[ScriptedAlchemy]

Merge #4478 first, followed by #4526, #4510, #4524, #4489, #4485, #4527, #4528, #4530, #4531, #4529, #4534, and #4541.

This PR is number 2 of 13, based on #4478. Merge #4510 next.

#4515 has already merged into this branch.

Use merge commits. After merging a parent PR, change its child's base to master before merging the child. Auto-merge is disabled.

Cursor and Codex summarizer timeouts were reported as unavailable providers. LCM then parked the session without counting a failure or setting a retry deadline. A slow run could stop summary generation until the binding changed.

Timeouts now return the existing LcmError::DeadlineExceeded. Convergence counts the failure and retries through its durable queue and capped backoff. The persisted delay starts after the failed run finishes, and the scheduler calculates the remaining delay at the end of the page. Provider runtime no longer consumes the retry delay.

Missing or unconfigured providers still park. Other failure classifications, cancellation, and conditional queue updates keep their existing behavior.

The four new regressions use executable adapters and a registered session store. They cover Cursor and Codex timeouts, missing executables, persisted failure counts, retry eligibility, status, and successful summary publication without a binding change.

Fixes #4522.

Validation

Local Rust builds and tests used Cargo with the pinned nightly toolchain.

• Before the fix, the two timeout regressions failed with unavailable or pending outcomes. Both missing-executable controls passed.
• After the fix, all 4 new regressions passed. The timeout test covers both providers. Its baseline run stopped at the first Cursor assertion.
• The affected LCM selection passed 63 tests, with none failed or ignored.
• The first session-runtime run passed 165 tests. Four profile fixtures failed because the store authority rejects the cloud machine's overlay filesystem. Those four passed with the same executable and TMPDIR on supported local tmpfs. All 169 selected tests therefore passed, with none ignored.
• One existing test was excluded from both selections: lcm_effects::tests::refresh_begins_at_the_committed_frontier_after_background_summaries_publish. test flake: refresh_begins_at_the_committed_frontier_after_background_summaries_publish times out under load #4518 and PR test(session-runtime): keep fake summarizer under default budget #4521 track that flake.
• Rustfmt, whitespace checks, and commit-message lint passed.

The recorded build command was:

cargo --config profile.test.package.asupersync.codegen-units=256 test \
  --locked --offline \
  -p tracedecay-session-runtime \
  -p tracedecay-rusqlite-runtime \
  --features tracedecay-rusqlite-runtime/test-runtime \
  --lib --no-run --message-format=json

The second package enables the existing native test-runtime bootstrap without manifest changes. The session-runtime executable from Cargo's artifact records ran with --test-threads=1, an isolated profile, and the global database disabled.

Compilation used one job, no incremental compilation, no test optimization or debug information, and 256 codegen units for asupersync. The fixture rerun changed no source or filesystem policy. Production timeout and retry limits remain unchanged.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 1733323

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR