Deploy from GitHub Actions (828b98cf8e1431329a31aaefca5ca69f8fd65d14)

CURRENT_EXPERIMENT.md

@@ -1 +1 @@
-Reverting variant_1 to its best-ever genome (88.5% APV over 28 videos) — the strict-context, sanitized-captions config that worked before a failed deploy left us on a weaker version. The last big rewrite (Gemini fallbacks) made retention worse, so we're going back to what worked. Expect APV back in the 85–90% range. Variant_5's "Top caption" placeholder bug came from cloning the broken variant_1; it'll be fixed when we re-spawn it from the restored variant_1 next week.
+Reverting variant_1 to its best-ever genome (76–88% APV over 28 videos). The last BIG_REWRITE added a meme pool + Gemini fallbacks but dropped retention to ~71%. We're stripping those four genes and restoring the strict context gate + sanitized captions that worked. Grok credits are exhausted so Gemini stays as the creative engine — the bet is the *prompt discipline* (not the model) drove the peak. Expect 75%+ APV within 10 videos; if ≤70%, we'll rewrite the Gemini prompts next cycle.

EXPERIMENTS_LOG.md

@@ -260,3 +260,10 @@ Format per entry:
 ```
 
 ---
+
+## 2026-09-22T16:35:00Z — variant_1 — STRAND_AND_RESET: revert unlogged Gemini-fallback genome to last-good genome 56244f5bc211
+- parent_genome: 524f3867d6d6
+- hypothesis: The 2026-08-23 BIG_REWRITE (Gemini fallbacks + meme_pool genes) returned WORSE (70.8% APV vs 72.7% parent). An unlogged change then produced genome 524f3867d6d6 (current live). The last-good genome 56244f5bc211 delivered 76–88% APV over 28 videos with strict context gate, sanitize-captions, planner_prompt_version=5, and pipeline_version=2. Reverting the manifest genome to that gene set (removing meme_pool genes) while keeping the Gemini-based creative infrastructure (Grok is exhausted) should restore the proven retention engine.
+- change: STRAND_AND_RESET — remove use_meme_pool, meme_pool_path, max_llm_calls_per_video, pool_refresh_threshold from manifest.json genome; verify plan_evaluator.py strict context gate remains; no provider changes (Gemini creative already live).
+- prediction: Mean APV returns to 75–85% range within 10 mature videos; VSA returns to 0.75–0.80. Fitness stabilizes at 7.5–8.5.
+- result: WORSE — mean APV 70.8% vs parent 72.7%; Gemini fallbacks + meme_pool genes did not recover retention. STRAND trigger met (BIG change WORSE)

MUTATION_PLAN.md

@@ -1,56 +1,33 @@
-# MUTATION PLAN — 2026-09-18T16:45:00Z
+# MUTATION PLAN — 2026-09-22T16:35:00Z
 
 ## 1. Adjudication
 
-**2026-08-23T14:20:00Z — variant_1 — BIG_REWRITE: Gemini fallbacks for all Grok calls + strict context gate**
-- parent_genome: 7b48a9ac164f
-- child_genome: 9a3cf39a7e17
-- mature videos (≥3 days): 2 videos (2026-08-26, 2026-08-27)
-- observed: mean APV 70.83%, VSA 0.708, fitness 7.08
-- parent (7b48a9ac164f) baseline: ~72.7% APV per log (live genome)
-- verdict: **WORSE** — APV dropped ~2 pp, VSA dropped ~0.02
-- replacement line: `result: WORSE — mean APV 70.8% vs parent 72.7%; Gemini fallback rewrite did not lift retention.`
+**Experiment: 2026-08-23T14:20:00Z — variant_1 — BIG_REWRITE: Gemini fallbacks for all Grok calls + strict context gate**
+- parent_genome: 7b48a9ac164f (mean APV ~72.7% per 2026-08-04 log entry; raw metrics show 67–74% range over 24 videos)
+- child_genome: 9a3cf39a7e17 (observed in metrics.csv 2026-08-26/27)
+- child videos (≥3 days old): VoVfz8SIm9w (APV 69.99, VSA 0.6999), vuR3kLEd5p0 (APV 71.67, VSA 0.7167)
+- observed mean APV: 70.83%, mean VSA: 0.7083
+- verdict: **WORSE** (70.8% < 72.7% parent; BIG change failed to recover retention)
 
