Beagle SCM

STATUS-005: merge conflicts must land in the wtlog and status as red con, not yellow mod

STALE (2026-07-29): durable con rows + red con landed; the marker-based liveness rule (markers gone → mod) is overturned by DIS-080 — markers are retired entirely, resolved==posted. See STATUS-017.

A merge that leaves conflict markers in a file records NOTHING: the wtlog gets only the bare get/patch row, and status shows the conflicted file as a plain yellow mod. Order: file conflicts must be recorded into the wtlog (.be) at merge time, and such files must be statused con and rendered red, not yellow. The existing cnf bucket (DIS-057) is a fragile mtime-stamp band tied only to patch rows — a get merge never gets it.

Input

Context

• Incident 2026-07-10 (work/JS-117): jab get ?#1a751c… merged JS-116 in, left ONE <<<</||||/>>>> block in shared/ingest.js; wtlog tail shows just the get row, jab status showed mod shared/ingest.js — indistinguishable from an ordinary edit.
• Bonus symptom: store-open maintenance require()d the conflicted .js and leaked idxmaint: SyntaxError: Unexpected token '<<' into status output — a red con would have explained the state at a glance.
• Existing pieces: shared/conflict.js hasConflictMarker (the POST-017 pre-flight scan, both WEAVEEmitMerged marker shapes); consumers verbs/get/get.js, verbs/patch/patch.js, verbs/post/post.js; shared/classify.js:150-170,241-260 the DIS-057 pat/mrg/cnf mtime-stamp band (ceil-2ms/-1ms/ceil on an in-scope patch row only).
• wtlog rows are <ron60> <verb> <uri> (shared/wtlog.js); per-file rows already exist (put <file>), so a conflict row has a natural shape.

Goals

1. The merging verb (get AND patch), right where weave emits a conflicted result, appends a per-file conflict row to the wtlog — durable, survives mtime churn, visible to any later session.
2. status classifies such files con: wtlog row says "merge left conflicts here", hasConflictMarker on the wt bytes confirms still-unresolved; markers gone → the file degrades to ordinary mod (resolved, uncommitted).
3. Render con red (mis/del-class severity), not the yellow mod class; summary line counts it (… 4 mod, 1 con, …).
4. post pre-flight keeps refusing marker'd adds (POST-017) — unchanged, but its error can now name the recorded con rows.

Constraints

• One state name must win: the order says con, the existing DIS-057 bucket is cnf — reconcile (rename or alias), do NOT ship both; flag the choice in the report for gritzko's confirmation.
• wtlog is append-only (Sniff owns it): resolution is detected by scan (goal 2), never by rewriting/deleting the conflict row.
• The DIS-057 mtime-stamp band stays for patch outcome reporting (pat/mrg) — this ticket only makes the CONFLICT state durable and verb-agnostic; don't vandalize classify.js's stamp logic.
• Marker-string false positives: prose <<<< is already excluded by the triple scan; --force posting stays the escape hatch.

WIP

Design decisions

• (proposed) row shape con <path> mirroring put <path>; one row per conflicted file per merge, duplicates across merges tolerated (newest wins, scan decides liveness).
• (open) whether con rows scope status's patch-band (classify.js) or ride a separate check in the classifier walk.

TODOs

• repro test: get-merge leaving a conflict block → status shows con (red), not mod; resolving the markers (edit) degrades to mod
• get.js + patch.js: append con <path> wtlog rows where weave reports conflicts
• classify.js/status.js: con bucket wired from wtlog row + marker scan; red rendering; summary count
• post.js: error message names the recorded con rows (no behavior change)
• cnf/con reconciliation, single name, report the choice

Blockers and bummers

• JS-116/JS-117 worktrees are in flight on overlapping shared/ files (store.js/ingest.js are NOT touched here, but classify/status/verbs are theirs-adjacent) — coordinate landing order.