id: 4cb7bf22214e46e186544c230b612a86
parent_id: ac46e4e60481424abf9c9f8b0f59d17d
item_type: 1
item_id: ec1518e71453490f8fe5ab75cff7180c
item_updated_time: 1787203030226
title_diff: "[{\"diffs\":[[0,\"-08-\"],[-1,\"19\"],[1,\"20\"],[0,\" — v19 \"],[-1,\"collection\"],[1,\"training\"],[0,\")\"]],\"start1\":26,\"start2\":26,\"length1\":24,\"length2\":22}]"
body_diff: "[{\"diffs\":[[0,\"2026-08-\"],[-1,\"19\"],[1,\"20\"],[0,\" — v19 c\"]],\"start1\":24,\"start2\":24,\"length1\":18,\"length2\":18},{\"diffs\":[[0,\"ion \"],[-1,\"RUNNING)\\\n\\\n> **Live: G51 (unchanged).** v18-e05 trained (parked as range model for v19 collector). **v19 dual-purpose collection running** (8×150K hands, ETA ~8-9h)\"],[1,\"DONE, training running)\\\n\\\n> **Live: paused (user decision 2026-08-19, until NN work lands).** v19 corpus collected; oracle table regenerated; RangeNet v19 A/B training running\"],[0,\".\\\n\\\n#\"]],\"start1\":48,\"start2\":48,\"length1\":171,\"length2\":182},{\"diffs\":[[0,\"nkt \"],[-1,\"3\"],[1,\"1\"],[0,\": v19 \"],[-1,\"Range \"],[0,\"Coll\"]],\"start1\":234,\"start2\":234,\"length1\":21,\"length2\":15},{\"diffs\":[[0,\"ion \"],[-1,\"(RUNNING — started 2026-08-19 14:33)\\\n- **Ziel**: RangeNet v19 mit board-exakten Features (A: 5×52+2×52 One-Hots) + Verteilungs-Targets gegen Überkonzentration; Oracle-Tabelle regeneriert aus denselben Daten (Sizing-Änderung invalidiert alte Tabelle)\\\n- **Recorder dual logging** (commit 1e3144f): `RangeSample.raw` = `{hero_cards[2], opp_\"],[1,\"— COMPLETE (2026-08-20 06:24)\\\n- 10.2M range samples + 4.0M equity samples, 13.4h runtime, ALL tables healthy (98.9-99.9% flop+)\\\n- `/home/jan/gen5_data/equity_v19` (70G): 6× 9-max (all-8-personalities gold pool) + 4× 6-max (3-loose quorum), light collector (corr/eq_cc off)\\\n- Every sample has raw block: hero/board/opp \"],[0,\"card\"],[-1,\"s[2], board[n], actions[(street,code,\"],[1,\" ids (rank*4+suit) + per-action \"],[0,\"chips\"],[-1,\")],\"],[1,\" +\"],[0,\" fac\"]],\"start1\":252,\"start2\":252,\"length1\":394,\"length2\":369},{\"diffs\":[[0,\"hips\"],[-1,\"[4], pot, bb}` — Karten-IDs `rank*4+suit`, Chips roh (Xpot-Re-Bucketing offline). Alte Corpora bleiben deserialisierbar (serde default). `RANGE_RECORDER_MAX_SAMPLES` env cap neu.\\\n- **Sizing-Diversität** (füllt leere medium/large Oracle-Zellen): Lag 0.55/0.35, Maniac 0.75/0.50, neu **g51_tagbig_nn** 0.45/0.33\\\n- **Run**: `scripts/v19_collect.sh` → `/home/jan/gen5_data/equity_v19`, g51-NN-Pool + TagBig, 9/6/3-Max gemischt, Caps range 1.6M / equity 400K pro Tisch, Collector `gen5_equity_collect_v19` (range_v18_e05 + corr + eq_cc)\\\n- **KRITISCHER BUG gefixt** (dc92fc0): `min_raise = 20` (vom alten gen5_equity_collect.sh geerbt) degeneriert das Spiel bei Blinds 500/1000 → 1.1% Flop-Rate, 100x Hand-Rate. **Immer min_raise = BB = 1000.** Validierung nach Fix: Flop 15%, Street-Verteilung {3:45%,4:31%,5:24%} = v6-Referenz, ~5 Hände/s/Tisch = v7_fit-Referenz.\\\n- Call-Realismus des Pools: pro Klasse steil & differenziert (Tag/Reg/Nit 180-380x, Fish/Station 6-7x weak→strong) — KEINE Persönlichkeits-Chirurgie nötig; frühere \\\"2.7:1 flach\\\"-Messung war Aggregations-Artefakt.