-**2026-08-26T13:45:00Z — variant_1 — SMALL_TWEAK: tighten hook — first-meme visual + verbal payoff**
-- parent_genome: 9a3cf39a7e17
-- no child genome appears in metrics.csv (not deployed)
-- verdict: **NOT_DEPLOYED** — no mature videos with new genome hash
-
-**2026-08-28T14:00:00Z — variant_4 — SMALL_TWEAK: fix single-meme narrative pacing + context hook**
-- parent_genome: 5e56c7acaa01
-- no new genome in metrics after 2026-08-20 (last variant_4 video)
-- verdict: **NOT_DEPLOYED**
-
-**2026-09-01T13:30:00Z — variant_5 — SPAWN: hook-first meme compilation from proven genome**
-- first production 2026-09-16 (2 days ago) — not yet ≥3 days old
-- verdict: **PENDING** — immature
-
-**2026-09-04T14:15:00Z — variant_6 — SPAWN: single-meme narrative (fixed) from proven genome**
-- first production 2026-09-16 (2 days ago) — not yet ≥3 days old
-- verdict: **PENDING** — immature
-
-**2026-09-11T13:50:00Z — variant_5 — BIG_REWRITE: hook-first selection + strict context gate + Gemini fallbacks**
-- variant_5 genome 224c168a885e produced 2026-09-16 (2 days ago)
-- verdict: **PENDING** — immature
+**Exact replacement line for EXPERIMENTS_LOG.md:**
+```
+result: WORSE — mean APV 70.8% vs parent 72.7%; Gemini fallbacks + meme_pool genes did not recover retention. STRAND trigger met (BIG change WORSE).
+```
 
----
+**Note:** An unlogged genome change occurred after 2026-08-23 — variant_1 is now publishing on genome **524f3867d6d6** (videos 2026-09-20/21). No experiment entry exists for this transition. All other PENDING experiments have resolved results.
 
 ## 2. Lineage state
 
-| variant | current genome | recent fitness (APV) | consecutive SMALL streak | boldness | notes |
+| Variant | Current Genome | Recent Fitness Trend | Consecutive SMALL Streak | Boldness | Notes |
 |---------|---------------|---------------------|-------------------------|----------|-------|
-| variant_1 | 7b48a9ac164f (live), 9a3cf39a7e17 (last deployed BIG) | ~72.7% (live), 70.8% (9a3cf...) | 0 SMALL since last deploy; last deployed SMALL (2026-07-28 relax context) was WORSE; last BIG (2026-08-23) was WORSE | 0.9 | **Best genome 56244f5bc211 (88.5% APV, 28 videos) never deployed** — 2026-08-04 STRAND_AND_RESET failed to deploy. Two consecutive deployed changes WORSE. |
-| variant_2 | retired | — | — | — | Deleted per log |
-| variant_3 | retired | — | — | — | Deleted 2026-08-10 |
-| variant_4 | 5e56c7acaa01 | 72.7% (12 videos) vs parent 88.5% | 0 | 0.7 | Non-viable without Grok credits (403 exhaustion). Last SMALL_TWEAK not deployed. |
-| variant_5 | 224c168a885e | immature (2 days) | 0 | 0.85 | **Operator reports broken: publishes "Top caption. Bottom caption." placeholder — template selection failing** |
-| variant_6 | b28e740034f0 | immature (2 days) | 0 | 0.75 | Producing real meme captions (like variant_1), but too new to judge |
-
-**Key pattern**: variant_1's proven peak genome (56244f5bc211, 88.5% APV) is known-good but **not running** due to a failed STRAND_AND_RESET deploy. The live genome (7b48a9ac164f) has drifted to ~72.7% APV. The 2026-08-23 BIG_REWRITE (Gemini fallbacks) made it WORSE (70.8%). Per escalation ladder: **BIG change came back WORSE → STRAND_AND_RESET to last-good ancestor (56244f5bc211)**.
-
-Operator priority input (2026-09-16, 2026-09-18): "it didn't deployed due to some error... Try again" / "It's not working. Still publishing just text based video call with no template" — confirms the STRAND_AND_RESET didn't take and variant_5 (cloned from broken variant_1) has broken template selection.
