GET-049: get stamps NOTHING — every file it writes must get the get row's ts as mtime

Now: OPEN
Sev: HIGH

INVARIANT (gritzko, 2026-07-17): every file affected by get/patch and every file undirtied by post carries that wtlog record's ts as its mtime, so the stamp-set fast path (STATUS-011) confirms it clean with zero content reads. get is the outlier: NO io.setMtime site exists in the whole get path (put/patch/post all stamp, BE-011), so every fresh clone / checkout / update re-hashes wholesale on every status, forever. Sibling verify tickets: PATCH-013, POST-029.

Input

Context

Found while profiling status (STATUS-011 Blockers): checkout mtimes are natural write times, not stamps, so wtlogReader.has(mtime) misses every get-written file.

be/verbs/get/get.js + its checkout path: zero setMtime calls; compare put.js:235 trySetMtime, patch.js:417, post.js:942 (BE-011).
The get row ts is the stamp value: wtlog stamp-set membership is exact (wtlog.js has(ts)), same policy as put's restamp-to-row-ts.

Goals

After any get (fresh clone, update, fast-forward), every file get WROTE has mtime == the get row ts; an immediate jab status does zero content reads on them (the STATUS-011 observable).

Verify first: a repro test asserting the invariant on a fresh clone + an update — RED today.
Fix: stamp each checkout-written file to the get row ts (one shared helper with put's restamp preferred).

Constraints

Stamp ONLY files whose written bytes == the new baseline blob. A GET-048 merge OUTPUT (weave of pending changes) differs from base — stamping it would lie ok; leave merge outputs unstamped (or route them to the DIS-057-style band as a follow-up ruling).
The row must be appended BEFORE stamping (ULOGAppendAt refuses ts<=tail; same ordering discipline as DIS-057's band-under-ceiling).
Untouched-by-get files keep their mtimes; deletes need no stamp.

WIP

Design decisions

REJECTED (gritzko): reading the assigned ts back off the log (tailTs) — no log read-back, ever; ts lives in the environment.
be.now = the verb invocation's start ts, minted per run in loop.js mintBe (resident procs re-sample); ambient.now() reads it; get's ctx.T0 = ambient.now() — ONE ts for rows AND stamps, reused across subs when the sub-mount follow-up lands.
ulog.append now honours an explicit MONOTONIC ts verbatim (bump only on use<=tail/prev, no silent bump-to-now) and RETURNS the assigned ts per row — the collision-edge stamp. POST-029 owns the canonical hunk; duplicated here byte-close.
fanoutWholeTree pins ctx._get.stampTs = the assigned row ts; null (restore/pick/mergeWorktreeTo: no row) = no stamp.
Only the clean-overwrite leaf stamps (bytes == new blob); mrg/con weave outputs + symlinks (setMtime follows) unstamped.
Repro = test/get/stamp (run.sh + assert.js): mtime==row ts + the statusfast io.open hook (0 reads), clone + update legs.

TODOs

Verify/repro test: fresh clone → all tracked mtimes == get row ts, status does 0 content reads; update-get → newly written files stamped, untouched files' mtimes intact.
Fix get's checkout/update write path to restamp; merge outputs excluded.
Full suite + end-to-end: clone, status timing before/after; DIS-023 idempotency intact (status writes no stamps).

Blockers and bummers

The anticipated band follow-up ruled 2026-07-19: carried/woven get outputs ride the DIS-057 band — GET-050 (stale put stamps false-clean carried files; incident work/STATUS-016).
Sub MOUNTS still unstamped: submount.mount/checkout.apply write the sub wt with natural mtimes — the sub re-hashes; follow-up (ruling: the SAME be.now stamps parent + every sub in one run).
Same-tip re-get restamps nothing (leafUnchanged skips the write): an mtime-churned clean wt stays slow until re-written.
shared/ulog.js append() hunk is DUPLICATED with POST-029 (they own the canonical one) — expect a weave dedup at merge time.
Fresh-clone anchor row rides ronStepMs(T0,-1) (1ms before the tip row's T0) so the tip ts == be.now exactly; cosmetic only.