id: 84e7fedea08b44349d8c1c247f08795e
parent_id: 10723d70e38a409ab42e7c9c00a88388
item_type: 1
item_id: ec1518e71453490f8fe5ab75cff7180c
item_updated_time: 1787386693072
title_diff: "[{\"diffs\":[[0,\"08-2\"],[-1,\"1\"],[1,\"2\"],[0,\" — c\"],[-1,\"all model\"],[1,\"hampionship, eq_cc\"],[0,\" fix\"],[-1,\"ed\"],[0,\")\"]],\"start1\":27,\"start2\":27,\"length1\":25,\"length2\":32}]"
body_diff: "[{\"diffs\":[[0,\"08-2\"],[-1,\"1\"],[1,\"2\"],[0,\" — c\"],[-1,\"all model fixed, pool v2 stats\"],[1,\"hampionship + eq_cc fix + live adapter\"],[0,\")\\\n\\\n>\"]],\"start1\":29,\"start2\":29,\"length1\":43,\"length2\":51},{\"diffs\":[[0,\"paused, \"],[-1,\"G\"],[1,\"g\"],[0,\"51 binar\"]],\"start1\":89,\"start2\":89,\"length1\":17,\"length2\":17},{\"diffs\":[[0,\"ary \"],[-1,\"unchanged on server (stopped anyway).** Preflop \"],[1,\"now STALE vs master** (\"],[0,\"call\"],[-1,\" \"],[1,\"-\"],[0,\"model \"],[-1,\"structurally fixed; pool personalities recalibrated to human stat ranges. Sanity gates running on g51-with-fix.\\\n\\\n## Punkt 5: Preflop Call Model Fix — DONE (commit a5b6264)\\\n- **Root cause of call-fest**: facing-raise path double-counted matched money as phantom implied odds (all behind at min(vpip,0.5)); opponent ranges capped at 0.50 (matrix supports 0-1) → ~0.30 equity floor under trash → VPIP 0.85-0.94 pool-wide, 99.5% flops\\\n- **Fix**: call range = raiser's PFR (same estimator as reraise path, epf_raise_range_factor removed); implied money only from NOT-yet-matched players at P(call)=max(vpip−pfr,0)/sizing (measured when observer warm, config priors when cold); all arbitrary 0.50 caps → [0.02, 1.0]; call_thresh cap removed\\\n- **4 regression tests** in call_model_tests (trash HU fold, no-phantom-money, 55 set-mine, K8o vs shove); 538 lib tests gre\"],[1,\"+ position + eq_cc fixes + live range collection in tree; live_server rebuilt — resume = data engine now works). **Production sim reference: g51-with-fixes, all fixes uncommitted-to-live.**\\\n\\\n## Championship-Ergebnis (Idee #9, 100K Hände, 2026-08-22)\\\nStation +20.4 / Maniac +5.2 / Tag −1.1 / **G51 −1.3** / G48 −2.3 / Whale −3.1 / **G58 −4.0** / Gen3 −6.0 / Gen2 −7.8\\\n1. **Station gewinnt** — Calling Station schlägt alle smarten Bots im losen Mixed Field. Die Live-Fisch-Feld-Lücke ist messbar: zu viel Bluff in Caller, zu wenig Thin Value. DAS ist der nächste strategische Hebel (Gen 6 mirrors/rollouts oder Thin-Value-Verbesserung).\\\n2. G58 < G51 konsistent (Championship + Isolation: g51 +1.80 vs g58 +1.22 vs G\"],[0,\"en\"],[1,\"2\"],[0,\", \"],[-1,\"52/52 Harrington replays pass\\\n- **Live che\"],[1,\"selber Harness). v22-Modell = kein Promotionskandidat. Isolation-vs-Sanity Zahlenlü\"],[0,\"ck\"],[1,\"e\"],[0,\" (\"],[-1,\"241 hands)**: hero VPIP 32.4%, NOT table-scaled — leak never ignited live (vpip-gated implied odds held). Problem was sim-pool + training-data realism.\\\n\\\n## Punkt 6: Pool stat recalibration — DONE (probe-verified)\\\n- Fish-tier knobs loosened (safety 0.55-0.65, domination 0.03-0.05, npot 0.03-0.04): **VPIP spread now Tag 0.23 / Lag 0.38 / Maniac 0.46 / Whale 0.66 / Fish 0.71 / Station 0.78 / Gen2 0.13** (was 0.85-0.94 all), flop 86.5%\\\n- Per-class targets now TRAINABLE (classes separable in feature space). Machinery ready: opp_name in RawRangeContext, per-class oracle rows in build_range_targets (17d4d21)\\\n- **Next major cycle (v22)**: c\"],[1,\"isolate ~4x höher) unverklärt — Vergleiche NUR innerhalb eines Harness.