주시
1
0
포크
이미 SniperGold_ML을(를) 포크했습니다
1
원본 프로젝트 chiki2bum2/SniperGold_ML
SniperGold_ML/docs/SESSION_HANDOVER_2026-08-23_P3_S17R1.md

10 KiB

SESSION HANDOVER — P3-S.17R.1 RESEARCH SWEEP EVENT CONTRACT REPAIR + FULL GATE WIRING

Date       : 2026-08-23
Session    : P3-S.17R.1 — Research Sweep Event Contract Repair + Full Gate Wiring
Baseline   : P3-S.17R close checkpoint 92e5bd6a8fa065c4be5509ad6c2100c1893739da
             VERIFIED (authoritative workspace; local == origin/main, branch
             main, origin = forge.mql5.io/chiki2bum2/SniperGold_ML.git).
Scope      : RESEARCH-ONLY. XAUUSD M15 (Files/AlgoForge/Data), authorized
             2017-01-03..2026-07-21 (194,985 bars). NO Tickstory/Dukascopy.
Human verif: REMAINS CANCELLED. NO AUC/PF/backtest/opt/ML training.
Verdict    : PARITY ACHIEVED WITH CAVEATS (see §1).
Next gate  : NOT training (P3-S.18 NOT ready). Full-scope vectorization of
             CHoCH/OB/FVG kernels -> re-run staged parity on the complete
             scope -> P3-S.17 re-audit on the reconciled population.

1. What P3-S.17R.1 established

A. SWEEP EVENT PARITY — ACHIEVED (full authorized scope):
     research event onsets 2,938 == runtime 2,938 (byte-locked sliding-window
     reference); runtime-only 0, research-only 0; contract-active bars
     194,985 / 194,985 exact; event identity (onset+dir+valid_until+
     superseded) identical.
   The legacy persistent-latch proxy (`smc_semantic_common.sweep_state`,
   1,896 active / 2,000 bars) is REPLACED in the setup path by the semantic
   EVENT port `ml/p3/smc_semantic/sweep_event.py`.
   NOTE: the P3-S.17R "58 onsets / 2,000" measurement was itself an
   approximation (PAR-suite 4/4 pivot port without begin=100); the faithful
   runtime pivot (internal len=5, begin=100) gives 25 onsets / 2,000, exactly
   matching the runtime reference.

B. GATE SERIES — WIRED (full scope) with runtime as-of semantics:
   H4 = Narrative agent dir, M30 = Context agent dir (newest CLOSED HTF bar
   with close_time <= decision-bar close), M15 = Entry agent dir (decision
   bar). Port: `ml/p3/smc_semantic/spec_tests_engine2_gates.py` (frozen
   Engine-2 agent rules, AF_E2_DIR_TOL, E2_MIN_BARS, Engine-1 cache as-of).
   In-scope non-zero: h4 163,830 / m30 178,764 / m15 145,220 of 194,985.
   Boundary (GR-T11): exactly at HTF close the just-closed HTF bar IS visible
   (runtime Engine-1 cache; contract §L text "close_time <= t" vs runtime
   "close_time <= decision close" — deterministic, documented, not a blocker).

C. CANDIDATE SETUP PARITY — runtime == research on every deterministic
   window (11 windows x 2,000 bars sampled; 8 setups total in 3 windows,
   identity + per-bar outcome match everywhere; the rest 0 by genuine frozen
   lifecycle filtering — e.g., 90000..92000: 19 single-bar complete chains,
   13 slot-blocked, 6 started and discarded at STRUCTURE_CONFIRMED by
   context/event lapse).
   The previous 0-population (P3-S.17) was an EXTRACTOR ARTIFACT:
   proxy latch + proxy mitigation + neutral gates + a frame/decision
   coordinate bug in the research runner (fixed).

