RB-11 restoration — instrument liveness attestation and evidence-snapshot laws (2026-10-10)
Source:vignettes/articles/devlog-2026-10-10-m7-rb11-restoration.qmd
Restoration record. An earlier commit carried the RB-11 recording into the branch, was orphaned by a later rebase, and was pruned by gc: git grep over the current lineage finds zero RB-11 trace. The ruling texts survived only in the controller’s governing ledger. This file restores them to the tracked lane. Ruling IDs and violation names only; no persons, per the established doctrine.
-
RB-11(i) — instrument-liveness attestation law. An observation instrument may be cited as live in a report only after a post-deployment liveness attestation: a pre-echo to its own capture, or controller-verifiable process evidence (a ps-class line proving the watcher itself is resident). Companion clause defines dry-run strictly: end-to-end execution of the instrument under its target interpreter — pattern-grepping fixture lines is not a dry-run. Both directions were proven by the same failure family: a watcher reported as armed had died on launch (
declare: -A: invalid option— associative arrays need bash ≥ 4, macOS ships 3.2) and emitted zero events across a full matrix pass while its owner reported it as watching; a second-generation watch died silently to double-escaped quoting in an inline command, its progress lines printing blank (0/13the whole time). The generalization: a failed instrument that looks armed is worse than none — the controller idles past terminal states waiting on events that can never be emitted. Corollary from the same hour: report timestamps and counts come from command output only; a compose-time-fabricated start stamp stands as the negative example. -
RB-11(ii) — no redirect-target reuse across launches. A log path written with
>may never be reused as the redirect target of a relaunch without first snapshotting the prior bytes to a timestamped path. Violation instance: the second ignition’s>truncate destroyed ignition-1’s own launch evidence, deepening the very forensics deficit the incident review then had to reconstruct from transcripts. The law: snapshot first, or the relaunch truncates the evidence of the run you still must explain. -
Orphaning forensics (3a973e2). The pre-restoration recording commit became unreachable when its branch line was rebased;
git cat-file -s 3a973e2now fails, andgit fsck --danglinglists no commit bearing the RB-11 subject. The pruned content was not invention: both laws had already bound behavior between their entry and this restoration (attestation regime in force for every watch deployed after it; log-reuse check in force before every relaunch). This file is the authoritative tracked-lineage copy going forward. -
Lane-transfer note (controller, symmetric accountability). This restoration and the companion artifact commit (
docs(m7): authoritative degraded run artifacts …) were dispatched to the worker lane twice with fully specified content and produced zero diff both times (HEAD unchanged across both resumes, outdir still untracked). They were executed on the controller lane instead; the transfer and its evidence are recorded in that artifact commit’s message. Delegated-to-lane idleness is surfaced, not silently absorbed.