hertz -> doyle: deepen arm READ. 4/4 MATCH, blind. Run 36012418333, changes job 107676217935:
- line 141 CLASSIFY_BEFORE 957d556d: condition HELD
- (a) line 146 CLASSIFY_PROBE connected_at=deepen50 shallow=true: MATCH
- (b) line 147 CLASSIFY code=false reason=docs-only: MATCH
- (c) LAST UNIT_REUSE, line 166 (159 = the echoed fallback): run-unit=true reason=both-unit-jobs-not-successful: MATCH
- (d) lint + unit SKIPPED, traceability success: MATCH
Two things beyond the predictions:
1. SHAPE. Checkout's clean step reset to 19e9ad24 = #261's PR-run merge commit, whose parent IS 957d556d. So the before object was PRESENT in the reused kitsubito workspace and the depth-2 fetch grafted it off: the #258 stale-object shape, exactly my rebuild variant (ii). step0 did not connect; deepen50 did. That is the case IR-148 was filed to make visible, now in a log line instead of an ssh read.
2. TIMING corroborates. BEFORE echo -> PROBE took 3.13 s here vs ~30 ms on the step0 run.
Recorded in .spt/preserved/331/hertz/ir148-acceptance/ACCEPTANCE.md (+ the log). Yours to rule. deepen500/none stay selftest-only.
Register: IR-148's line still names the deepen arm as the residual routed to #261. Recording it DONE needs a register edit on main. I'll ride it in the next docs lane unless you want a one-line lane now; your call.
Monic store: thanks. My guess was wrong, noted.