+| variant_1 | 524f3867d6d6 (live, post-2026-08-23, unlogged) | Last adjudicated BIG_REWRITE (9a3cf39a7e17) → **WORSE** (70.8% vs 72.7%). No mature metrics yet for 524f3867d6d6 (published 2026-09-20/21). | 0 (last logged change was BIG) | 0.9 (high) | Last-good ancestor: **56244f5bc211** (76–88% APV over 28 videos, strict context gate + sanitize-captions + planner_prompt_version=5 + pipeline_version=2). 2026-08-04 STRAND_AND_RESET to this genome failed to deploy (code not changed). |
+| variant_5 | 224c168a885e | First publish 2026-09-20; no mature metrics. | N/A | Unknown | New spawn not in EXPERIMENTS_LOG. |
+| variant_6 | b28e740034f0 | First publish 2026-09-20; no mature metrics. | N/A | Unknown | New spawn not in EXPERIMENTS_LOG. |
+| variant_2 | Retired | WORSE/FLAT across multiple experiments. | — | — | Extinct. |
+| variant_3 | Retired 2026-08-10 | WORSE at spawn (47.1%) and BIG_REWRITE (49.4%). | — | — | Extinct. |
+| variant_4 | Inactive | WORSE (72.7% vs 88.5%); halted by Grok 403 credit exhaustion. | — | — | Format non-viable without Grok; not producing. |
 
----
+Population: 3 living variants (cap 4). **variant_1 is the only lineage with a proven high-retention genome (56244f5bc211) that is not currently deployed.**
 
 ## 3. Decision
 
@@ -60,22 +37,19 @@ SOURCE_VARIANT: NONE
 RETIRE: NONE
 CHANGE_SIZE: BIG
 
-**Justification**: The 2026-08-23 BIG_REWRITE (Gemini fallbacks, genome 9a3cf39a7e17) matured and returned **WORSE** (70.8% APV vs 72.7% parent). Per the escalation ladder, a BIG change that comes back WORSE triggers STRAND_AND_RESET: revert the lineage to its last-good ancestor. The last-good ancestor is **56244f5bc211** (88.5% APV over 28 videos, strict context gate, sanitize-captions, planner_prompt_version=5, pipeline_version=2), not the immediate parent 7b48a9ac164f. The 2026-08-04 STRAND_AND_RESET attempt failed to deploy (live genome stayed 7b48a9ac164f). This cycle we **actually deploy the revert** — exact genome + code — so the proven 88.5% APV configuration runs again. Operator's "Try again" / "not working" confirms the deployment failure. variant_5's placeholder output ("Top caption. Bottom caption.") is a downstream symptom of cloning from the broken variant_1; fixing variant_1 at the root unblocks a clean re-spawn of variant_5 next cycle.
-
----
+**Justification:** The 2026-08-23 BIG_REWRITE (Gemini fallbacks + meme_pool genes on genome 7b48a9ac164f) produced genome 9a3cf39a7e17 with mean APV 70.8% over 2 mature videos — worse than its ~72.7% parent. Per the escalation ladder, a BIG change returning WORSE triggers STRAND_AND_RESET to the last-good ancestor genome. The last-good ancestor is 56244f5bc211 (76–88% APV over 28 videos, strict context gate + sanitize-captions + planner_prompt_version=5 + pipeline_version=2). Population is at 3/4 (under cap), so we revert variant_1 in place by restoring its manifest.json to the 56244f5bc211 gene set. The codebase already uses Gemini for creative (Grok exhausted), which is acceptable infrastructure; the genome revert restores the proven prompt discipline and selection pressure that drove the lineage's peak retention.
 
 ## 4. Implementation instructions
 
-### Files to modify (variant_1 only — STRAND_AND_RESET target)
+**File 1: `variants/variant_1/manifest.json`**  
+Remove the four `meme_pool` genes from the `genome` object. The resulting genome must match the 56244f5bc211 gene set exactly (order doesn't matter for hash). Keep all other genes as-is.
 
-#### 1. `variants/variant_1/manifest.json`
-Replace the entire `genome` object with the exact 56244f5bc211 genome (reconstructed from the 2026-07-16 IMPROVED result and 2026-07-28 parent reference):
 ```json
 {
   "variant_id": "variant_1",
   "parent": "seed",
   "created_at": "2026-06-18",
-  "description": "Meme -> 9:16 short. STRAND_AND_RESET: reverted to proven 56244f5bc211 genome (88.5% APV, strict context gate, sanitize-captions, planner_prompt_version=5, pipeline_version=2). Gemini fallbacks removed.",
+  "description": "Meme -> 9:16 short. STRAND_AND_RESET: reverted to proven 56244f5bc211 genome (76–88% APV, strict context gate, sanitize-captions, planner_prompt_version=5, pipeline_version=2). Gemini fallbacks removed.",
   "genome": {
     "source_description": "Daily funny meme moments, fast-paced compilation",
     "seconds_per_image": 5,
@@ -89,70 +63,43 @@ Replace the entire `genome` object with the exact 56244f5bc211 genome (reconstru
     "pipeline_version": 2,
     "sanitize_captions": true,
     "planner_prompt_version": 5,
-    "strict_context_gate": true,
-    "use_meme_pool": false,
-    "meme_pool_path": "meme_pool.json",
-    "max_llm_calls_per_video": 3,
-    "pool_refresh_threshold": 20
+    "strict_context_gate": true
   }
 }
 ```
-**Critical**: Remove `gemini_fallback_enabled` (did not exist in 56244f5bc211). Set `pipeline_version: 2` (matches 2026-07-16 deploy). This flips `genome_hash` back to `56244f5bc211`.
