62 lines
No EOL
3.7 KiB
Markdown
62 lines
No EOL
3.7 KiB
Markdown
# HANDOVER_R1F.md
|
|
|
|
Session closure / handover for the ARIMA + SAX hybrid forecasting research.
|
|
|
|
## CURRENT STATE
|
|
- Walk-forward causality: PASS (no evidence of lookahead anywhere in target/ATR/pipeline).
|
|
- Preprocessing / ARIMA / SAX / Hybrid boundaries: PASS.
|
|
- Stride selection deterministic: PASS. Stride sample representative: PASS (with a small-n sampling limitation).
|
|
- **OPEN MAJOR FINDING (F8): stride reproducibility / provenance gap.**
|
|
|
|
## AUTHORITATIVE COMMITS
|
|
- Authoritative R1: `af3f699`
|
|
- Previous checkpoints: `2360cca` -> `c08b74e` -> `bd323a6` (amended closure commit).
|
|
- Current HEAD / remote main: `bd323a68d2ce2c1af78f833282d1c30df270b066`.
|
|
- Prior commits: `ff95e11` (DEFECT-001 fix), `1ea9b41`, `0e56f73`, `3fbd268`.
|
|
- Branch: `main`. Remote: `origin` -> `https://forge.mql5.io/chiki2bum2/ARIMA_SAX_Hybrid_Forecaster.git`.
|
|
|
|
## R1 STATUS
|
|
`NOT SUPPORTED — PROVISIONAL PENDING R1-F`
|
|
|
|
## R1-F STATUS
|
|
`INCONCLUSIVE — INCOMPLETE`
|
|
|
|
## COMPLETED GATES
|
|
- **R1-F.1** — `COMPLETE`: PASS with MINOR ATR-convention limitation (ATR is causal close-diff, not OHLC true-range; internally consistent). Target/horizon correct, no future leakage.
|
|
- **R1-F.2** — `COMPLETE`: F8 MAJOR reproducibility gap (below). All other R1-F.2 checks PASS.
|
|
|
|
## OPEN FINDING — F8 (MAJOR, REPRODUCIBILITY / PROVENANCE)
|
|
- Committed stride defaults in `scripts/run_r1.py`: M1=60, M5=12, M15=4 -> would select **6519 / 6387 / 6428** origins.
|
|
- Published R1 results (`R1_RESULTS.csv` n_total) used: **489 / 511 / 515** origins.
|
|
- Effective strides approximately **M1=800, M5=150, M15=50** reproduce the published counts exactly.
|
|
- **The exact runtime stride value used for the authoritative R1 run is NOT recorded** in committed configuration/provenance.
|
|
- F8 is a **reproducibility/provenance** issue. It is **NOT** evidence of lookahead and **NOT** evidence the R1 sample was biased.
|
|
- F7 found the actual sample **reasonably representative** of the eligible population.
|
|
- Exact provenance remains **unresolved**.
|
|
|
|
## NEXT GATE
|
|
`R1-F.2a — STRIDE PROVENANCE TRIAGE`
|
|
Purpose: recover the exact runtime stride or exact selected-origin set from existing evidence WITHOUT changing or rerunning R1.
|
|
Outcomes: A) runtime stride recoverable; B) exact origin list recoverable; C) neither recoverable.
|
|
|
|
## BLOCKED ACTIONS (until separately authorized)
|
|
- DO NOT rerun R1. DO NOT modify the stride. DO NOT change configuration.
|
|
- DO NOT perform R1-F.3 / R2 yet. R2 stays BLOCKED until R1-F verdict.
|
|
- DO NOT tune parameters. DO NOT add MALE5 / new models/features.
|
|
- DO NOT edit R1 evidence to hide F8.
|
|
|
|
## EVIDENCE LOCATIONS (results/ is git-ignored, local only)
|
|
- `results/R1_PROTOCOL_MANIFEST.json`, `results/R1_RESULTS.csv`, `results/R1_COST_SENSITIVITY.json`, `results/R1_SUMMARY.md`, `results/R1_real_data/`, per-resolution `results/R1_m1|m5|m15/`.
|
|
- `results/R1_FORENSIC/R1_FORENSIC_REPORT.md` (scope-limited, superseded).
|
|
- `results/R1_FORENSIC/R1F1_TARGET_AUDIT.md` + `.json`.
|
|
- `results/R1_FORENSIC/R1F2_WALKFORWARD_AUDIT.md` + `.json`.
|
|
- `results/R1_FORENSIC/R1F2b_EFFECTIVE_STRIDE.json` (supplement for F8).
|
|
|
|
## SOURCE / TESTS (committed)
|
|
- Source: `src/`, runner: `scripts/run_r1.py` (stride defaults 60/12/4), configs, tests (`tests`), docs.
|
|
|
|
## REPOSITORY STATE
|
|
- **PUBLISHED on Forge.** Remote repository `chiki2bum2/ARIMA_SAX_Hybrid_Forecaster` created (public, empty) and local `main` pushed normally (no force).
|
|
- Remote `main` == local HEAD == `bd323a68d2ce2c1af78f833282d1c30df270b066`; main tracks `origin/main`.
|
|
- Working tree CLEAN at close. No source/config changes this closure session; only this documentation handover was committed.
|
|
- Authoritative R1 `af3f699` is an ancestor of `bd323a6`. Do not force-push or rewrite `af3f699`. |