Beagle SCM

GET-058: get must DECIDE by the quad table, not by ad-hoc flags

get already RENDERS the quad (/blog/status.mkd, shared/quad.js) but does not DECIDE by it: verbs/get/get.js:1465 quadReport() calls quadOf on the POST-get tree as a report only, in a try/catch that downgrades failure to a warning, while the D5 leaf and shared/checkout.js apply() roll their own oldMap/dirty/isBinary logic with a trackUpdate flag bolted on by /todo/GET/GET-040. Same vocabulary on output, different mechanism on the deciding side — which is why cells drift (/todo/GET/GET-056, /todo/GET/GET-057, the DIS-082-jab loss). RULED (gritzko 2026-07-31): the per-file verdict is a 13-cell track x work table in the status letters against the root; and /todo/GET/GET-053 is RULED FF-ONLY, so the diverged branch is deleted, not rewritten.

Input

Context

• shared/quad.js already computes every input the table needs: root = dag.mergeBase(track, base), four columns . same / x removed / o created / v advanced, and the wt column measured vs BASE not root (gritzko 2026-07-17) — one require away from get, nothing to re-derive.
• The landed /todo/GET/GET-040 fix (///be/#978b8a52, 2026-07-31 12:16, main tree tip) added the diverged TRACK UPDATE path: trackUpdate at get.js:453,475,536,794, divMerge (6 uses, ~:628-636), local-only-add and ours-deleted special cases at :1031,:1068, and the merge-base-blob weave leg at :1079. All of it serves the diverged case ONLY.
• The :1079 leg is wrong on its own terms: a dirty file's 3-way ancestor is the curTip blob (what the edit is a delta against), not the LCA blob — mixing the committed leg and the uncommitted leg into one weave.
• Under FF-only the four history shapes are: behind/FF (root = base, base column .), ahead (root = track, track column all . so every row is a no-op), same (all .), diverged (REFUSE on non-force; get! is the any-ref reset and consults no table).

Goals

• ONE decision table for both paths — the flat D5 leaf (get.js :1054) and the sub checkout (shared/checkout.js :227) index the SAME 13 cells; no second weave, no second binary heuristic.
• quadOf the wt BEFORE the checkout, not only after; each row's track+base+wt chars pick THEIRS / OURS / WEAVE3 / DELETE / no-op.
• Delete the diverged machinery listed in Context, replaced by an FF reachability test that REFUSES (/todo/GET/GET-053); dag.mergeBase survives demoted to that test.
• The post-get quadReport stays, and now reports in the same terms the decision was made in; conflicts print in the glyphs status already prints (/todo/DIS/DIS-080).

Constraints

• JS only, be/, run via jab. Do NOT run native be; do NOT open beagle/.
• Repro-first: a test per table cell BEFORE the rework; the existing get/sub buckets must stay green (ctest -j16, 328/328 at 978b8a52).
• Reuse only: weave3 from be/shared/weave.js, isBinary from be/views/diff/diff.js, quadOf/quadModel from be/shared/quad.js, URI/hash via be/core/resolve_hash.js. No new mechanism, no new deps.
• Clone ///be/ at 978b8a52 (it holds the code being deleted). Norms: 2-line GET-058 comments, 1-line 64-char ticket-coded commit.

WIP

Design decisions

• THE TABLE (gritzko 2026-07-31), FF slice, rows = track vs root, cols = wt vs base: - in base: T. W. no-op / T. Wv OURS / T. Wx stays deleted - in base: Tv W. THEIRS / Tv Wv WEAVE3 / Tv Wx THEIRS - in base: Tx W. DELETE / Tx Wv OURS keep file / Tx Wx no-op - not in base: To W- THEIRS / To Wo WEAVE3 (add/add) / T- Wo OURS untracked / T- W- no-op
• BINARY is a modifier on the both-sides-edited cells ONLY (Tv Wv, To Wo) → OURS + conflict: binary is never woven, a dirty binary is never reset. The six clean cells are byte-agnostic.
• Tv Wx is THEIRS regardless of bytes (gritzko 2026-07-31): a local delete carries no content, so nothing is destroyed, only an intent that costs one rm to reassert; a local EDIT carries content and wins.
• Presence lives inside o/x, so the 12 illegal presence combinations of a bitmask cannot be written down, and edited-vs-locally-deleted stop sharing one "dirty" bit — the two defects of the current oldMap model.
• DIVERGED = both track and base columns lit at once; that is the only shape needing a two-leg fold, and FF-only removes it from the non-force path. AHEAD lights the base column too but reads no-op via T., so it needs no special case.
• get! bypasses the table entirely: unconditional THEIRS plus the untracked sweep — force stays the SOLE cleaning path (/todo/GET/GET-040's invariant).

TODOs

• T0 — cell tests first: one repro per table cell, flat and sub, RED where trunk disagrees
• T1 — quadOf the PRE-get wt; thread the model into the D5 leaf and checkout.apply
• T2 — one verdict(track, base, wt, binary) helper; both paths call it, nothing else decides
• T3 — FF gate: reachability test, REFUSE a diverged non-force get (/todo/GET/GET-053)
• T4 — delete trackUpdate (:453,:475,:536,:794), divMerge, :1031, :1068, :1079
• T5 — conflict rows in status glyphs (/todo/DIS/DIS-080); quadReport unchanged
• T6 — full ctest -j16, no regression in get/sub/patch/status/post buckets

Blockers and bummers

• quad.js THROWS on "no common ancestor" by design and get's clone/bootstrap path has no base at all — that path must BYPASS the table, never acquire a degenerate root.
• /todo/GET/GET-048 (HIGH, divergent+clean must be a plain reset) loses its premise here: there is no diverged non-force path left to reset wrongly, and Tv W. → THEIRS satisfies it by construction. Close it against this ticket, do not fix it separately.
• Sequencing: T3 and T4 are only correct under the FF-only ruling; if it is ever revisited, the two-leg fold (root→base committed + base→work dirty, same table applied twice) is what comes back.
• Staged rows must survive an FF base advance (gritzko 2026-07-29, /todo/GET/GET-053) — the ...V → ...v demotion is that ticket's, not a table cell.
• A wt based BEFORE 5c0d5c21 runs its own pre-fix jsrc during its first get and still clobbers dirty exec files — back the wt up before the first update of any stale clone.

Outcome

• Suggested commit: GET-058: get decides by the quad table, no diverged path