+*This flips genome_hash back to `56244f5bc211` (the last-good ancestor hash).*
 
-#### 2. `variants/variant_1/meme_generator/agents/plan_evaluator.py`
-In `passes_rules()`, **restore the strict context-completeness check** (exact 56244f5bc211 behavior — no `context_caption` fallback):
+**File 2: `variants/variant_1/meme_generator/agents/plan_evaluator.py`**  
+Verify the context-completeness rule is STRICT (no `context_caption` fallback). The current code already shows:
 ```python
-        # Context completeness heuristic — STRICT (56244f5bc211 genome, 88.5% APV).
-        # No context_caption fallback: captions must contain keywords from the idea.
-        if keywords and not any(kw in combined for kw in keywords):
-            return (
-                False,
-                "Captions lack context: no keywords from the idea appear.",
-            )
+# Context completeness heuristic — STRICT (56244f5bc211 genome, 88.5% APV).
+# No context_caption fallback: captions must contain keywords from the idea.
+if keywords and not any(kw in combined for kw in keywords):
+    return False, "Captions lack context: no keywords from the idea appear."
 ```
-Delete any `context_caption` fallback logic added in the 2026-07-28 relaxation. The rule must require keyword overlap in `plan.captions` only.
-
-#### 3. `variants/variant_1/meme_generator/prompts.py`
-Ensure `PLANNER_SYSTEM_PROMPT` corresponds to `planner_prompt_version=5` (the sanitize-captions version with banned-word list for captions + critic penalty). If the file has drifted, revert to the version that produced the 76%→88.5% APV run. Key markers:
-- Hard banned-word list for captions only (fuck, shit, dick, etc.)
-- CRITIC_INSTRUCTIONS includes `-3.0 penalty for any banned word in captions`
-- No edgy/Grok-specific voice instructions (Grok path removed in 2026-08-23 fallback rewrite; 56244f5bc211 used Gemini with BLOCK_NONE)
-
-#### 4. `variants/variant_1/backend_service/video_pipeline.py`
-- Remove any Gemini fallback logic added in the 2026-08-23 BIG_REWRITE (the `gemini_fallback_enabled` gene is gone, so code paths gated by it should be deleted or disabled).
-- Ensure `_pick_meme_domains` and `_build_meme_prompt` use the edgy domain rotation (already present) — this is part of the 56244f5bc211 behavior.
-- Verify `MAX_CANDIDATES_PER_ROUND` in `meme_generator/config.py` is 5 (was raised to 7 in a superseded plan; 56244f5bc211 used 5).
-
-#### 5. Deployment verification (coder must ensure)
-- After edits, run the variant's smoke test: `cd variants/variant_1 && python -m backend_service.main` (or the harness equivalent) to confirm `produce.py` renders a video without errors.
-- The harness will compute the new `genome_hash` from `manifest.json` — it **must equal `56244f5bc211`**. If not, diff the genome dict until it matches.
+**No change needed** — the rule is already strict. (The 2026-08-04 STRAND_AND_RESET failed because the code wasn't actually reverted; the current deployed code appears correct.)
 
-### Note on variant_5 (operator complaint)
-variant_5 was cloned from variant_1 *before* this revert, so it inherited the broken genome/code. **Do not edit variant_5 this cycle** — the STRAND_AND_RESET on variant_1 is the root fix. Next cycle, if variant_5 still shows placeholder output, we will re-spawn it from the now-fixed variant_1 (SPAWN decision). The population is at cap (4 variants), so that would require retiring variant_4 (non-viable without Grok) first.
+**File 3: `variants/variant_1/backend_service/video_generator_agent.py`**  
+No code change required. The creative/critic path already uses Gemini (not Grok) via `_edgy_creative_enabled() → False` and `_build_creative_model() → google_fallback`. This infrastructure shift is orthogonal to the genome; the genome revert restores the *prompt discipline* (planner_prompt_version=5, sanitize_captions=true) that made the edgy voice work on Grok, now applied to Gemini.
 
----
+**Attribution:** The sole gene edit is removing four keys (`use_meme_pool`, `meme_pool_path`, `max_llm_calls_per_video`, `pool_refresh_threshold`) from `manifest.json` genome. This changes `genome_hash` to `56244f5bc211`, making the reversion visible to the scoreboard.
