id: 5e40d11e78fc4b0dac951b9c69fc866d
parent_id: f6a520f8a6e2406a94d135a0ddf5a66c
item_type: 1
item_id: d105cf38c62146daa0b7339bdc64aba5
item_updated_time: 1781710226395
title_diff: "[]"
body_diff: "[{\"diffs\":[[0,\"marvin (\"],[-1,\"3\"],[1,\"4\"],[0,\" commits\"]],\"start1\":54,\"start2\":54,\"length1\":17,\"length2\":17},{\"diffs\":[[0,\"ng`\\\n\"],[-1,\"   - Server: `infer_bb_seat()` + synthesize BB when pot < sb+bb at decision time\\\n   - Gen1: HU loosening block (any pair, ace, broadway, suited, K-high, connectors ≥7-high)\\\n\\\n2. **`e3104d8`** — `fix: add draw_raise_min_strength and bet_adj_max thresholds`\\\n   - `draw_raise_min_strength` (default 0.0, prod 0.15): prevents pure draws from raising\\\n   - `bet_adj_max` (default 99.0, prod 3.0): caps bet adjustment power factor\\\n\\\n### super-marvin-userscripts (1 commit)\\\n1. **`887f8f6`** — `fix: synthesize missing BB + guard reconstruction re-entry`\\\n   - `forceFlush()`: synthesize BB player_action when SB seen but BB missed\\\n   - Ante handler: reclassify ante as BB when amount matches (≥ sb, ≤ sb×3)\\\n   - `attachMessageObserver()`: skip reconstruction on re-attach\"],[1,\"2. **`e3104d8`** — `fix: add draw_raise_min_strength and bet_adj_max thresholds`\\\n3. **`e7709cb`** — `refactor: expand blind reconciliation + add sweep utility scripts`\\\n\\\n### super-marvin-userscripts (1 commit)\\\n1. **`887f8f6`** — `fix: synthesize missing BB + guard reconstruction re-entry`\"],[0,\"\\\n\\\n##\"]],\"start1\":148,\"start2\":148,\"length1\":768,\"length2\":296},{\"diffs\":[[0,\"ue |\"],[-1,\" Root Cause |\"],[0,\" Fix\"]],\"start1\":464,\"start2\":464,\"length1\":21,\"length2\":8},{\"diffs\":[[0,\"--|-----\"],[-1,\"------|-----\"],[0,\"|\\\n| BB p\"]],\"start1\":481,\"start2\":481,\"length1\":28,\"length2\":16},{\"diffs\":[[0,\"ed |\"],[-1,\" Torn message missing \\\"big blind\\\" keyword |\"],[0,\" Use\"]],\"start1\":514,\"start2\":514,\"length1\":51,\"length2\":8},{\"diffs\":[[0,\"ipt \"],[-1,\"+ server \"],[0,\"BB s\"]],\"start1\":526,\"start2\":526,\"length1\":17,\"length2\":8},{\"diffs\":[[0,\"sis \"],[-1,\"|\\\n| Wrong pot/cur_bet/num_raises | BB missing → cascade | Same fix\"],[1,\"+ server reconcile_missing_blinds\"],[0,\" |\\\n|\"]],\"start1\":539,\"start2\":539,\"length1\":74,\"length2\":41},{\"diffs\":[[0,\"Gen1\"],[-1,\" no HU handling |\"],[0,\" HU \"]],\"start1\":602,\"start2\":602,\"length1\":25,\"length2\":8},{\"diffs\":[[0,\"83 |\"],[-1,\" No min strength for draw raises |\"],[0,\" `dr\"]],\"start1\":651,\"start2\":651,\"length1\":42,\"length2\":8},{\"diffs\":[[0,\"_adj\"],[-1,\" = 4.0` crushed value betting | `bet_adj_max = 3.0` |\\\n| Duplicate game_start | Reconstruction replayed on re-connect | `initialized` guard |\\\n\\\n## Sweep Results\\\n\\\n### Big Swing (30K, seed=42)\\\nValue-Bet variant (+0.71) won; Aggressive (-0.90) worst. Direction: value betting improvements.