| field | _template |
|---|---|
| description | Domain-neutral starting point for a cc-tree field profile. Copy to <your-field>.md and fill in. |
A field profile gives the engine domain-aware reviewer weighting (loaded via
--field <name|path>, seedocs/ENGINE.md§2.2). It does not relax any universal rule — it only re-prioritizes which branches the 12 framings explore first. Keep every list short and concrete; vague entries ("be rigorous") add no weighting signal.
What does a senior reviewer in this field reliably attack first? List the recurring failure-patterns they look for, each phrased as a checkable concern.
- <e.g. "data leakage across the train/test split">
- <e.g. "multiple-hypothesis testing without correction">
- <e.g. "effect size reported without confidence interval">
What does the field currently take for granted that a result might quietly depend on? For each, name the regime where it is known (or suspected) to break.
- — breaks when
- — breaks when
- — breaks when
Concrete, recurring ways work in this field goes wrong in practice (not hypotheticals).
- <e.g. "pipeline silently drops rows with missing covariates">
- <e.g. "baseline tuned harder than the proposed method">
What counts as strong evidence in this field, versus weak/anecdotal? The engine raises its citation standard to match.
- Strong: <e.g. "pre-registered replication on an independent cohort">
- Weak (insufficient on its own): <e.g. "single-run result, no error bars">