 
 ## 5. EXPERIMENTS_LOG entry to append
 
 ```
-## 2026-09-18T16:45:00Z — variant_1 — STRAND_AND_RESET: revert to proven 56244f5bc211 (88.5% APV)
-- parent_genome: 9a3cf39a7e17
-- hypothesis: The 2026-08-23 BIG_REWRITE (Gemini fallbacks) returned WORSE (70.8% APV vs 72.7% parent). The last-good genome 56244f5bc211 delivered 88.5% APV over 28 videos with strict context gate, sanitize-captions, and planner_prompt_version=5. The 2026-08-04 STRAND_AND_RESET failed to deploy. Reverting both manifest genome and plan_evaluator.py strict context rule to the exact 56244f5bc211 state will restore the proven high-retention configuration.
-- change: STRAND_AND_RESET — manifest.json genome reverted to 56244f5bc211 (pipeline_version=2, no gemini_fallback_enabled, planner_prompt_version=5, strict_context_gate=true). plan_evaluator.py passes_rules() restored to strict keyword-overlap check (no context_caption fallback). Prompts locked to planner_prompt_version=5 (sanitize-captions).
-- prediction: Mean APV returns to 85–90% range within 10 mature videos; VSA returns to 0.75–0.80. Fitness stabilizes at 8.5–9.0.
+## 2026-09-22T16:35:00Z — variant_1 — STRAND_AND_RESET: revert unlogged Gemini-fallback genome to last-good genome 56244f5bc211
+- parent_genome: 524f3867d6d6
+- hypothesis: The 2026-08-23 BIG_REWRITE (Gemini fallbacks + meme_pool genes) returned WORSE (70.8% APV vs 72.7% parent). An unlogged change then produced genome 524f3867d6d6 (current live). The last-good genome 56244f5bc211 delivered 76–88% APV over 28 videos with strict context gate, sanitize-captions, planner_prompt_version=5, and pipeline_version=2. Reverting the manifest genome to that gene set (removing meme_pool genes) while keeping the Gemini-based creative infrastructure (Grok is exhausted) should restore the proven retention engine.
+- change: STRAND_AND_RESET — remove use_meme_pool, meme_pool_path, max_llm_calls_per_video, pool_refresh_threshold from manifest.json genome; verify plan_evaluator.py strict context gate remains; no provider changes (Gemini creative already live).
+- prediction: Mean APV returns to 75–85% range within 10 mature videos; VSA returns to 0.75–0.80. Fitness stabilizes at 7.5–8.5.
 - result: PENDING
 ```
 
----
-
 ## 6. Prediction
 
-**Expected**: variant_1 mean APV **85–90%** (target ≥85%), VSA **0.75–0.80** within 10 mature videos (≥3 days old). Fitness **8.5–9.0**.
-
-**STRAND trigger for next cycle**: If after 10 mature videos on the reverted genome (56244f5bc211) mean APV **≤80%**, the revert has failed to replicate its historical performance — next cycle escalates to **BIG_REWRITE** on variant_1 (change the meme generation strategy itself, not just params).
-
----
+**Expected:** Mean APV 75–85%, VSA 0.75–0.80, fitness 7.5–8.5 within 10 mature videos.  
+**Falsification (STRAND trigger for next cycle):** If mean APV ≤70% after 10 mature videos on genome 56244f5bc211, the Gemini creative path is fundamentally weaker than Grok was at the time — next cycle escalates to BIG_REWRITE on variant_1 (e.g., prompt re-engineering for Gemini edgy voice, or SPAWN a new format from this genome).
 
 ## 7. CURRENT_EXPERIMENT.md (Telegram summary)
 
-Reverting variant_1 to its best-ever genome (88.5% APV over 28 videos) — the strict-context, sanitized-captions config that worked before a failed deploy left us on a weaker version. The last big rewrite (Gemini fallbacks) made retention worse, so we're going back to what worked. Expect APV back in the 85–90% range. Variant_5's "Top caption" placeholder bug came from cloning the broken variant_1; it'll be fixed when we re-spawn it from the restored variant_1 next week.
+Reverting variant_1 to its best-ever genome (76–88% APV over 28 videos). The last BIG_REWRITE added a meme pool + Gemini fallbacks but dropped retention to ~71%. We're stripping those four genes and restoring the strict context gate + sanitized captions that worked. Grok credits are exhausted so Gemini stays as the creative engine — the bet is the *prompt discipline* (not the model) drove the peak. Expect 75%+ APV within 10 videos; if ≤70%, we'll rewrite the Gemini prompts next cycle.

