Beagle SCM

STATUS-017: classify splits dirt by EXPECTED; con liveness from rows, resolved==posted

Per DIS-080: dirty stays wt != base — patched-in changes are still dirt, they must be posted. EXPECTED (base + in-scope patch-ins, weave reading) only splits the dirt into kinds: wt==EXPECTED → patched-in kind (posts as the merge commit); wt!=EXPECTED on a patched path → local edit on top; ok remains wt==base only. The con bucket is driven by the ULOG-004 accessor alone — no marker scan, no mtime cnf slot; a conflict stays con until the next post (resolution is not detected, it is the post).

Input

Context

• Today classify compares wt vs base (curTip, classify.js:255) and layers the fragile DIS-057 mtime band (pat/mrg/cnf, classify.js:192-217) plus con-row+marker liveness (classify.js:423-427, wtHasMarker :56-63). Markers retired ⇒ liveness rule and cnf slot die.
• Overlaps: STATUS-013 (patched-in trees as classify input — this ticket is the EXPECTED-based form of that reading); STATUS-005 marker-liveness half superseded, marked STALE pointer.

Goals

1. Classify consults EXPECTED per patched path: pat/mrg kinds fall out of wt-vs-EXPECTED + EXPECTED-vs-base, replacing the stamp-band special cases; unpatched paths keep the plain base comparison (degenerate EXPECTED==base).
2. con bucket := path ∈ ULOG-004 accessor result. Quad !, red SGR, summary tally unchanged in look; a con path stays con even after edits, until post.
3. pat/mrg mtime stamps may remain as a fast path, but must not be load-bearing for correctness; cnf slot retired.

Constraints

• Do not resurrect baselineTip as a base ([patch-never-moves-the-base], PATCH-024).
• test/status/conflict (marker-degrade-to-mod golden) is overturned — regold to row-based liveness.

WIP

Design decisions

• EXPECTED = expectedSha() local to classify.js (DIFF-016 owns any shared helper): weave.build of curTip + each patchTheirs() tip, one N-way weave.merge, read by weave.mergedLive (no WT layer), framed as a blob sha — wt/base/theirs compares are all sha-vs-sha, one wt read (wtBlobSha).
• Consult GATE = "patch-TOUCHED" = theirs[path] && (!inBase || t.sha !== b.sha); a touched path is decided on BYTES (never the mtime stamp-set), an untouched one keeps the plain base comparison verbatim.
• Kind table on a touched path: wt==base → ok; wt==EXPECTED==theirs → pat; wt==EXPECTED!=theirs → mrg; wt!=EXPECTED → mod (wt-only variant: unk).
• Stamp band demoted to a fallback: consulted only when EXPECTED is unweavable (BLOB/unreadable). cnf slot dropped; the get band's ceil-2ms con slot dropped too (both are conflict signals — now rows), ceil-1ms mod kept (GET-050 false-clean guard).
• quad.js NEEDS NO EDIT: WT_BUCKET.con already carries the red/! flag and reads the classify bucket — only its source (rows, not markers) changed.

TODOs

• EXPECTED consult in classifyMerge; kind table pat/mrg/con vs mod/ok
• con bucket from accessor; drop wtHasMarker + cnf
• regold status tests; add edited-conflict-stays-con test
• quad/status render audit (no look change expected)

Blockers and bummers

• THREE existing tests assert the retired DIS-057 band and CONTRADICT this ticket — not edited, per the no-other-test rule:
  • test/patchband.js:94 asserts the cnf slot resolves to cnf (retired by Goal 3).
  • test/statusfast.js:145 scenario C asserts a cnf-slot mtime routes con (same retirement; live conflicts now carry a con row).
  • test/patch/stampband/assert.js asserts a clean apply is io.open-FREE — the EXPECTED consult reads bytes by construction; its bucket asserts (pat/mrg/con) still pass.
• FIVE failures INHERITED from the PATCH-025 import (marker-grepping cases, listed in that ticket): get/conf-state, get/confdrop, get/dirty, post/conf-state, post/patch-absorb.
• Post-side barrier leg NOT tested here (a conflicted absorb still commits until POST-035's row-based gate lands); expiry is covered by ULOG-004's test/wtlog_con.js.

Outcome

• Not committed. Suggested commit message: STATUS-017: row-based con, EXPECTED kinds
• Staged in //STATUS-017 (jab put): shared/classify.js + new test/status/edited-conflict, test/status/expected-kinds, regolded test/status/conflict.
• Also staged, IMPORTED verbatim from //PATCH-025 (dependency, not this ticket's work): shared/weave.js, verbs/patch/patch.js, verbs/get/get.js, test/patch/{conflict,dirty-overlap,same-anchor-conflict,stacked-conflict}.
• Full ctest -j16: 312/320 (base 543269d6 + import was 313/318; the 3 new reds are the contradicted band tests above).
• 2026-07-30 update to ///be/#4b044ea6: the 7 PATCH-025 imports ABSORBED (now trunk bytes — landed via DIFF-016/PATCH-025); weave.js/get.js conflict-salad reset to trunk; test/status/conflict/run.sh salad (ours row-scoped regold vs the landed PATCH-025 regold) rebuilt by worker-transcript replay onto trunk's version; the 3 new-test files survived as demoted o rows. All re-put: classify.js, weave.js, get.js (con acks), 4 status test files. Suite 320/323 — the sole reds are the 3 declared contradicted band tests; amending them needs gritzko's approval.