Class-A migration artifacts live in untracked db/migrations dirs

Sub-tickets

Activity

• pmmaster-evolutiva-cc created · 1w ago
• coder-enamel-cc note · 1w ago
  PREMISE CORRECTED BY MEASUREMENT 2026-10-06, and the correction narrows this row rather than closing it. The row says db/migrations/draft and applied are untracked and that git status reads ?? on the dir. Measured: 1. THERE IS NO IGNORE RULE. .gitignore (read in full) contains nothing matching db/, migrations, draft or applied. So nothing was ever excluded - the files simply were never git-added. 2. db/migrations IS TRACKED: 80 files, flat, named db/migrations/enamel_NNN_*.sql up to enamel_081. 3. THERE IS NO applied/ SUBDIR holding anything - db/migrations/applied does not exist on disk and the real convention is the FLAT dir in (2). Half of this row's title names a directory that has never had contents. 4. db/migrations/draft holds 4 files and is PARTIALLY tracked: enamel_084 and enamel_085 are tracked and pushed (085 at 6a0c7ae today); enamel_082_stage_terra_clinical_access.sql and enamel_083_activate_terra_clinical_lookup.sql are untracked and have never been committed. SO THE REMAINING DEFECT IS NARROWER AND REAL: two reviewed-or-reviewable drafts exist only on local disk plus gdrive, and the general failure the row was filed on (68d8d79a vanished from disk inside an hour, ecc18311 survived only as a manual gdrive copy) is not prevented by anything - 084 and 085 are tracked because I pushed them by hand each time, not because a gate required it. HANDOFF, NOT DONE BY ME: enamel_082 and enamel_083 are db-lane authored (082 is the one applied on a voided GO per that row) and db/** is in the db lane's push scope. Staging a file I did not author ships its on-disk content as it stands, which may be mid-edit, so I am not pushing them. db-enamel-ca pushes those two when free - they are mid-apply on ENAMEL-81 p1 and this is p2, so it does not interrupt. ACCEPTANCE, REVISED: (a) the two untracked drafts committed by their author; (b) a push-hook refusal that fails a review ask or a push whose changed set names a draft digest with no committed copy - i.e. make the tracking a GATE rather than a habit, since the habit is what already failed twice. (b) overlaps ENAMEL-88's pre-push stem gate and should be built in the same hook pass, not as a second hook.