches all 82 observed train EOL devices; censored devices
+  remain censored.
+- Runtime uses `batteryswap_public==0.3.4`, CPU only, no network, and emits every
+  live battery exactly once with a valid plan date.
 
-## Artifacts
+## Evidence discipline
 
-- `submission_artifacts/temperature_planner.joblib`: primary candidate and the default in
-  `script.py`.
-- `submission_artifacts/identity_planner.joblib`: identity-only fallback.
-- `submission_artifacts/planner.joblib`: byte-preserved V06 full-lookback base, SHA-256
-  `3cfca2e7dd2c05ddd84f2eca42a484168a1ab4ad3cd454806ef4e687fe8f1569`.
-- During the Hub upload, the public V05 artifact is archived byte-for-byte as
-  `submission_artifacts/v05_planner.joblib` before the local V06 base is placed at
-  `submission_artifacts/planner.joblib`.
+Local totals are not comparable with the public leaderboard, and neither 010 nor 011
+carries a validated local gain: measured against the shipped 008 configuration, 010 is
+-5.25 (paired t = -1.49, worse in the first chronological half) and 011 is a wash locally,
+motivated only by the public/local difference in early-swap pricing. 008 (1517.22) remains
+the anchor. Every artifact passes 57/57 tests with byte-identical replays and 19,890 rows
+well inside the 30-minute budget.
 
-Each candidate manifest records its artifact hash, raw dataset hashes, source hashes, fitted AFT
-parameters, donor-library hashes, temperature coefficient/gate, dependency versions, and the
-unchanged embedded V06 base hash. Exact release commands and evidence are in `REPRODUCIBILITY.md`.
+Local totals are not directly comparable with the public leaderboard; candidates
+are supported by paired deltas against the same clean reference, never by claiming
+a public score in advance.
 
-## External eligibility checks
+## License
 
-Participant code is MIT. The organizer-required `batteryswap_public` package does not declare a
-license in its PyPI metadata, and the official evaluator base includes an NVIDIA container license
-notice. These are recorded in `THIRD_PARTY_LICENSES.md`; organizer confirmation is the only
-remaining user-owned licensing check.
+Participant code is MIT (`LICENSE`). Third-party notices are recorded in
+`THIRD_PARTY_LICENSES.md`; reproduction steps are in `REPRODUCIBILITY.md`.

script.py

@@ -45,7 +45,7 @@ def main() -> None:
     artifact_path = Path(
         os.environ.get(
             "BATTERYSWAP_PLANNER_PATH",
-            "submission_artifacts/weekly_planner.joblib",
+            "submission_artifacts/workload_planner.joblib",
         )
     )
     splits = [

submission_artifacts/workload_planner.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:336713ecfec29f070e2209b71906d1016755363d9fd570a80e40040e7ec2db13
+size 16272000

submission_artifacts/workload_planner.json

@@ -0,0 +1,43 @@
+{
+  "artifact": "submission_artifacts/workload_planner.joblib",
+  "artifact_sha256": "336713ecfec29f070e2209b71906d1016755363d9fd570a80e40040e7ec2db13",
+  "schema_version": 1,
+  "base_artifact": "public/v05:submission_artifacts/planner.joblib",
+  "base_artifact_sha256": "63d71091d7f46e8fb11798b7c2be936cc7e0022954824a4eb98c07dc1bf9b0b3",
+  "configuration": {
+    "eol_voltage": 2.4,
+    "horizon_days": 42.0,
+    "terminal_nonoverlap_weeks": 12,
+    "minimum_weekly_increments": 4,
+    "maximum_state_staleness_days": 7.0,
+    "rank_residual_weight": 0.15,
+    "capacity_lookback_days": 42,
+    "emergency_operational_scale": 0.5,
+    "scheduled_fraction": 0.046,
+    "outer_pool": "leave current target building out",
+    "smoothing": "strict 10<T<30; daily median count>=5; rolling seven-calendar-day median min_periods=3; terminal calendar day strictly before cutoff"
+  },
+  "promotion_evidence": {
+    "official_48": "artifacts/hierarchical_fpt_official.json",
+    "robust_48x27": "artifacts/hierarchical_fpt_grid.cases.json"
+  },
+  "source_sha256": {
+    "src/batteryswapai/hierarchical_fpt.py": "51874f5fde70a2c868e5300cce3648d778279cfdb1d64ee3af5f6a30683eca0b",
+    "src/batteryswapai/identity_ensemble.py": "fa1f9aeaa89fba1bc5fc437c478c876544f2973960db18ef2c83f0042cc9df4a",
+    "src/batteryswapai/competition_planner.py": "a56bfe549a3d6fcfe59217f7bd8a647533a4484fdbb26fe3ef8f1d6c84035a98",
+    "scripts/experiment_hierarchical_fpt.py": "1ce854654813752930194413bcd5dff44984d105493a36c42e10945e23fa6b60",
+    "scripts/evaluate_hierarchical_fpt_grid.py": "90f764da5df050e37af15fc47524c629cce2b021de34b0eab39095383639a51e",
+    "scripts/build_hierarchical_fpt_submission.py": "d931e7b2fd2b220ae7fca8b27e267afd74f9ad31e4609ea9ecdce5f2ae6e07da",
+    "script.py": "3b95b55eed638fbe90f17efc0fea62866ebc6266f6222738263b0d55ab07cecb",
+    "Dockerfile": "84a3f71b2ba664f91ceb8a2980dd5397a002eed02af9fdd264325caf89158573",
+    "requirements.txt": "53d9c582b98b392c0517dc87c78949cd13b99b3f09ebfce601d0c8eadd4f2db3"
+  },
+  "runtime_versions": {
+    "batteryswap_public": "0.3.4",
+    "joblib": "1.5.3",
+    "numpy": "2.2.6",
+    "pandas": "2.3.3",
+    "scipy": "1.14.1",
+    "scikit-learn": "1.7.2"
+  }
+}