| name | research-claims |
| description | Maintain the research workspace belief layer: open questions, claims, decisions, confidence levels, evidence links, and revisit conditions. |
Research Claims
Use when interpreting evidence, updating beliefs, closing questions, or changing
research direction.
Files
OPEN_QUESTIONS.md: uncertainties ranked by information gain.
CLAIMS.md: compressed belief state.
DECISIONS.md: choices that change future behavior.
RESEARCH.md: short dashboard summary.
Claim Rules
- No claim without evidence IDs.
- No high confidence without multiple converging runs or one decisive blocker.
- Every claim needs an implication.
- Use statuses:
supported, weakened, mixed, refuted, superseded,
unknown. Mark falsified beliefs refuted; do not soften a dead idea to
weakened.
- Use confidence:
low, medium, high.
- The ledgers are repo-global across programs. When a claim lands or changes
program relevance, keep the program card's
Primary claims: line current;
it is the per-program view of CLAIMS.md.
Question Rules
Each Q### should have:
- measurable answer;
- next experiment or reason it is paused;
- related claims;
- status:
open, answered, paused, or superseded.
Decision Rules
A decision changes future behavior. Every D### needs:
- rationale;
- evidence;
- revisit condition.
Do not create a decision for every run. Create one when the next branch changes,
when a path stops, or when a default practice changes.
Update Pattern
After evidence:
facts -> claim status/confidence -> question status -> decision if behavior changes
Then update the RESEARCH.md evidence snapshot if the current thesis changes.