D. P3-S.17 RE-AUDIT: NOT RUN (requires full-scope vectorization; see §5).

2. Deliverables

docs/P3_S17R1_RESEARCH_SWEEP_GATE_REPAIR.md                (this parity report)
ml/p3/smc_semantic/sweep_event.py                          (semantic EVENT port)
ml/p3/smc_semantic/spec_tests_engine2_gates.py             (Engine-2 gate port)
ml/p3/setup_dataset/spec_tests_sweep_research_parity.py    (SR-T01..T15)
ml/p3/setup_dataset/spec_tests_gate_research_parity.py     (GR-T01..T14)
ml/p3/setup_dataset/spec_tests_p3_s17r1_full_parity.py     (real-data runner)
ml/p3/setup_dataset/output/p3_s17r1_sweep_parity.json      (SR report)
ml/p3/setup_dataset/output/p3_s17r1_gate_parity.json       (GR report)
ml/p3/setup_dataset/output/p3_s17r1_full_parity.json       (staged full parity
                                                            + sweep detail +
                                                            gate evidence +
                                                            verdict)
docs/SESSION_HANDOVER_2026-08-23_P3_S17R1.md               (this file)
(gitignored: p3_s17r1_gate_series_cache.npz, *.csv)

3. Regression status (after research-only changes)

P3-S.2 17/17 | P3-S.3 20/20 | P3-S.4 20/20 | P3-S.5 20/20 | P3-S.6 20/20
P3-S.7 22/22 | P3-S.8 25/25 | F1 8/8 | F2 20/20 | F3 25/25 | F4 16/16
P3-S.16 20/20 | P3-S.17R 17/17 | P3-S.17R.1 SR 15/15 | P3-S.17R.1 GR 14/14
No historical suite weakened or modified. Historical report JSON timestamp
side-effects reverted.
Namespace: the new FVG/OB-referencing research files live in the spec_tests_*
namespace (frozen P3-S.17R decision) so P3-S.4 FVG-T20 / P3-S.5 OB-T20 stay
green (verified 20/20 each).

4. Explicitly unchanged areas

- MQL5 detectors / AF_Engine2_Setup / AF_Engine2_Agents / F1/F2/F3/F4 : unchanged
- FEATURE_CONTRACT.md / SniperGold_ML.mqh / weights / labels / legacy MLP   : unchanged
- P3-S.16 entry timestamp / TP-SL / horizon / WIN-LOSS-UNRESOLVED-AMBIGUOUS /
  de-overlap / feature timing                                               : unchanged
- No AUC/PF/backtest/optimization/ML/human annotation used.

5. Open for a FUTURE phase (NOT resolved in P3-S.17R.1)

1. Full-scope CHoCH/OB/FVG detection kernels need vectorization for the
   complete 194,985-bar Candidate Setup pass (the staged chain currently runs
   on 11 deterministic 2,000-bar windows).
2. After (1): re-run this staged parity on the COMPLETE scope, then the
   P3-S.17 re-audit (de-overlap, class distribution, censoring, ambiguity,
   lifetime, feature snapshot, reproducibility) on the reconciled full
   population.
3. Tickstory/Dukascopy = future independent data-source robustness study only.

6. Handoff

DO NOT train ML (P3-S.18 NOT ready). The sweep byte-lock and the H4/M30/M15
gate wiring are done; the research side now reproduces the frozen Candidate
Setup semantics wherever the chain is computable. The full-population
certification requires the deferred vectorization, then the P3-S.17 re-audit.
The legacy MLP remains frozen reference.

7. Provenance / Checkpoint

Forge HEAD (start) : 92e5bd6a8fa065c4be5509ad6c2100c1893739da (P3-S.17R close)
Branch / remote    : main == origin/main (forge.mql5.io/chiki2bum2/SniperGold_ML.git)
Working tree       : CLEAN at start (research artifacts adopted); CLEAN at close
P3_S17R1_FINAL_SHA : 13a23df0d87e1cda0178b2924dae2a5c1e143b46
Human verification : CANCELLED

End P3-S.17R.1 handover. Sweep event parity byte-locked; H4/M30/M15 gate series wired; Candidate Setup parity runtime==research on the computable scope; full-population certification deferred to the vectorization phase.


