research

Question: $ARGUMENTS `ROADMAP.md` is a direction with exit criteria and a sourced evidence table; `BACKLOG.md` is the pool behind it, with pools already born from research ("Pool — from the 2026-08 needs research"). This skill produces the next such input, never a rewrite of either file. ## 1. Frame it State the decision the answer would change (keep / expand / revise / stop which workstream), the milestone it belongs to, and what would count as evidence against your own expectation. ## 2. Gather, with sources - **External**: vendor docs, published studies, competitor surfaces — each with URL, date, who published it and their interest in the result. Vendor research is directional, never causal (the roadmap's own rule). - **Internal**: what the code already does (`scout`), what the corpus already answers (`corpus-prover` or `signals coverage` on a throwaway store), what `BACKLOG.md` already holds on the topic (ids, not paraphrase), what `docs/corrections.md` says went wrong nearby. - **Users**: issues and PRs (`gh issue list --search`), `SUPPORT.md` routes, pilot feedback the maintainer names. ## 3. Write the note `docs/work/<date>-research-<slug>.md`: question → decision → evidence table (claim, source, date, weight) → what it changes for the product → proposals. Proposals are concrete: a new ROADMAP evidence row with its footnote, a BACKLOG item in the pool's shape (next free `B` id, effort S/M/L, scope solo/team/both, the honesty caveat), a gate to add or d...