market-researchlisted

# Market research The method M12 ran on 2026-09-18, captured so the next pass re-runs it instead of re-deriving it worse. Adapted from `akitaonrails/ai-memory`'s `docs/` — the same artifact set one project-stage later. **Read `docs/research/2026-landscape.md` first.** It is the previous pass. A new pass is a follow-up that says what it is a follow-up to and what moved, not a fresh start. ## The rules, before the artifacts These are what make the output worth anything. Breaking one quietly is worse than not running the pass. 1. **A capture date in every header**, and the sentence that goes with it: *"every load-bearing claim was checked against a live page on <date>, not remembered."* If you did not open it, do not write it. 2. **Primary sources only.** The repository, its README, its issue tracker, its releases, the vendor's own documentation, a published paper. **Never a third-party "X alternatives" or "best tools for Y" page.** This category is full of them and they are all published by a vendor that appears in its own ranking. Named, so the next pass recognises them on sight: `abralo.com/alternatives`, `runpane.com/alternatives/*`, `nimbalyst.com/blog/*`, `codeagentswarm.com/guides/*`, `munderdiffl.in/blog/*`. They may be cited **only** as evidence of how a vendor positions itself, labelled as such. 3. **Read the code, not the marketing.** Every claim about how a competitor's feature works names the file it was read in. M12's sharpest