forked from chiki2bum2/SniperGold_ML
9.5 KiB
9.5 KiB
SESSION HANDOVER — P3-S.17R.2 VECTORIZE CHoCH / OB / FVG (FULL-SCOPE PARITY)
Date : 2026-08-23
Session : P3-S.17R.2 — Vectorize CHoCH / OB / FVG for Full-Scope
Runtime <-> Research Parity
Status : COMPLETE (officially closed)
Verdict : FULL PARITY ACHIEVED
P3-S.18 : NOT STARTED (blocked; requires separate, explicit authorization)
Next : NEW DECISION SESSION
A. Session
P3-S.17R.2 — Vectorize CHoCH / OB / FVG Full-Scope Parity
Final status : COMPLETE
Final verdict : FULL PARITY ACHIEVED
CHoCH : vectorized
OB : vectorized
FVG : vectorized
B. Starting Checkpoint
P3-S.17R.1 close checkpoint : 5cd1e7eed9cefbc91f6975de3cb9c1cda2755046
Verified at session start : local == origin/main, branch main, working tree
CLEAN, origin = https://forge.mql5.io/chiki2bum2/SniperGold_ML.git.
Authoritative handover read first :
docs/SESSION_HANDOVER_2026-08-23_P3_S17R1_RESEARCH_SWEEP_GATE_REPAIR.md
(newest authoritative by commit ancestry; supersedes P3_S17R1.md).
C. Final Forge HEAD
Final Forge HEAD (verified) : 7cc8af6ab9047f565f883683721d0ee9d11a3342
(docs: record P3_S17R2_FINAL_SHA ff3202c)
Meaningful P3-S.17R.2 parity-close commit :
ff3202c8be59d9f1073b2f286e219c54f96eb44a
Verified conditions (close, pre-closing-commit):
local HEAD == remote HEAD == 7cc8af6...
branch = main
working tree = CLEAN
origin = https://forge.mql5.io/chiki2bum2/SniperGold_ML.git
D. CHoCH vectorization
Engine-2 CHoCH (fractal 2/2, newest-4 trend, close break) transcribed into a
vectorized numpy kernel:
reference == vectorized : 197,032 / 197,032 exact
dir / trend / newest-levels / events : 0 mismatches (22,798 onsets in feed)
Approach: pivot masks via neighbor comparisons, newest-pivot selection via
searchsorted/gather, per-bar event with latest-frames masks. Semantics
consumed by the F3 Candidate Setup chain reproduced identically.
E. OB vectorization
Order Block (opposite-color candle B before a strong-move candle M with
|body| >= 1.5x avg-body(20); zone = full B; CLOSE-THROUGH mitigation with
strict inequality, includes the move candle; partial-fill overlap stays
ACTIVE) fully vectorized:
reference == current : 197,032 / 197,032 bars
dir / formation / mit / bounds : 0 mismatches (196,628 active bars)
Bit-exact average-body arithmetic achieved by summing newest-first sequences
identically to the runtime (IEEE-754). Sliding min/max + cumulative sums drive
the full-range zone and close-through/partial-fill boundary.
F. FVG vectorization
Fair-value-gap (3-candle wick geometry, bull Low(C3) > High(C1); zone
[High(C1), Low(C3)]; strict zero-gap; newest closed bar eligible as C3 (no
1-bar lag); wick full-fill mitigation) is reproduced:
reference == current : 197,032 / 197,032 bars
dir / formation / bounds : 0 mismatches (180,251 active bars)
Loops over gap offsets with incremental min-L/max-H sliding windows; zone
type identifier assembled in the chained namespace (parity-absence guards
kept green).
G. Reference-vs-vectorized evidence
Full-scope audit over the homogeneous 197,032 feed bars (in-scope identical):
CHoCH : 0 mismatches (dir / trend / events / bounds)
OB : 0 mismatches (dir / formation / mit / bounds)
FVG : 0 mismatches (dir / formation / bounds)
No tolerance hides the threshold semantics; the OB average-body boundary-flip
was found and removed (bit-exact newest-first sum matching the runtime).
H. Full runtime <-> research parity
Full authorized scope (194,985 bars; chain consumed on the 197,032 feed):
runtime_created (canonical_oracle) : 694
research_created (F3SetupEngine) : 694
identity matches : 694 / 694
runtime-only : 0
research-only : 0
per-bar state mismatch (in scope) : 0
Result: 694 runtime == 694 research, identity 694/694, zero runtime-only /
research-only, zero per-bar mismatch. Full Candidate Setup parity.
I. Candidate Setup population
Authorized P3-S.17 in-scope population : 686 Candidate Setups
PRESERVED DISTINCTION (never collapse):
full runtime/research computable population : 694
P3-S.17 authorized in-scope population : 686
De-overlap :
leads : 594
follow-ons : 92
duplicate IDs : 0
Historical correction (recorded, not silent): the P3-S.17R.1 "8 setups on 11
windows" came from pre-vectorization research ports whose CHoCH trend was
inverting, OB mitigation scanned older bars, and FVG window was detached from
the decision bar. Corrected kernels are validated against the frozen runtime
transcription and yield the population above.