\\\n- **Nach Abschluss**: (1) merge + train v19 (A/B: board-exakt vs 85-dim, Log-S\"],[1,\"/pot/bb\\\n\\\n## Punkt 2: Oracle Table Regen — COMPLETE (commit cab5173)\\\n- **Bug found+fixed**: street_sizes lookup used board length against 0=preflop..3=river index — all calls/folds bucketed \\\"small\\\", medium/large cells empty since v8. Offset fix → all 4 size contexts populated (medium 4-22K/class/street, large 1.5-8K)\\\n- New: `data/oracle_table_v19.json`. Call steepness now size-dependent: Fish 5.3x(small)→8.1x(large), Tag 43-188x\\\n- Old table (data/oracle_table.json, built on pre-sizing-fix pool) superseded\\\n\\\n## Punkt 3: RangeNet v19 A/B — RUNNING (started 2026-08-20 ~07:00)\\\n- Arm 1 (baseline): 85-dim, eps 0.05, 30 epochs → models/range_v19_85d.safetensors\\\n- Arm 2 (board-exact): RANGE_BOARD_EXACT=1 appends hero 2×52 + board 5×52 one-hots from raw block (85→449 dims) → models/range_v19_board.safetensors\\\n- Both on v19 mixed-size data (fixes v17's 9-max-only limitation)\\\n- Model selection: NLL + marginal calErr (v18 trainer)\\\n- **Next after training**: compare log-s\"],[0,\"core\"],[-1,\"-Metrik), (2) Oracle-Tabelle aus v19-equity_data regenerieren, (3) Verteilungs-Targets via Kontext-Clustering auf raw-Block.\\\n\\\n## Punkt 1: RangeNet v18 (DONE — e05 geparkt)\\\n- NLL+calErr-Checkpoint-Selection, ε-Sweep auf 15M: **e05 gewinnt** (NLL 4.555). v18-e05 verbessert Produktions-Kalibrierung, ist jetzt das Range-Modell \"],[1,\"/calibration → winner becomes g55 candidate → Harrington/KK gates, sanity_check, A/B vs g51\\\n\\\n## Gen 6 (renamed 2026-08-19): Opponent Mirrors + Decision-Time Rollouts\\\n- Old PPO \\\"Gen 6\\\" numbering retired. Gen 6 = Jan's Idea #8 (details in I\"],[0,\"de\"],[1,\"a\"],[0,\"s \"],[-1,\"v19-Collectors.\\\n- Überkonzentration (v17-Krankheit) nur teilweise behoben → v19-Ansätze (board-exakt + Verteilungs-Targets) sind die Fortsetzung.\\\n\\\n## Punkt 2: Oracle-in-Corrector (DONE — parked with finding)\\\n- corr_use_oracle regrediert KK-Gate (Pool-P(call|strength) zu flach vs hand-gefitte 8:1 — Bots sind schlechte Verhaltens-Vorlage für Menschen). Tabelle + Lookup-Maschinerie bleiben für FutureActionNet.\\\n- g54 existiert, kein Kandidat.\\\n\\\n## Session context\\\n- Corpus-Aufräumen durchgeführt 2026-08-19: equity_v6_mc/nocorr/g53audit gelöscht (~43G frei). Behalten: v6, v7_fit, v8_oracle, v19.\\\n- CUDA OOM bei 8 Paralleltischen → CPU-Fallback ist normal/erwartet.\"],[1,\"note f3c42f44): observer stats → personality mirror; range posterior → sample holdings; behavioral oracle → sample actions; EV per hero action. Distilled EV/action NN (revived EquityNet) replaces rollouts for live.\\\n\\\n## Session context / lessons (2026-08-19 collection saga)\\\n- Tightness death spiral: tight-heavy table draws + observer-adaptive tightening → near-zero flops. Fix: loose quorum in compose_table (v19_collect.sh, commit 3d97fd9)\\\n- 3-max unusable: G47 preflop = raise-war at 3-handed (78% raise-first, zero limps, 5-20% flops)\\\n- Parallel CUDA init OOM → 20s launch stagger\\\n- Health validation: per-table street distribution vs gold {24/17/59}% flop/turn/river-ended, ~15 equity taps/hand\"],[0,\"\\\n\\\n##\"]],\"start1\":625,\"start2\":625,\"length1\":2156,\"length2\":1925}]"
metadata_diff: {"new":{},"deleted":[]}
encryption_cipher_text: 
encryption_applied: 0
updated_time: 2026-08-20T05:26:10.155Z
created_time: 2026-08-20T05:26:10.155Z
is_locked: 0
type_: 13