91 lines
No EOL
3.6 KiB
Markdown
91 lines
No EOL
3.6 KiB
Markdown
# RESEARCH RULES
|
|
|
|
| field | value |
|
|
|---|---|
|
|
| Document | RESEARCH_RULES.md |
|
|
| Version | 1.0 (2026-08-28) |
|
|
| Status | **CONTROLLED** |
|
|
| Project | ARIMA_SAX_Hybrid_Forecaster |
|
|
| Authoritative workspace | `D:\TradingTerminal\MetaTrader 5\MQL5\Shared Projects\ARIMA_SAX_Hybrid_Forecaster` |
|
|
|
|
These rules govern every session and every gate in this repository. They are
|
|
read together with `RESEARCH_CHARTER.md` (the permanent purpose) and
|
|
`RESEARCH_STATE.md` (the compact machine-readable checkpoint). Rules are
|
|
mandatory. A change to these rules is a controlled change, documented and
|
|
versioned as described in §"Rule-change governance".
|
|
|
|
---
|
|
|
|
## Mandatory rules
|
|
|
|
1. Read `RESEARCH_CHARTER.md` and `RESEARCH_STATE.md` before any session.
|
|
2. Use repository documents as the source of truth, not conversation history.
|
|
3. One session = one gate.
|
|
4. Do not repeat completed gates.
|
|
5. Do not broaden scientific scope without explicit authorization.
|
|
6. Frozen protocols are immutable.
|
|
7. Any scientific-scope change requires protocol amendment.
|
|
8. SHA proves identity, not scientific correctness.
|
|
9. Any quantity that can change the scientific conclusion requires
|
|
independent verification.
|
|
10. A test written by the same implementation is not sufficient by itself
|
|
for scientific validation.
|
|
11. Never select models, parameters, horizons, features, or costs because
|
|
they produce a preferred result.
|
|
12. Engineering failure is not a scientific result.
|
|
13. Scientific failure is not automatically an engineering defect.
|
|
14. Never claim success from an execution that fails an integrity gate.
|
|
15. Never infer execution authorization from context or previous approval.
|
|
16. Current gate must always be completed before NEXT_GATE begins.
|
|
17. If a gate fails, STOP and preserve evidence.
|
|
18. Do not overwrite historical evidence.
|
|
19. Do not silently repair discrepancies.
|
|
20. Results may update project STATE but may not redefine the research
|
|
question without authorization.
|
|
|
|
---
|
|
|
|
## Interpretation notes
|
|
|
|
- **Rules 6 and 18** protect frozen/immutable evidence: R1-R evidence, R2-H
|
|
evidence, frozen MALE5-ML protocols and manifests, and reference MALE5
|
|
v3.2.0 are preserved byte-for-byte.
|
|
- **Rules 11, 5, 7** together prohibit every form of "model shopping," of
|
|
outcome-driven selection, and of silent scope broadening.
|
|
- **Rule 8** reminds that a correct SHA-256 (identity) does not by itself
|
|
establish scientific correctness.
|
|
- **Rules 9 and 10** guard the independence of verification: conclusions that
|
|
can flip must be confirmed independently of the code that produced them.
|
|
- **Rules 12 and 13** keep engineering defects and scientific results
|
|
distinct so that one is not mislabeled as the other.
|
|
- **Rules 15 and 20** bind the gate discipline: authorization for execution is
|
|
obtained per gate, and results update the state checkpoint only.
|
|
|
|
---
|
|
|
|
## Controlled classifications
|
|
|
|
- **IMMUTABLE:** R1-R evidence, R2-H evidence, frozen MALE5-ML protocols,
|
|
frozen manifests, reference MALE5 v3.2.0.
|
|
- **MUTABLE:** `RESEARCH_STATE.md`.
|
|
- **CONTROLLED:** `RESEARCH_CHARTER.md`, `RESEARCH_RULES.md`.
|
|
|
|
---
|
|
|
|
## Rules-change governance
|
|
|
|
A change to this document must:
|
|
|
|
1. Be proposed, documented, and explicitly authorized;
|
|
2. Increment its version and add a dated entry to the table below; and
|
|
3. Preserve prior wording in the change record rather than silently rewriting.
|
|
|
|
### Change table
|
|
|
|
| Version | Date | Scope | Authorized by |
|
|
|---------|------|-------|---------------|
|
|
| 1.0 | 2026-08-28 | Initial permanent research rules (20 mandatory rules) | RESEARCH GOVERNANCE SETUP gate |
|
|
|
|
---
|
|
|
|
_End of RESEARCH_RULES.md — version 1.0._ |