POST-006's worker proved the ticket's premise (a byte-shape discriminator between a real conflict and a quoted marker) is impossible. be post's refusal calls SNIFFHasConflictMarker(body) (sniff/POST.c:2921), the SAME predicate PATCH uses to classify its own WEAVE output as conf (sniff/PATCH.c:352, used at :667/:724/:1795). The critical finding: a genuine WEAVE conflict is emitted mid-line, abutting the divergent tokens — WEAVEEmitMerged (graf/WEAVE.c:1138-1148) produces e.g. const char *greet = <<<<"hello"||||"salaam">>>>; (test be-patch-31-force-past-conflict). That is byte-for-byte the shape of a doc comment quoting the markers (sniff/PATCH.c:340). The worker implemented the ticket's suggested "markers at line-start" discriminator: it cleared the doc false-positives but broke 4 genuine-conflict tests (be-patch-31, 16/17-conflict-*, branches-08-conflict) and was reverted. Because PATCH's conf and POST's refusal share one byte-scanning predicate and the real output is inline, any byte-narrowing that clears the doc case misclassifies real conflicts. Actively biting: during today's merge, landing PATCH-002 and PATCH-001 (both edit sniff/PATCH.c, which legitimately contains marker literals in its detection code) each forced a verified be post --force. See POST, PATCH.
A shared byte-level predicate cannot tell a real inline conflict from a quoted-marker literal; both are identical bytes.
SNIFFHasConflictMarker flags ANY <<<<…||||…>>>> triple; genuine WEAVE conflicts are inline (graf/WEAVE.c:1138-1148), so there is no positional or byte discriminator vs a comment/string quoting the markers.sniff/POST.c:2921) and PATCH conf-classification (sniff/PATCH.c:667/724/1795) call the same predicate; narrowing it for POST's benefit breaks PATCH's conf detection (4 tests proved this).sniff/PATCH.c itself) false-fires POSTCFLCT and demands a blunt --force, exactly the blast radius the original POST-006 warned about.Make POST's refusal provenance-scoped, not byte-scoped; this overlaps POST-016's commit-classification pipeline.
conf — thread per-file conflict provenance from PATCH's wtlog patch row through to POST's commit-time classification (sniff/POST.c change-set pipeline, the same path POST-016 (b99e921e) reworked for in-scope stamps).sniff/test/SNIFF.c): the doc_comment_inline case (the sniff/PATCH.c:339 shape) must classify NO-conflict for POST while a genuine PATCH-conf file still trips it; the worker staged this repro and reverted it to keep the tree green.