topic-story

name	topic-story
description	Explore one to three evidence-bounded research directions, apply non-compensable qualification gates, and build a scientific story only for a direction that can continue or pivot credibly. Use for topic selection, question refinement, feasibility triage, gap mapping, falsifier design, and early go-pivot-stop decisions.

Topic & Story

Turn an early research idea into a bounded decision. Narrative appeal never compensates for missing evidence, inaccessible data, invalid measurement, or an unfalsifiable claim.

Inputs and artifact

Use the supplied material to create research-brief.md from ../../templates/research-brief.md. Treat anything not supplied or retrieved through an actually available tool as unknown.

The brief is also the requirement boundary for downstream work. Freeze the intended deliverables, explicit non-goals, observable acceptance criteria, human decisions that cannot be delegated, and unresolved inputs. Do not turn a model inference into an agreed requirement.

Evidence classes

Label every substantive statement as one of:

• provided evidence: directly anchored to supplied material;
• model inference: reasoned but not independently verified;
• unknown: required information is absent.

Preserve supplied identifiers and source anchors. Never invent searches, citations, datasets, access, results, approvals, or domain consensus. If no retrieval was performed, state that literature coverage is unverified.

Candidate exploration

Generate one to three meaningfully distinct candidates. For each:

1. define the question, unit or population, focal construct, comparator, outcome, and intended contribution when applicable;
2. identify supporting evidence and model inference separately;
3. assess data access, measurement, design fit, ethics, feasibility, and the smallest credible test;
4. state the decisive evidence gaps;
5. define at least one observable falsifier;
6. state claims the candidate must not make.

Do not collapse hard gates into a weighted score. Evaluate evidence, data access, measurement, design fit, ethics, feasibility, and falsifiability separately. A blocked hard gate cannot be offset by a persuasive story.

Decision

Choose exactly one:

• CONTINUE: all non-compensable gates pass and a credible falsifier exists;
• PIVOT: the underlying problem remains plausible, but a specific question, scope, population, outcome, design, or evidence plan must change;
• STOP: a critical premise is unsupported, the claim is not falsifiable, required data or safeguards are unavailable, or credibility cannot be reached within the stated constraints.

Missing information is uncertainty, not negative evidence. State what would change the decision. Recommend at most one primary candidate.

Story gate

Build a story spine only for a qualified CONTINUE candidate or a specifically bounded PIVOT:

situation → tension → falsifiable question → evidence arc → narrow contribution → claim boundary

For STOP, withhold the story and return the evidence or redesign plan.

Before finishing, call record_checkpoint with the decisive evidence, uncertainty, or stop rule. Prefix its statement with [topic-decision]. A checkpoint records the decision boundary; it is not research evidence.

Output

Return:

1. evidence boundary;
2. candidates and hard-gate table;
3. comparison and decisive gaps;
4. CONTINUE, PIVOT, or STOP;
5. falsifiers, repair actions, and stop rule;
6. story spine or the reason it is withheld;
7. bounded delivery contract and unresolved human decisions;
8. the completed research-brief.md artifact.

Match the user's language.