J. P3-S.17 re-audit
Rerun on the corrected population:
raw Candidate Setups (in scope) : 686 leads (de-overlapped) : 594
follow-ons : 92 de-overlap separation : OK
duplicate setup_ids rejected : 0
class distribution (leads) : WIN 167 | LOSS 404 | UNRESOLVED 18 |
AMBIGUOUS 5
censoring : 22 UNRESOLVED (18 leads), never forced to
LOSS
ambiguity : 5 AMBIGUOUS (same-bar TP+SL)
lifetime (sweep) : sweep_age mean 21.0, choch_age mean 4.8
leakage : NONE
reproducibility : PASS (deterministic re-run)
K. Label distribution (CONTRACT OBSERVATION only)
Leads, on open defaults (k_tp=1.5, k_sl=0.75, H=16):
WIN : 167 (28.1%)
LOSS : 404 (68.0%)
UNRESOLVED : 18 (3.0%)
AMBIGUOUS : 5 (0.8%)
This LOSS-heavy distribution is a DATASET CONTRACT observation about the
open-parameter candidate set. It MUST NOT be read as strategy / profitability
/ model failure, and no parameter was selected or optimized on it.
L. Open label parameters
Primary open label parameters :
k_tp = 1.5
k_sl = 0.75
H = 16
Robustness reference run(s) :
H = 8
symmetric 1.0 / 1.0
All label parameters remain OPEN, NOT OPTIMIZED, NOT AUC-SELECTED, NOT
PF-SELECTED. Any future review must be an authorized contract decision, never
a performance selection.
M. Production protection
NO MQL5 CHANGES. Unchanged : MQL5/** production files,
AF_Engine2_Setup.mqh, F1/F2/F3/F4 runtime semantics,
FEATURE_CONTRACT.md, SniperGold_ML.mqh, model weights, training pipeline,
legacy v4.x semantics. Research-side vectorization only.
N. Tickstory / Dukascopy status
DEFERRED. The current population was validated on the same source used for
runtime parity. Tickstory/Dukascopy remains a future independent
data-source robustness study and is NOT introduced automatically in the next
session.
O. Final verdict
FULL PARITY ACHIEVED.
CHoCH : FULL PARITY (reference == vectorized, 197,032/197,032)
OB : FULL PARITY (reference == current, 197,032/197,032)
FVG : FULL PARITY (reference == current, 197,032/197,032)
Candidate Setup : 694 runtime == 694 research (identity 694/694,
runtime-only 0, research-only 0, mismatch 0)
P3-S.17 : RE-AUDITED (686 in scope; 594 leads / 92 follow-ons)
Production : UNCHANGED
ML : NOT TOUCHED
P3-S.18 : NOT STARTED
P. Next-session authorization boundary
The next session is a NEW DECISION GATE. It must explicitly choose:
OPTION A — LABEL CONTRACT REVIEW
review k_tp / k_sl / H, the LOSS-heavy (68%) observation,
TP-before-SL suitability, and setup-outcome interpretation.
Rules: no AUC, no PF, no ML training, no parameter optimization
disguised as validation.
OPTION B — P3-S.18 SETUP-LEVEL BASELINE ML
permitted ONLY if the owner explicitly authorizes it.
Would use ONE Candidate Setup as one observation, the P3-S.16 label
contract, keep legacy MLP frozen as reference, create a separate
setup-level baseline, and NOT compare it directly as if it predicted
the legacy target.
OPTION C — DESIGN DECISION REQUIRED
P3-S.18 is NOT started automatically. Parity does NOT auto-select B.
Until explicit authorization: no ML training, no LSTM/Informer/regime/
calibration, no AUC/PF optimization, no TP/SL or horizon change because of
LOSS rate, no Candidate Setup redesign, no MQL5 runtime changes, no
Tickstory/Dukascopy substitution.
Regression (final)
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 | SR 15/15 | VEC 15/15
No historical suite was modified or weakened.
Provenance / Close Verification
Forge HEAD (start) : 5cd1e7e... (P3-S.17R.1 close)
Forge HEAD (close) : ff3202c8be59d9f1073b2f286e219c54f96eb44a
(recorded at final pre-close HEAD 7cc8af6...)
Branch / remote : main == origin/main (forge.mql5.io/chiki2bum2/SniperGold_ML.git)
Working tree : CLEAN
Close commit : P3_S17R2_CLOSE_SHA (filled after this closed-session commit)
End P3-S.17R.2 official close handover. Verdict: FULL PARITY ACHIEVED. CHoCH / OB / FVG vectorized == frozen reference over the complete authorized scope; full runtime <-> research parity (694 == 694, 0 runtime-only, 0 research-only, 0 mismatch) with authorized in-scope population 686 (594 leads, 92 follow-ons); P3-S.17 re-audit completed; P3-S.18 blocked until separately authorized.