Literature Review (rapid evidence map)

Return JSONL rows:
{doi,title,year,design,sample,geography,exposure,comparator,outcome,effect,unit,quote,locator,limits,confidence}
Then: risk-of-bias bullets; top-3 gaps; 2 must-read DOIs; ≤200-word synthesis.
Success: ≥15 resolvable DOIs; quotes+locators; 1 counter-claim.

Modelling / Coding

Spec: state vars, flows, parameters, equations/relations.
Parameter table (CSV): name,definition,unit,prior_low,prior_high,source,quote,locator.
Add 2 stress tests + unit/sanity checks. Success: runs locally; tests green.
Debugging: list failure modes; add 3 unit tests; create tiny synthetic dataset + expected outputs.

Draft & Review

Outline: Produce a 6-section outline of the idea in IMRaD style:
Introduction (problem framing, 1 sentence)
Background (state of knowledge, 1 sentence)
Methods/Approach (what is done, 1 sentence)
Results/Evidence (what is found, 1 sentence)
Discussion/Interpretation (what it means, 1 sentence)
Conclusion/Implications (so what, 1 sentence)
→ For each, add section purpose (why this section exists).
End with thesis/antithesis/synthesis (≤2 sentences).
Critique: As Reviewer-2, give 5 ranked criticisms (severity 1–5) + suggested fixes. At least 1 must be methodological.

Revision: Write a ≤120-word abstract, define 2 key terms, and include a statement: “What would change our mind?”
Blind spot: Name 1 important limitation not addressed.
Success: Coherent IMRaD outline, falsifiable abstract, ≥1 methodological critique, and 1 blind spot