forked from chiki2bum2/SniperGold_ML
61 lines
2.3 KiB
Markdown
61 lines
2.3 KiB
Markdown
# P3-S24 — RESEARCH VALUE ASSESSMENT & SIGNAL RECOVERY STRATEGY
| |||
| |||
```text
| |||
Session : P3-S24 — RESEARCH VALUE ASSESSMENT (ANALYSIS ONLY)
| |||
Status : COMPLETE
| |||
Policy : docs/CODE_VERIFICATION_AND_SILENT_BUG_POLICY_v1.md
| |||
Decision: D — CURRENT EDGE NOT DEMONSTRATED
| |||
Next : P3-S25 = NOT STARTED (owner authorization required)
| |||
```
| |||
| |||
## Purpose
| |||
| |||
After P3-S23 (corrected-population ML baseline: pooled OOS ROC-AUC 0.5235,
| |||
fold-3 collapse to 0.464, calibration worse than the constant prior) this
| |||
namespace answers the strategic question:
| |||
| |||
> Is the current SniperGold research framework limited by insufficient sample
| |||
> size, unstable market mechanism, population construction, or lack of
| |||
> predictive information?
| |||
| |||
The phase is ANALYSIS-ONLY: no ML training, no hyper-parameter tuning, no new
| |||
features, no label / TP / SL / horizon change, no MQL5 change, no deployment,
| |||
no trading, no automatic external-data import.
| |||
| |||
## Contents
| |||
| |||
| File | Role |
| |||
|---|---|
| |||
| `p3_s24_config.py` | Frozen inputs (populations, P3-S23 result, power config, external-data note, evidence phases) |
| |||
| `p3_s24_power.py` | Hypothesis A — Hanley-McNeil sample power / CI analysis (standalone, analytic) |
| |||
| `p3_s24_heterogeneity.py` | Hypothesis B — descriptive WIN-rate heterogeneity over the 5 frozen dimensions + year (corrected 579 rows) |
| |||
| `p3_s24_evidence.py` | Evidence matrix (P3-S18..S23) + research decision matrix + final classification logic |
| |||
| `p3_s24_run_main.py` | Orchestrates all analyses and writes `output/p3_s24_*.json` |
| |||
| `output/` | Machine evidence (power, heterogeneity, evidence matrix, decision matrix, classification, manifest) |
| |||
| |||
## Reproducibility
| |||
| |||
Run twice:
| |||
| |||
```text
| |||
python p3_s24_run_main.py
| |||
```
| |||
| |||
All substantive outputs are deterministic; only `generated_utc` differs.
| |||
`output/p3_s24_manifest.json` records a SHA256 of the timestamp-free content
| |||
per artifact and an aggregate; the aggregate must be identical across runs
| |||
(byte-identical except timestamps).
| |||
| |||
## Hard constraints honoured
| |||
| |||
```text
| |||
ML training / tuning : NONE
| |||
New features : NONE (bins of existing features only, descriptive)
| |||
Label / TP / SL / horizon : NONE changed
| |||
MQL5 / production : NONE changed
| |||
External data : NONE imported (MQL5/Files assessed only)
| |||
Deployment / trading : NONE
| |||
P3-S25 : NOT STARTED
| |||
```
| |||
| |||
*End P3-S24 namespace README.*
|