[loopx-agent]

Goal And Delivered Outcome

The PR-wait lifecycle tests stopped before checking their intended behavior: a compute-paused quota guard now returns observation diagnostics through its original capture, while the tests expected inline hook output. Read the digest-verified capture while preserving the rejected business decision and exact heartbeat receipt.

The consumer acceptance test also requested full presentation with the default native thin transport. Keep that refusal covered, use explicitly disabled context delivery for the compatibility full presentation, and retain default native delivery for thin. Both generated prequota commands still feed the same merge proof into actual validated protocol-host work and original-journal replay.

Author Declaration

• Written by: model_agent · Codex (GPT-6 family) · OpenAI.
• Specification: PR dependency observation and continuation readback, and the existing native heartbeat capture constraints at that revision.
• Intended base: main; work started from 396ff6655ce34831669c949b9f63ad1a576d4a96.

Criterion	Disposition	Evidence
Observation continues during paused business compute; original refusal remains binding	implemented	lifecycle guard helper asserts exit 1, ok=false, skip, should_run=false and unchanged receipt/reason before inspecting diagnostics
Shared durable cadence, permission revocation, read-only consumers and uncertain effects	implemented	complete lifecycle/observation suites with real isolated File and SQLite authority
Exact merge proof reaches selected Task, native input, validated protocol Host and replay	implemented	consumer adoption suite retains full requirements, action-signature negatives, one execution/debit, and unchanged replay
Native thin capture and compatibility full presentation remain distinct supported paths	implemented	unsupported native full is rejected without provider calls; generated full/thin prequota commands both run

Scope And Continuation

Complete within this test-fixture repair: two acceptance files, with no production implementation, default, permission or UI changes. The existing capture reader owns digest/privacy checks; the local helper owns the deliberately paused fixture and retains its rejection. Installed executor adoption, scientific effectiveness/cost claims and the wider CI qualification remain separate existing work.

Validation

• Tested revision: 99829eb32ab2392e1cd31066d4946d115544ef95.
• Run state: finished.
• Input classes: synthetic, public_fixture.

Check kind	Result	Evidence / limitation
regression_parity	passed	Unmodified shared-cadence acceptance fails on both File and SQLite with missing inline diagnostics. The first wider run preserved 70 passes and two heartbeat presentation failures before the second fixture repair.
real_backend / real_entrypoint	passed	Complete PR-wait lifecycle, observation, consumer adoption and retained-history suites: 72 passed in 76.81s; disposable File/SQLite and local protocol Host, synthetic GitHub provider.
unit	passed	Four typed PR-wait observation owner cases.
static	passed	Full contributor Ruff scope, configured mypy (17 source files), CLI output-budget smoke and diff/RFC checks.
static	passed	Changed-file public/private scan: no findings.
integration	not_run	Complete repository CI has not been qualified for this head; earlier full-shard failures remain historical failures. No passing full-suite claim.
integration	blocked	Exact-scope quality receipt remains failing on complete-CI qualification; standard premerge direct and compile checks pass, with no catalog canaries selected for this test-only diff.

• Visual evidence: not applicable; test fixtures only, no user interface changes.
• No merge, installation or release performed. Required checks and independent review remain prerequisites to merge.