[JacobStephens2]

Summary

0.31.0 records host Suspended time and accounts for it in stage views, health and timing comparisons (#2053). Sessions also wait for free disk space by default (#2104).

• Commands record Suspended time alongside full wall-clock stage durations in show, status and Console views. Expanded Command details wrap to keep the explanation readable on desktop and phone (Record host Suspended time in Command stages and show #2038, Explain Suspended time in stage views and exclude it from Usual ranges #2059, Spec: record host Suspended time, and keep it out of stage health and Usual ranges #2053).
• Slow and Silent exclude suspension from working and quiet time; Usual ranges omit suspended stage samples (Exclude Suspended time from Slow and Silent #2060, Explain Suspended time in stage views and exclude it from Usual ranges #2059). Experiments and Process evals exclude suspended Trials from timing comparisons and estimates while retaining outcomes, Scores, usage and cost. Reports disclose exclusions; Experiment results advance to format 11 (Exclude suspended Trials from Experiment and Process eval timing #2063).
• Host limits require 5120 MB free disk space by default. resources.min_available_disk_mb = 0 disables the disk limit; thirdshift resources and Console resource views show available disk and its minimum (Host limits wait for free disk space (#2091) #2104).
• Recognized host out-of-disk failures hand issues back as ready-for-agent, without agent-failed, even when work is on origin. Security fixes retain agent-failed (A Run the host ran out of disk space under hands its issue back, not agent-failed (#2092) #2111).
• A rate-limit probe showing at least 500 requests remaining clears the Quota wait and lets Commands retry once immediately. Stored waits with sufficient quota no longer skip Passes (Record a Quota wait only when the probed quota is low #2143).
• Unchanged CI, mergeability and existing Base fix waits back off from 10 to 30 to 60 seconds, resetting on observed changes; CI polls reuse one PR observation (Back off unchanged GitHub waits and reuse PR observations #2142). Check and Abandoned job re-runs keep watching dependent jobs awaiting new attempts, with a final grace period after other jobs settle (Keep a Check re-run going while dependent jobs await new attempts #2137).
• Fix Pass checks closing-issue Claims before and after reserving a PR, skipping claimed candidates (fix: skip Fix Pass adoption while a linked issue is claimed #2134). Ticket runs confirm their PR is open before creating a Base fix issue (Confirm the pull request is open before a Base fix #2152). Delivery, Fix and Review share readiness polling while retaining their existing rules (Share pull-request readiness with Review runs #2150).
• Fix CI sessions keep validation logs local, use targeted reproductions and run validation in the foreground. Review instructions limit PR-state checks to once immediately before a multi-minute suite (Keep Fix run validation logs local and scope Review PR checks #2101). Review reports list and count actionable findings only; findings later fixed still count, and historical metrics remain unchanged (Count only actionable code review findings #2129).
• Security runs that pass their skip gates fail before any Session when repository Issues are off (Fail a Security run before any session when Issues are off #2133). Custom process Security Steps skip fixing without suppressing audits when Issues are off (A Security run on a repository with issues turned off audits and skips fixing (#2094) #2125).
• Both Triage modes receive repository categories before their Session; invalid proposed-issue categories get the correction turn (Validate Triage categories in the correction turn and give Report Triage the repository categories #2139). Blank questionnaires count as absent; non-empty questionnaires are accepted only for needs-info, including their rejection for ready-for-human (Treat blank Triage questionnaires as absent #2135). Recommendation interpretation moves into one shared module, preserving behavior (Interpret the Triage recommendation in one module #2148).
• Fleet metrics capture saved sources once per operation so machine choices and results stay consistent (Capture Fleet metrics saved sources once per operation #2120). Unreadable OS accounts remain attached to each machine across live reads, cached snapshots and pushed reports, appearing in JSON, text and the Console (Fix unreadable OS accounts missing from Fleet metrics #2136).
• New failure records and Where runs failed fallbacks use the stage before trailing cleanup; explicit Recovery stages retain precedence (Attribute unrecovered failures to the stage before trailing cleanup #2140).
• Fleet schedule repository groups and Other start closed on every load, with independent expansion for the current visit; Machines status stays visible (Fold Fleet schedule repositories closed on every load #2141).

Also: site/docs updates (#2101, #2129), Pass and host-expansion notes (#2119, #2132), Release run design docs (#2155), project-board Claim and PR status synchronization (#2123, #2124, #2130), and test-fixture cleanup after killed harnesses (#2117).

Version diff

diff --git a/Cargo.lock b/Cargo.lock
index 76472595..404aa854 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2432,3 +2432,3 @@ dependencies = [
 name = "thirdshift"
-version = "0.30.0"
+version = "0.31.0"
 dependencies = [
diff --git a/Cargo.toml b/Cargo.toml
index 79612376..73425937 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -2,3 +2,3 @@
 name = "thirdshift"
-version = "0.30.0"
+version = "0.31.0"
 edition = "2024"

[JacobStephens2]

Review run. No fixes to push. Spec axis used the pull request body alone; no closing issue was linked.