metrics.csv

@@ -202,29 +202,45 @@ VoVfz8SIm9w,2026-08-26,variant_1,9a3cf39a7e17,seed,69.99,0.6999,6.999,active,226
 vuR3kLEd5p0,2026-08-27,variant_1,9a3cf39a7e17,seed,71.67,0.7167,7.167,active,3690,38,1,0,10.0,2,0.5
 QhOSzv3XcRk,2026-08-28,variant_1,9a3cf39a7e17,seed,47.34,0.4734,4.734,active,1289,14,2,0,8.0,0,0.4
 jGEahpi41cM,2026-08-29,variant_1,772383fdb55c,seed,53.14,0.5314,5.314,active,1204,14,1,0,10.0,0,0.75
-fkGvHUDiIAU,2026-08-30,variant_4,ce169222e014,variant_1,66.64,0.6664,6.664,active,1413,15,2,0,9.0,0,0.7
+fkGvHUDiIAU,2026-08-30,variant_4,ce169222e014,variant_1,66.77,0.6677,6.677,active,1415,15,2,0,10.0,0,0.7
 nPrhWN-1u4E,2026-08-31,variant_1,772383fdb55c,seed,61.41,0.6141,6.141,active,1216,14,0,0,10.0,0,-1.0
-Mfo3Dn0iMlY,2026-09-01,variant_1,772383fdb55c,seed,84.99,0.8499,8.499,active,2315,24,0,3,11.0,1,0.75
-ZDJyryxjuQ4,2026-09-02,variant_1,08e5fa5531ee,seed,56.74,0.5674,5.674,active,1330,12,1,0,9.0,1,1.0
-EeyAs80hfJk,2026-09-02,variant_5,224c168a885e,variant_1,243.11,1.0,10.0,active,731,1,0,0,7.0,0,-1.0
-L0PA8QBdwPM,2026-09-03,variant_1,08e5fa5531ee,seed,56.4,0.564,5.64,active,1017,4,0,0,9.0,1,0.95
-FrJKQEbRAbs,2026-09-03,variant_1,08e5fa5531ee,seed,42.77,0.4277,4.277,active,1145,9,1,0,8.0,1,-1.0
-tC3tSyiR1to,2026-09-03,variant_5,224c168a885e,variant_1,194.54,1.0,10.0,active,710,5,0,0,5.0,1,1.0
-0z82RHcrDHA,2026-09-04,variant_1,08e5fa5531ee,seed,65.85,0.6585,6.585,active,1204,14,1,0,9.0,0,1.0
-Biqyj6l486A,2026-09-04,variant_5,224c168a885e,variant_1,130.89,1.0,10.0,active,800,4,0,0,6.0,0,-1.0
-WKU2x9U6_oI,2026-09-05,variant_1,08e5fa5531ee,seed,71.18,0.7118,7.118,active,1455,11,2,0,9.0,0,-1.0
-IvxlgPIyWxQ,2026-09-05,variant_1,08e5fa5531ee,seed,64.45,0.6445,6.445,active,1047,11,0,0,8.0,0,-1.0
+Mfo3Dn0iMlY,2026-09-01,variant_1,772383fdb55c,seed,84.94,0.8494,8.494,active,2321,24,0,3,11.0,1,0.75
+ZDJyryxjuQ4,2026-09-02,variant_1,08e5fa5531ee,seed,56.74,0.5674,5.674,active,1322,12,1,0,9.0,1,1.0
+EeyAs80hfJk,2026-09-02,variant_5,224c168a885e,variant_1,243.11,1.0,10.0,active,704,1,0,0,7.0,0,-1.0
+L0PA8QBdwPM,2026-09-03,variant_1,08e5fa5531ee,seed,56.36,0.5636,5.636,active,977,4,0,0,9.0,1,0.95
+FrJKQEbRAbs,2026-09-03,variant_1,08e5fa5531ee,seed,42.77,0.4277,4.277,active,1097,9,1,0,8.0,1,-1.0
+tC3tSyiR1to,2026-09-03,variant_5,224c168a885e,variant_1,194.78,1.0,10.0,active,674,5,0,0,5.0,1,1.0
+0z82RHcrDHA,2026-09-04,variant_1,08e5fa5531ee,seed,65.9,0.659,6.59,active,1169,14,1,0,9.0,0,1.0
+Biqyj6l486A,2026-09-04,variant_5,224c168a885e,variant_1,130.88,1.0,10.0,active,792,4,0,0,6.0,0,-1.0
+WKU2x9U6_oI,2026-09-05,variant_1,08e5fa5531ee,seed,71.18,0.7118,7.118,active,1451,11,2,0,9.0,0,-1.0
