DIS-080: markerless conflicts — RGA weave bytes, EXPECTED tree, resolved==posted

Now: OPEN

RULING (gritzko, 2026-07-29): conflict fence markers (<<<</||||/>>>>) are retired — fenced bytes cannot be cleanly diffed or merged again, which breaks multi-patch stacking. On conflict the wt file gets the weave's LIVE BYTES per RGA rules (both sides' insertions in deterministic weave order, no fences). Conflict state lives in wtlog rows + render, never in file bytes. Umbrella; implementation split: PATCH-025, ULOG-004, DIFF-016, STATUS-017, POST-035.

Input

Context

Today weave.merge output carries fences; shared/conflict.js hasConflictMarker scans them; con liveness = durable con row AND markers-on-disk (classify.js:423-427); resolution is detected by markers vanishing. All of that assumes fenced bytes.
Fenced files poison every later weave/diff: a second patch run folds the fences as content; multi-patch is the driving case.
Supersedes the marker-liveness half of STATUS-005 (its durable con rows stay).

Goals — the rulings

Wt bytes on conflict = RGA live reading of the merged weave. No fences, file always tokenizes, always re-weavable.
EXPECTED tree = the straightforward weave reading of base + all in-scope patch-ins (theirs layers, no WT layer). Everything is weave-based; no separate diff3.
Durable per-file con <path> wtlog rows stay (append-only). Cheap accessor: conflict paths = con rows scanned back to the last get/post barrier.
Resolution (AMENDED, gritzko 2026-07-29): a registered conflict is resolved by (1) an explicit put/delete row LATER than its con row (row order decides), or (2) a forced post! — believed all-resolved, its row is the barrier; else (3) still pending, post refuses. No content-based detection. Amendment lands in POST-035's wt (conflicts() shadows con rows by later put/delete).
Diff colors by weave token provenance: local edits (wt vs EXPECTED) green/red; patched-in tokens pale blue/orange — they are edits, the other kind of in/rm; overlap/conflict yellow. Patched-in tokens weigh as edits in the max(in,rm) row metric.
Dirty stays wt != base. EXPECTED only splits dirt into kinds: wt==expected → patched-in dirt (posts as the merge commit); wt!=expected on a patched path → local edit on top; ok remains wt==base only.
be patch keeps returning non-zero on any conflict (PATCHCONFLICT) — unattended patch && post chains must stop.

Constraints

Region/provenance info comes from the weave (per-token origin ids), not from stored rows — rows stay path-granular.
Retirements once landed: hasConflictMarker scan in post pre-flight (goes row-based), the cnf mtime stamp slot, marker-based tests.

WIP

TODOs

PATCH-025 RGA live bytes out of the merge, no fences
ULOG-004 conflicts accessor (paths since last get/post)
DIFF-016 EXPECTED-based coloring + metric
STATUS-017 classify kinds vs EXPECTED, row-based con liveness
POST-035 row-based conflict gate, post resolves