I reviewed Release PR #2165 against fixed point 4d24f79d. The code itself is correct, but the release summary in the PR body states the wrong Experiment results format number. I filed that as issue #2166, since I'm not allowed to edit the PR. The review changed no files, so there is no commit.

The diff is commit abb0ac30 Release 0.31.0, which bumps version from 0.30.0 to 0.31.0 in Cargo.toml and Cargo.lock. No closing issue is linked, so the Spec is the PR body alone.

Standards: no findings

• Release rule: scripts/release.sh says a release PR "bumps the version and nothing else", limited to Cargo.toml and Cargo.lock. The diff does exactly that.
• Other release files:
  • There is no changelog.
  • Release notes are generated into the PR body, not committed.
  • dist-workspace.toml pins only the cargo-dist tool version, so it needs no change.
• Lockfile: cargo metadata --locked --offline succeeds, so Cargo.lock matches Cargo.toml.
• Other checks:
  • Version 0.31.0 is plain X.Y.Z and higher than v0.30.0.
  • None of the baseline code smells apply.
• Files read: Cargo.toml, Cargo.lock, scripts/release.sh, dist-workspace.toml, AGENTS.md, CODING_STANDARDS.md, docs/testing.md, docs/releases/0.1.0-acceptance.md, scripts/release-summary.md.

Spec: 1 finding

• What matches:
  • The diff matches the body's "Version diff" exactly.
  • HEAD's parent is the stated release-content-base c84de08c.
  • Every merge in v0.30.0..c84de08c is cited, except Release 0.30.0 #2102, which is the 0.30.0 release PR itself.
  • Spot checks hold, for example the 5120 MB disk minimum in src/resources.rs:22.
• (c) Wrong: the summary says "Experiment results and the Console schema advance to format 11 (Exclude suspended Trials from Experiment and Process eval timing #2063)". On the branch they move from 11 to 12:
  • docs/experiment-results.md says format 11 at v0.30.0 and format 12 now.
  • src/console/openapi.rs:608 has "const": 12.
  • The reviewer's check command exits 1, which confirms the finding.
• Missing requirements or scope creep: none.
• Files read: .thirdshift-review-input.md, Cargo.toml, Cargo.lock, AGENTS.md, scripts/release-summary.md, scripts/release.sh, src/resources.rs, src/console/openapi.rs, docs/experiment-results.md, plus the git history v0.30.0..c84de08c.

Fixed

Nothing. The only finding is in the PR body, which I may not edit; no code change fixes it.

Unaddressed findings

• Spec, "format 11" should be "format 12": confirmed by the run above. Fixing it means editing the body of Release 0.31.0 #2165, which this run may not do, so a maintainer needs to correct it before merging. That keeps the GitHub Release notes right. Filed as Release 0.31.0 summary says Experiment results format 11; the branch has format 12 #2166 with only the needs-triage label.

Review coverage

• Both reviewers read both changed files.
• No generated output was excluded.
• No changed file was left unread.

Validation

1. scripts/affected-tests main exited 1, but the result is invalid. The local main (69c8689) is behind the fixed point, so the script picked up unrelated files and ran npm test --prefix site. That failed because the site's npm dependencies aren't installed here (Cannot find package 'marked').
2. scripts/affected-tests origin/main (origin/main is the fixed point) exited 1. Because the Cargo files changed, it selected the full Rust suite, and the full-suite cargo test ran 25 test binaries: 2774 passed and 1 failed.
   • The failure was process_session_evidence_check_all_private_fix_identifiers_and_abbreviated_commits in --test experiments, at tests/experiments/process.rs:111.
   • It is a flaky test, not a regression. The test rejects any request.json containing the substring 421, and that run's random Experiment hash (…944421ec…) contained it.
   • cargo test stopped at that binary, so the later integration binaries didn't run in this pass.
3. Rerunning the failing test alone (cargo test --test experiments process_session_evidence_check_all_private_fix_identifiers_and_abbreviated_commits) exited 0: 1 passed.
4. The binaries cargo test skipped, from experiments onward, now cover the whole suite:
   • experiments through quota_wait (cargo test --no-fail-fast --test experiments --test export_trace … --test release … --test user_config): 37 binaries, 987 passed, 0 failed. This run hit the 10-minute background limit while running release.
   • release through user_config, rerun as cargo test --no-fail-fast --test release … --test user_config: exit 0, 49 binaries, 962 passed, 0 failed.
5. The flaky test is filed as Flaky test: private-identifier check matches '421' inside random Experiment hashes #2167, also with only needs-triage.

The extra command list was empty. Logs are in .review-logs/ inside the worktree. PR CI is still running; its one failing test check is from a cancelled run that a newer run replaced, not a test failure.

The reports are standards.md and spec.md in .thirdshift-review-hqkr72/, next to summary.json ({"standards":0,"spec":1}).

Summary: Standards 0 findings. Spec 1 finding, the worst being the release summary's wrong format number, which is unaddressed and tracked in #2166.

Security

No unaddressed introduced findings.

CI passed or absent at abb0ac3