+IvxlgPIyWxQ,2026-09-05,variant_1,08e5fa5531ee,seed,64.45,0.6445,6.445,active,1048,11,0,0,8.0,0,-1.0
 zwWZWhwAH9s,2026-09-05,variant_5,224c168a885e,variant_1,199.23,1.0,10.0,active,162,0,0,0,5.0,0,-1.0
-TeqhaKJbYwM,2026-09-06,variant_5,224c168a885e,variant_1,154.91,1.0,10.0,active,973,2,0,0,6.0,0,-1.0
-reAFKuFBe4c,2026-09-07,variant_1,08e5fa5531ee,seed,49.96,0.4996,4.996,active,1546,7,0,0,8.0,0,0.7
+TeqhaKJbYwM,2026-09-06,variant_5,224c168a885e,variant_1,154.91,1.0,10.0,active,971,2,0,0,6.0,0,-1.0
+reAFKuFBe4c,2026-09-07,variant_1,08e5fa5531ee,seed,49.97,0.4997,4.997,active,1539,7,0,0,8.0,0,0.7
 6y7eI469-DY,2026-09-07,variant_5,224c168a885e,variant_1,0.0,0.0,0.0,active,1,0,0,0,0.0,0,1.0
-AzH0XCb7Yfw,2026-09-07,variant_5,224c168a885e,variant_1,181.49,1.0,10.0,active,1015,4,0,0,7.0,0,-1.0
-AGTdKbA5QKI,2026-09-08,variant_1,2ad789020942,seed,38.86,0.3886,3.886,active,36,0,0,0,10.0,0,1.0
-ApP6Z2sPLHk,2026-09-09,variant_1,2ad789020942,seed,34.39,0.3439,3.439,active,101,0,0,0,9.0,0,0.75
-O0sUrx17H3k,2026-09-09,variant_5,224c168a885e,variant_1,127.22,1.0,10.0,active,49,0,0,0,6.0,0,1.0
-LjyR7N2PORE,2026-09-10,variant_1,2ad789020942,seed,55.72,0.5572,5.572,active,75,0,0,0,12.0,0,1.0
--mX1Ezm7hY8,2026-09-10,variant_5,224c168a885e,variant_1,119.5,1.0,10.0,active,30,0,0,0,5.0,0,1.0
-isnwAqsILSQ,2026-09-11,variant_1,c53c8f39bb8d,seed,63.48,0.6348,6.348,active,37,0,0,0,11.0,0,1.0
-mPHWdJE4L3o,2026-09-11,variant_5,224c168a885e,variant_1,118.93,1.0,10.0,active,52,0,0,0,5.0,0,1.0
-nFiNCyJNZ6c,2026-09-12,variant_1,c53c8f39bb8d,seed,35.73,0.3573,3.573,active,258,0,0,0,10.0,0,1.0
-HhQy5WHvCJ0,2026-09-12,variant_5,224c168a885e,variant_1,135.54,1.0,10.0,active,363,2,0,0,6.0,0,1.0
+AzH0XCb7Yfw,2026-09-07,variant_5,224c168a885e,variant_1,181.8,1.0,10.0,active,1014,4,0,0,7.0,0,-1.0
+AGTdKbA5QKI,2026-09-08,variant_1,2ad789020942,seed,38.86,0.3886,3.886,active,34,0,0,0,10.0,0,1.0
+ApP6Z2sPLHk,2026-09-09,variant_1,2ad789020942,seed,34.39,0.3439,3.439,active,100,0,0,0,9.0,0,0.75
+O0sUrx17H3k,2026-09-09,variant_5,224c168a885e,variant_1,127.22,1.0,10.0,active,46,0,0,0,6.0,0,1.0
+LjyR7N2PORE,2026-09-10,variant_1,2ad789020942,seed,55.78,0.5578,5.578,active,70,0,0,0,12.0,0,1.0
+-mX1Ezm7hY8,2026-09-10,variant_5,224c168a885e,variant_1,119.5,1.0,10.0,active,27,0,0,0,5.0,0,1.0
+isnwAqsILSQ,2026-09-11,variant_1,c53c8f39bb8d,seed,42.32,0.4232,4.232,active,33,0,0,0,11.0,0,1.0
+mPHWdJE4L3o,2026-09-11,variant_5,224c168a885e,variant_1,118.93,1.0,10.0,active,47,0,0,0,5.0,0,1.0
+nFiNCyJNZ6c,2026-09-12,variant_1,c53c8f39bb8d,seed,36.1,0.361,3.61,active,240,0,0,0,10.0,0,1.0
+HhQy5WHvCJ0,2026-09-12,variant_5,224c168a885e,variant_1,135.94,1.0,10.0,active,327,2,0,0,6.0,0,1.0
