GloriousFlywheel Default Branch Stability Recovery 2026-04-23

Snapshot date: 2026-04-23

Starting red-main commit under investigation: c6d6ebc6b38cd61f81f70c3de6ce5e3b31a00951

Target proof package

• Validate
• Secret Detection
• Platform Proof
• Source Bazel Proof

Failing runs at start

• Platform Proof run 24849468454
• Source Bazel Proof run 24849468531

Failure Classification

Evidence Summary

Platform Proof

• run 24849468454 failed on c6d6ebc6b38cd61f81f70c3de6ce5e3b31a00951
• failed jobs: Prove tinyland-docker contract, Prove tinyland-nix-heavy contract
• bounded rerun of the failed jobs on the same run id succeeded with no repo or infra changes:
  • tinyland-nix-heavy job 72782938961 completed 2026-04-23T21:36:10Z on runner tinyland-nix-heavy-jtq2j-runner-lwn75
  • tinyland-docker job 72782939003 completed 2026-04-23T21:36:41Z on runner tinyland-docker-vvx6d-runner-558cj
• overall run 24849468454 conclusion is now success
• Prove tinyland-nix contract succeeded in the same run, so the instability hypothesis is limited to the two failed jobs
• captured job tails ended before an application assertion or test failure
• final assessment: external/transient until disproven; rerun-only recovery is not evidence of a repo regression or workflow-patch incident

Source Bazel Proof

• run 24849468531 failed on c6d6ebc6b38cd61f81f70c3de6ce5e3b31a00951
• the failure path was inside the repo’s own Attic package build
• fetch attempts for https://static.crates.io/crates/aws-types/1.3.7/download failed with repeated curl: (6) Could not resolve host: static.crates.io
• a later mirror check against https://tarballs.nixos.org/... failed with curl: (28) Resolving timed out after 15001 milliseconds
• the missing crate was aws-types-1.3.7
• current working hypothesis: external crates resolution failure first, with a possible repo-owned resilience gap if it proves repeatable

Corrective Action

• recorded that Platform Proof recovered via bounded rerun of the failed jobs on run 24849468454, with no repo or infra changes, so the incident stays in the external/transient lane unless it reproduces
• reran Source Bazel Proof after confirming the failure was a crates/DNS timeout and not a GloriousFlywheel assertion failure
• tightened the active docs and admin surfaces so they now say explicitly that current dogfood still depends on self-hosted cache/env injection, cluster-local reachability, and finite runner pressure
• updated the orgwide runner scoreboard and queue to distinguish declared runner intent from counted authority, so blocked or hybrid repos stop reading like clean default-branch authority

Residual Risk

• monitor for recurrence; if tinyland-docker or tinyland-nix-heavy fail again without a repo assertion, capture runner evidence before changing workflow code
• runner pressure is still active reality on shared lanes; the recovery should not be misread as proof that capacity and cache reachability are background solved problems

Current Dogfooded State After Recovery

• the originally red main commit c6d6ebc6b38cd61f81f70c3de6ce5e3b31a00951 is green across the target proof package after bounded reruns only
• the newer main head bfc8bc73c625f12476d2cda610dd4e03740cadc5 is also green on Validate, Secret Detection, Platform Proof, Source Bazel Proof, and Publish to FlakeHub
• current management truth is now narrower and more honest:
  • tinyland-inc/lab is the counted shared-authority repo in the recent tinyland window
  • Jesssullivan/yt-text is a real shared-runner canary, but still hybrid by policy
  • Jesssullivan/acuity-middleware and Jesssullivan/scheduling-kit remain the counted repo-owned canaries in the recent Jess window
  • tinyland-inc/rockies, tinyland-inc/betterkvm, tinyland-inc/elders.tinyland.dev, and Jesssullivan/MassageIthaca are currently nominal-only because accessible runner inventory is zero