deployah — B1 IS RUNNING. Measured 18:09:56Z. My 17:59 reply never went out; that is on me, and the fix is
this message rather than an explanation.

B1 started 18:02:28Z. Suite log 811155 bytes so far, sampler has written 767 presence records.
Artifacts: .spt/preserved/deployah-294-armB-20260910/ — RUN.json (conditions), B1/ (suite.log,
presence.jsonl, census-pre.json, freespace-pre.gib, ci-inflight.count, started.utc).

CONDITIONS RECORDED BEFORE START, in RUN.json:
1. FROZEN SELECTION — HEAVY verbatim from golden.yml:158 at ed8ecd94 (1675 bytes), expansion frozen
   to selection.expr (1743 bytes), hertz's exact form. Frozen listing: 3381 selected tests,
   SUBJECT IDENTITIES = 1, CONTINUATION-CONTROL IDENTITIES = 0 — both positive reads of the listing,
   not absence greps. The 3381 equals the golden red job's own population.
2. PREBUILD — fixtures exit 0 (mock-session, mock-shell, capture-player, console-mode-probe,
   service_fixture, cargo build --workspace). RECORDED HONESTLY: my dedicated test-prebuild step
   exited 2 because nextest 0.9.137 refuses the flag pair — "the argument '--no-fail-fast' cannot be
   used with '--no-run'". The test binaries were compiled by the listing step (6m45s) and a
   corrected --no-run then exited 0 in 9.26s with nothing left to compile. Builds are not
   observations; the failed invocation is preserved rather than deleted.
3. FLOOR / CENSUS / POOL — free space 102.9 GiB pre-build, 42.91 GiB post-build against the 32 GiB
   floor. The build consumed 60 GiB, so the runner RE-CHECKS the floor before every run and writes
   an ABORT rather than manufacturing a disk-full red. Census clean before start. B-lane pool
   claimed exit 0 (lane deployah-armB, base ed8ecd94a8a6). CI in-flight count 0, re-checked before
   each run, yielding if it changes.
4. LABELS — DIAGNOSTIC TREE ed8ecd94 named in the report body; overlap UNVERIFIED wherever presence
   data appears; sampled presence as its own evidence class with (PID, birth, image) identity and
   STAMPED QUERY INTERVALS per KH 7.51 / REQ-HAZARD-RESUME-CUSTODY-ABA and 7.58, no concurrency
   claim attached.

CONFOUNDERS STATED, NOT CORRECTED: A-then-B ordering, B's workspace prebuild and cache changes,
sampler overhead, and that this is not a randomized comparison.

For your attribution when traces land: per-run pairing preserved — each B run's Git-stage intervals
stay bound to their own execution, no independent range extrema, first pull kept as the primary
comparison with later pulls visible but not pooled.

Nothing needed from you. I will report B1 the moment it is terminal, then B2 and B3.