Prove abrupt recovery with live jobs in released and managed containers

Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a13590-a16f-7f60-be60-fd71627bc99d
This commit is contained in:
tegwick 2026-10-13 22:32:10 +02:00
parent 5b8e798419
commit 2cb2b654e0
25 changed files with 1954 additions and 4 deletions

View file

@ -0,0 +1,37 @@
# INTENT against SCOPE — 2026-10-13 22:31 Europe/Berlin
The abrupt-interruption gap is now closed with real local and managed execution.
Published release 0.1.0 recovered after SIGKILL during an actual browser job;
a queued job was also preserved as interrupted. Partial artifacts were removed,
completed hashes survived, and a fresh 14-step screenshot/keyboard trial passed.
Managed recovery used the same pod, PVC and image with a new container. See
[execution evidence](../docs/evidence/2026-10-13-crash-recovery.md).
Most important remaining gaps:
1. **Real self-service human and coding-agent access.** Identity registration,
governed agent credentials, native API authentication and per-caller transport
still need owner activation and native acceptance receipts. The completed
recovery check used an existing operator identity. It does not satisfy this
gap. RPF-WP-0056-T03 owns delivery; CUST-WP-0080-T05 coordinates it. Membership
is decided, so no additional eligibility question is needed.
2. **Cross-version rollback.** This is the first release. A same-image restart
proves crash recovery, not a version transition. Exercise actual rollback
when a second reviewed compatible release exists; retain the open requirement.
3. **Final managed availability and discovery.** A5 local-container guidance is
published. Real consumer receipts must precede final readiness and managed
discovery promotion under WP-0004-T01/T03.
4. **Consumer-specific acceptance.** Content-explorer owns its broader route,
keyboard and visual requirements. Existing provider evidence cannot close
that consumer task.
WP-0001–0003 remain finished. WP-0004-T02 is done; T01/T03 retain their narrower
external waits. SCOPE now includes the managed abrupt-recovery proof. The repo
contains repeatable consumer API and crash-recovery commands and sanitized
receipts. The managed pilot is Ready; temporary verification connections were
stopped. No runtime code, image version, grants or shared controls changed.
Validation: three unit tests and syntax checks passed; both real released-image
crash checks passed. An initial managed transport failure stopped before the
crash step and is retained explicitly. The overall implementation is not complete
until the outstanding consumer-access and acceptance requirements are satisfied.