first commit

文档润色流和知识库构建流/claude-scholar/skills/research-ideation/references/research-contract.md

@@ -0,0 +1,111 @@
+# Research Contract
+
+Use this contract to carry research state across Claude Scholar workflows.
+
+The goal is to preserve:
+- what question is being studied,
+- what evidence currently exists,
+- what claim is allowed by that evidence,
+- what uncertainty remains,
+- what decision or next action should happen.
+
+## Research Question Card
+
+Use this card when a vague idea becomes a research direction.
+
+```md
+## Research Question Card
+
+Question:
+Type: exploratory | confirmatory | applied
+Hypothesis:
+Why it matters:
+Current evidence:
+Missing evidence:
+What would support it:
+What would falsify it:
+Minimal next action:
+Decision: explore | read more | run experiment | stop
+```
+
+## Evidence Record
+
+Use this record for paper evidence, project notes, experiment outputs, and analysis artifacts.
+
+```md
+## Evidence Record
+
+Evidence ID:
+Source:
+Source type: full paper | preprint | dataset | experiment artifact | project note | abstract-only | webpage placeholder
+Supports:
+Contradicts:
+Method / dataset / metric:
+Limitation:
+Project relevance:
+Claim strength: speculative | observed | supported | strong
+```
+
+Evidence ID format:
+- Use `ER-YYYYMMDD-shortslug-NN`, for example `ER-20260513-tta-eeg-01`.
+- Keep IDs unique within the project or research thread.
+- Use stable, human-readable slugs. Do not use vague IDs such as `E1`, `paper1`, or `source-a`.
+- Reuse the same Evidence ID when the same evidence record is referenced downstream; create a new ID only for a distinct source, artifact, or analysis result.
+
+## Claim Candidate
+
+Use this candidate when an analysis or synthesis suggests language that may later enter a report, paper, rebuttal, or project plan.
+
+```md
+## Claim Candidate
+
+Claim:
+Source evidence:
+Allowed wording:
+Forbidden stronger wording:
+Uncertainty:
+Next check:
+Decision: keep | weaken | revise | discard
+```
+
+## Source Trust Levels
+
+Use source trust to decide whether a note can support downstream synthesis.
+
+- `full paper` / `preprint`: can support `observed`, `supported`, or `strong` claims when the relevant method, dataset, metric, and limitation are named.
+- `dataset` / `experiment artifact`: can support project claims when the unit of analysis, metric, provenance, and analysis limits are named.
+- `project note`: can support hypotheses and plans, but not literature-backed claims unless it links to separate evidence records.
+- `abstract-only` / `webpage placeholder`: can support discovery and `To-Read` routing only. Do not use it to support `Knowledge`, manuscript, or rebuttal claims unless it is later replaced by a full paper, preprint, or verified artifact.
+
+## Claim Promotion Gate
+
+Before a claim moves into `Knowledge`, `Writing`, a report, a manuscript draft, or a rebuttal, check:
+
+1. The claim has at least one Evidence Record ID.
+2. The source type is strong enough for the intended claim.
+3. The claim strength is not silently upgraded.
+4. The allowed wording and forbidden stronger wording are both recorded.
+5. Contradictory evidence or missing evidence is preserved.
+
+If any item fails, keep the claim as a hypothesis, motivation, warning, or `To-Read` item. Do not polish it into a durable conclusion.
+
+## Proposal Readiness Gate
+
+Generate a `research-proposal.md` only when:
+
+- one Research Question Card is selected,
+- current evidence is enough to justify the question and method,
+- missing evidence is explicit and tractable,
+- the minimal next action is more specific than "read more",
+- citations or evidence records are available for the key motivation claims.
+
+If these conditions are not met, generate `research-question-card.md`, a gap note, or an intake summary instead of a proposal.
+
+## Strength Rules
+
+- `speculative`: plausible idea, weak or indirect evidence only.
+- `observed`: seen in a paper, note, or experiment, but not yet enough for a durable conclusion.
+- `supported`: backed by explicit evidence such as a paper result, experiment, or analysis bundle.
+- `strong`: supported by multiple evidence anchors or statistically rigorous project evidence.
+
+Do not promote a claim to a stronger level without naming the evidence that justifies the upgrade.