Accept resident maintenance cadence proof

This commit is contained in:
Meta Vens 2026-08-11 10:26:01 +02:00
commit 2155d87a2a
4 changed files with 47 additions and 19 deletions

View file

@ -3966,10 +3966,11 @@ state is now the sole final-action owner for these review paths.
own what may be retired, not how often a stable runtime is rescanned.
- Inputs: cognitive brake, shutdown signal, resident revision, prepared/active
lease state, and producer-owned typed source stores.
- Outputs: the first stable braked revision receives normal bounded
maintenance; later identical braked pulses only observe/reconcile resident
authority and publish operator physiology. New domain and continuation
pressure are ingested only after brake release.
- Outputs: maintenance runs on startup, resident revision change, or a bounded
60-second deadline. Stable braked pulses only observe/reconcile resident
authority and publish operator physiology. Stable released polls still scan
producer ingress at the configured wake cadence, but do not repeat retention.
New domain and continuation pressure are ingested only after brake release.
- Derived state: poll interval, readiness projection, and pulse count are not
retention or cognition-ingress authority.
- Forbidden writers: a braked pulse cannot bind a new repository domain,
@ -3978,9 +3979,11 @@ state is now the sole final-action owner for these review paths.
- Shared paths: brake and shutdown use the same ingress refusal boundary;
prepared/active child reconciliation remains live so the brake can drain and
terminalize exact existing authority.
- Cut line: unconditional retention scans no longer repeat at an unchanged
braked revision, and `cycle` no longer runs repository/domain/continuation
ingress while braked.
- Cut line: unconditional retention scans no longer follow poll cadence, and
`cycle` no longer runs repository/domain/continuation ingress while braked.
- Verification: exact packaged `3f37503e` on copied 23 MiB runtime state held
revision and runtime/local-Verse hashes stable, created no grant, exited zero,
and measured 0.29-0.33% CPU at one-second cadence versus prior 10-24% spikes.
Exact packaged `f04abd6b` then proved released sleeping at five-second
cadence: 91 process ticks over 30 seconds (about 3.0% of one core), stable
revision/runtime hash, no grant, restored brake, and clean exit zero.