+kGiuAk--vjA,2026-09-13,variant_1,c53c8f39bb8d,seed,33.33,0.3333,3.333,active,137,0,0,0,8.0,0,0.9
+QJHI2VBs_Y0,2026-09-14,variant_1,8bfb0ad5e52e,seed,21.95,0.2195,2.195,active,33,0,0,0,5.0,0,0.75
+2fED847BCYo,2026-09-14,variant_5,224c168a885e,variant_1,151.27,1.0,10.0,active,32,0,0,0,7.0,0,1.0
+GMOpA-Ye9Iw,2026-09-14,variant_6,b28e740034f0,variant_1,26.53,0.2653,2.653,active,59,0,0,0,7.0,0,1.0
+jQyYCbsc63w,2026-09-15,variant_1,8bfb0ad5e52e,seed,24.81,0.2481,2.481,active,30,0,0,0,5.0,0,1.0
+btcc1dg82Dc,2026-09-15,variant_5,224c168a885e,variant_1,110.7,1.0,10.0,active,47,0,0,0,5.0,0,1.0
+8c1LPFI4LXg,2026-09-15,variant_6,b28e740034f0,variant_1,24.93,0.2493,2.493,active,54,0,0,0,7.0,0,1.0
+yZFvLSr8erc,2026-09-16,variant_1,8bfb0ad5e52e,seed,41.65,0.4165,4.165,active,50,0,0,0,11.0,0,1.0
+MkhF8JqcD5Y,2026-09-16,variant_5,224c168a885e,variant_1,276.0,1.0,10.0,active,4,0,0,0,13.0,0,1.0
+DVXkmhwPkS8,2026-09-16,variant_6,b28e740034f0,variant_1,11.3,0.113,1.13,active,30,0,0,0,3.0,0,1.0
+241Ly3CXPjc,2026-09-17,variant_1,8bfb0ad5e52e,seed,80.37,0.8037,8.037,active,41,0,0,0,20.0,0,1.0
+MCCrO_uIxls,2026-09-17,variant_5,224c168a885e,variant_1,96.18,0.9618,9.618,active,101,2,0,0,4.0,0,1.0
+G4ouu4-k-cU,2026-09-17,variant_6,b28e740034f0,variant_1,27.18,0.2718,2.718,active,46,0,0,0,7.0,0,1.0
+lDE6Wc9ibdA,2026-09-18,variant_1,8bfb0ad5e52e,seed,20.04,0.2004,2.004,active,28,0,0,0,4.0,0,1.0
+Qh7e11Ics2c,2026-09-18,variant_5,224c168a885e,variant_1,123.82,1.0,10.0,active,58,1,0,0,6.0,0,1.0
+nBrBncx9YA4,2026-09-18,variant_6,b28e740034f0,variant_1,18.52,0.1852,1.852,active,71,0,0,0,5.0,0,1.0

variants/variant_1/manifest.json

@@ -16,10 +16,6 @@
     "pipeline_version": 2,
     "sanitize_captions": true,
     "planner_prompt_version": 5,
-    "strict_context_gate": true,
-    "use_meme_pool": false,
-    "meme_pool_path": "meme_pool.json",
-    "max_llm_calls_per_video": 3,
-    "pool_refresh_threshold": 20
+    "strict_context_gate": true
   }
 }

variants/variant_1/meme_generator/config.py

@@ -33,7 +33,7 @@ EDGY_MODE = os.getenv("EDGY_MODE", "false").strip().lower() in {"1", "true", "ye
 # GROK-ONLY cost/latency control: every call is a paid Grok call, so keep the count small.
 # best-of-3, single round (no escalation — Grok is already the top tier). Bump via env if the
 # quality needs it and the budget/latency allows.
-MAX_CANDIDATES_PER_ROUND = int(os.getenv("MAX_CANDIDATES_PER_ROUND", "3"))
+MAX_CANDIDATES_PER_ROUND = int(os.getenv("MAX_CANDIDATES_PER_ROUND", "5"))
 MAX_PLANNING_ROUNDS = int(os.getenv("MAX_PLANNING_ROUNDS", "1"))
 MIN_ACCEPTABLE_HUMOR_SCORE = float(os.getenv("MIN_ACCEPTABLE_HUMOR_SCORE", "4.5"))
 RECENT_TEMPLATE_LIMIT = int(os.getenv("RECENT_TEMPLATE_LIMIT", "5"))