\\\n3. Generations-Ranking: G51≈G48>G58>Gen3>Gen2.\\\n\\\n## Fixes 2026-08-22 (commits)\\\n- **eq_cc River-Finish** (decision_math): equity_vs_continue war Current-Board-hs → Flush-Draws = High Card am Turn → jedes Paar pass den Gate. Jetzt E[river hs] = hs(1−npot)+(1−hs)ppot. 4_5_99: Flop-Check + Delayed-Turn-Bet (legitime Linie; Test jetzt konditional: Turn-Assertion nur bei echtem Flop-Cbet). 53/53 Replays, 538 Lib-Tests.\\\n- **Live Range C\"],[0,\"ollect\"],[-1,\" on recalibrated pool → per-class soft targets → train → g58 → gates\\\n\\\n## Earlier points (2026-08-20, status unchanged)\\\n- v19 corpus (10.2M) + v21 Gen2-cove\"],[1,\"ion Adapter**: Gen5 puffert showdown_event-Reveals, baut Samples bei on_hand_end (~300-500/h Tag). Live-Config: \"],[0,\"ra\"],[1,\"n\"],[0,\"ge\"],[-1,\" (3.2M): g57 = best new-gen candidate (Gen2 isolation +1.25 vs g55 −0.27, G48 +1.7/TAG +1.5 retained) but FAILED Gate 1 (KQs top-pair passive, 4_5_99 turn barrel) — pool-average labels too wide for described opponents; per-class fix is the answer\\\n- g57nc (corr off): worse everywhere → corrector stays\\\n- Oracle table v19 regenerated (street_sizes index bug fixed, cab5173)\\\n- Training RAM ceiling: 13.4M samples ≈ 80.5GB; RANGE_MAX_SAMPLES cap; u8 one-hot storage would 7x it\\\n\\\n## Gen 6 (renamed): mirrors + rollouts — details in Ideas note f3c42f44.\\\n\\\n## Key lessons (2026-08-19..21)\\\n- Tightness death spiral / call-fest equilibrium: p\"],[1,\"_data_path=/home/jan/gen5_data/live_history/range_data.jsonl. live_server rebuilt.\\\n- VPID-Sensitivität 4_5_99: NEIN — Barrel war VPID-unabhängig (0.12..0.32), Wurzel war der eq_cc-Gate, nicht der Range-Read.\\\n\\\n## Status Modelle\\\n- v22 (range_v22_board): offline besser (tgtNLL 4.8447), im Spiel schlechter als v17. Ursachen-Hypothesen: calErr 0.135, nur 32.7% soft targets, Realismus-Pool ≠ Pool der Sanity-Gegner (G48/TAG sind alte Formula-Bots mit altem Call-Verhalten → Distribution-Mismatch Training-vs-Eval).\\\n- Nächster Trainingszyklus (v23): mehr Samples (Caps hoch), evtl. calErr-gewichtete Selection, gemischtes Corpus (v22 + Anteil alte Formula-Bots für Eval-Distribution).\\\n\\\n## Punkte 1-6 (2026-08-21, kompakt)\\\n- Call-Model-Fix a5b6264 + Position-Ranges 7aa9ad1: P\"],[0,\"ool \"],[-1,\"equilibria are structural, not configurable — validate flop rate + per-class stats with probes before collecting\\\n- Symmetric-pool leaks invisible in A/B (Gen2 = only dissimilar opponent, always thinnest margin)\\\n- Live hand histories at /tmp/super_marvin_hand_history.log (hero = seat 0/Bolsa); 241 hands parsed OK\"],[1,\"VPIP 0.13-0.78, Flop 86.5%, 4 Regressionstests\\\n- G48/TAG Sanity-Gates: Ties (Extended: G48 +0.36, TAG −0.33) — Gates für ALLE Kandidaten incl. g51 failend seit Fix; Policy-Entscheidung offen (≥Tie statt strikt positiv?)\\\n- v22-Zyklus: collect→oracle→per-class targets→train→g58; KQs-Test gefixt, 4_5_99 via eq_cc-Fix\\\n\\\n## Gen 6: mirrors + rollouts (Ideen f3c42f44). Station-Befund gibt Gen 6 die stärkste Begründung: strukturelle Exploitation statt Threshold-Tuning.\"],[0,\"\\\n\\\n##\"]],\"start1\":104,\"start2\":104,\"length1\":2731,\"length2\":2714}]"
metadata_diff: {"new":{},"deleted":[]}
encryption_cipher_text: 
encryption_applied: 0
updated_time: 2026-08-22T08:26:27.516Z
created_time: 2026-08-22T08:26:27.516Z
is_locked: 0
type_: 13