GloriousFlywheel Longevity Sprint Retrospective

GloriousFlywheel Longevity Sprint Retrospective

Sprint window: 2026-04-19 to 2026-06-14

Closeout snapshot captured early on: 2026-04-20

This retrospective records the sprint state that was actually landed by the time the release, canary, and benchmark truthing slices were merged. It is a closeout note, not a promise to keep projecting optimism until the original June target.

What Is Now True

  • FOSS core is explicit: repo-owned docs now separate runners, cache, operator surfaces, and release boundaries from future control-plane work
  • GitHub ARC is the primary measured execution surface
  • Downstream adoption moved past proof-of-life into a repeatable first canary tranche and migration kit
  • Forgejo compatibility now has one honest disposable proof on honey instead of remaining purely notional
  • Compatibility-only paths are called out as compatibility-only instead of being counted as product truth
  • SaaS/control-plane work stays above the FOSS bootstrap boundary instead of being tangled into initial deployment truth
  • Full competitive benchmark coverage exists across GitHub-hosted and commercial baselines
  • Cache-hit-rate and reliability measurements are complete enough for broad performance positioning

Benchmark Evidence

Measured source: gloriousflywheel-benchmark-scorecard-2026-04-19.md

Metric Target Actual on merged main Status
Time to first step (nix) < 60s 48s on tinyland-nix; 30s on tinyland-nix-heavy met
Queue latency < 15s 47s on tinyland-nix; 29s on tinyland-nix-heavy missed
Nix bootstrap overhead track separately 12.070s on tinyland-nix; 14.850s on tinyland-nix-heavy met
gf-nix-build runtime n/a 1.222s on tinyland-nix; 1m 25.411s on tinyland-nix-heavy measured
gf-flake-check runtime n/a 12.708s on tinyland-nix; 21m 22.739s on tinyland-nix-heavy measured
Cold start (docker) < 30s not measured in current sprint closeout missing
Nix cache hit rate > 80% not measured; only coarse cache availability recorded missing
Bazel cache hit rate > 70% not measured; only coarse cache availability recorded missing
Hosted/commercial compare n/a not measured in current sprint closeout missing

Benchmark Read

What We Can Say

  • the repo-owned benchmark workflow is real and succeeded on merged main
  • both current Nix lanes are measured, including the heavy lane
  • runner bootstrap/setup overhead is separated from timed workload execution
  • release-baseline benchmark truth is stronger than template optimism

What We Cannot Say Yet

  • that GloriousFlywheel is faster than GitHub-hosted runners
  • that GloriousFlywheel beats Namespace, Blacksmith, or RWX on named workloads
  • that cache hit rate, reliability, and operator-debug cost are fully measured
  • that one source-repo benchmark pack generalizes to the broader consumer set

Canary Results

Repo Forge Result Evidence
Jesssullivan/XoxdWM GitHub landed canary Jesssullivan/XoxdWM#31 merged as 66c93aa; Runner Health Check run 24663084258 passed; Cache Warming run 24663085690 advanced into real build and Attic push steps
Jesssullivan/cmux GitHub landed cross-org proof GloriousFlywheel #271 merged as d838caa; targeted Platform Proof run 24661112554 passed
MassageIthaca GitHub explicit exception, not counted as default-branch canary remains a named branch-authority exception rather than a default-branch sprint canary

What Remains Compatibility-Only

  • Bzlmod overlay kit (lives in bzl-cross-repo)
  • GitLab runner fleet (Helm chart validates, no live fleet)
  • Forgejo adapter remains proof-only: one disposable repo-scoped honey proof, not a production fleet

What We Currently Win On

  • repo-owned runner and cache contracts are documented from actual merged runtime truth
  • GitHub ARC is primary, measured, and exercised through real canary repos
  • the first downstream tranche is now a repeatable migration kit instead of a loose set of one-off patches

What We Currently Lose On

  • queue latency is still too high for strong benchmark marketing claims
  • the competitive benchmark story is still narrower than the original GitHub-hosted and commercial comparison ambition
  • broader orgwide adoption remains much smaller than the overall recent repo footprint

What Moves Next

  • broader GitHub-hosted and commercial benchmark comparison becomes explicit follow-on product-positioning work instead of active sprint exit criteria
  • control-plane tenant features remain later work above the now-clarified FOSS core
  • broader canary rollout remains follow-on expansion, not unfinished tranche-1 cleanup
  • secrets and MCP wiring work remain separate platform backlog lanes

GloriousFlywheel