\\\n\\\n### Focused (30K × 3 seeds)\\\nHigh variance across seeds. `bam=1.5+low_bets` averaged +0.64 BB/100.\\\n\\\n### 100K Definitive Test (running)\\\nTesting bam=2.0, bam=1.5, bam=1.5+low_bets at 100K hands for reliable signal.\"],[1,\"_max = 3.0` |\\\n| Duplicate game_start on re-attach | `initialized` guard |\\\n\\\n## Sweep Results & Conclusions\\\n\\\n**Threshold tuning produces NO robust improvement.** Tested `bet_adj_max`, bet thresholds, call thresholds, draw restrictions across 30K × 3 seeds and 100K × 2 seeds. Results:\\\n\\\n| Test | Seed 42 | Seed 137 | Average |\\\n|------|---------|----------|---------|\\\n| bam=1.5+low_bets (30K) | +1.67 | +3.81 | +2.74 |\\\n| bam=1.5+low_bets (100K) | +5.12 | **-4.76** | **+0.18** |\\\n\\\nThe 30K results showed promise but 100K confirmation revealed it was seed-dependent variance. Average across 2 seeds at 100K hands is essentially zero.\\\n\\\n**Root cause of high variance**: With `reset_bankroll_each_hand = true`, BB/100 is dominated by a small number of large-pot outcomes. Parameter changes flip decisions in marginal spots, which cascades into different big-pot results, swinging the delta by ±5 BB/100 per seed.\\\n\\\n**Recommendation**: Stop threshold micro-tuning. Focus on structural improvements:\\\n- Gen3 opponent modeling (range estimation)\\\n- Gen4 RL\\\n- Preflop chart improvements (hand ranges vs heuristics)\\\n\\\n## Production Config (cash_nl.toml) — Current State\\\n```toml\\\nmc_samples = 50000\\\ncall_turn_base = 0.82\\\npower_callers_weight = 0.3\\\ncall_pot_odds_norm = 60.0\\\nraise_cbet_into_raiser = 0.82\\\nnpot_raise_ceiling = 0.35\\\ndraw_raise_min_strength = 0.15   # NEW: prevents pure-draw raises\\\nbet_adj_max = 3.0                # NEW: caps multi-way bet adjustment\\\n```\"],[0,\"\\\n\\\n##\"]],\"start1\":726,\"start2\":726,\"length1\":505,\"length2\":1457},{\"diffs\":[[0,\"otes\"],[-1,\" Cleanup\"],[0,\"\\\n- D\"]],\"start1\":2191,\"start2\":2191,\"length1\":16,\"length2\":8},{\"diffs\":[[0,\"udit\"],[1,\"s\"],[0,\" #4\"],[1,\", #5\"],[0,\" (all \"],[-1,\"data \"],[0,\"issu\"]],\"start1\":2207,\"start2\":2207,\"length1\":22,\"length2\":22},{\"diffs\":[[0,\")\\\n- \"],[-1,\"Deleted audit #5 (executor bug fixed, confirmed working)\\\n- Updated audit #2: QJo HU fold marked\"],[1,\"Audit #2 updated: QJo\"],[0,\" fixed\"],[-1,\";\"],[1,\",\"],[0,\" AQo\"],[-1,\" passive and QTs aggressive remain open (\"],[1,\"/QTs \"],[0,\"part\"]],\"start1\":2237,\"start2\":2237,\"length1\":155,\"length2\":45},{\"diffs\":[[0,\"resholds\"],[-1,\")\"]],\"start1\":2307,\"start2\":2307,\"length1\":9,\"length2\":8}]"
metadata_diff: {"new":{},"deleted":[]}
encryption_cipher_text: 
encryption_applied: 0
updated_time: 2026-06-17T15:37:12.000Z
created_time: 2026-06-17T15:37:12.000Z
type_: 13