OFFICIAL SESSION CLOSE — P3-S.17R.1

A. Session

P3-S.17R.1 — Research Sweep Event Contract Repair + Full Gate Wiring
Final status : COMPLETE / Verdict = PARITY ACHIEVED WITH CAVEATS

B. Starting Checkpoint

92e5bd6a8fa065c4be5509ad6c2100c1893739da VERIFIED
(local == origin/main, branch main, working tree clean; the 4 untracked
P3-S.17R.1 artifacts from the interrupted earlier attempt were reviewed,
adopted, and completed)

C. Final Forge HEAD

1165174fbaf2f1bcc9eb41cba6ec887f6c3cbd13 (local == origin/main, working tree CLEAN;
P3_S17R1_FINAL_SHA recorded = 13a23df0d87e1cda0178b2924dae2a5c1e143b46)

D. Data Source

SAME authorized P3-S.17 source: XAUUSD M15 (Files/AlgoForge/Data),
2017-01-03..2026-07-21, 194,985 bars. NO Tickstory/Dukascopy.

E. Parity Results

data identity      : PASS (194,985 bars == P3-S.17)
H4 gate            : WIRED + GR-T01..T03 (full scope, as-of)
M30 gate           : WIRED + GR-T04..T06 (full scope, as-of)
Sweep              : PARITY (2,938 == 2,938 onsets; 194,985/194,985 active;
                     0 runtime-only, 0 research-only; SR-T01..T15)
CHoCH              : subset parity (PAR-T06/T07; full-scope kernel deferred)
OB                 : subset parity (PAR-T08/T09; P3-S.5 20/20)
FVG                : subset parity (PAR-T10/T11; P3-S.4 20/20)
M15 entry          : WIRED + GR-T09/T10 (full scope, as-of)
Candidate Setup    : runtime == research (identity + per-bar outcome) on all
                     11 sampled windows; 8 setups total; previous 0 = artifact

F. First Divergence Resolution

Liquidity Sweep:
  legacy latch                 : 1,896 active bars / 2,000 (state, no expiry)
  faithful EVENT port          :    25 onsets / 2,000
  byte-locked runtime reference:    25 onsets / 2,000  -> IDENTICAL
  full-scope                   : 2,938 == 2,938 onsets; active 194,985/194,985
Classification: STATE/EVENT CONVENTION DIVERGENCE RESOLVED.

G. Second Blocker Resolution

H4/M30/M15 gate series wired into the research path with the runtime as-of
semantics (Engine-2 agents; newest CLOSED HTF bar <= decision close; no
future bar; GR-T11/T12/T13 pin the temporal contract).

H. Zero Population Interpretation

NOT CONFIRMED and now EXPLAINED: the previous 0 was an extractor artifact
(proxy latch + proxy mitigation + neutral gates + runner coordinate bug).
With faithful detectors + real gates the sampled population is non-zero
(8 setups / 22,000 sampled bars; runtime == research).

I. Production Changes

NONE. MQL5 detectors, AF_Engine2_Setup, F1/F2/F3/F4, FEATURE_CONTRACT.md,
SniperGold_ML.mqh, weights, training pipeline all UNCHANGED.

J. ML Status

NOT TOUCHED. No MLP/LSTM/Informer/regime training, no calibration, no
AUC/PF/profitability/backtest/parameter optimization.

K. Tickstory/Dukascopy

DEFERRED. Future order: SAME SOURCE runtime<->research parity -> valid
Candidate Setup population -> P3-S.17 re-audit -> only then independent
data-source robustness study.

L. Final Verdict

PARITY ACHIEVED WITH CAVEATS

M. Next Authorized Phase

P3-S.17R.2 (or P3-S.17R.1 follow-on):
  Vectorize CHoCH/OB/FVG detection kernels -> re-run the staged parity on the
  COMPLETE 194,985-bar scope -> P3-S.17 re-audit on the reconciled full
  population -> only then consider P3-S.18 baseline readiness.

N. Explicit Prohibitions for Next Phase

- NO MQL5 runtime changes
- NO Candidate Setup redesign
- NO F2/F3 semantic changes
- NO ML training
- NO optimization
- NO Tickstory/Dukascopy source substitution
- NO P3-S.16 label contract changes

P3-S.17R.1 is officially CLOSED. Next authorized phase: full-scope vectorization + P3-S.17 re-audit (NOT training).