Strategic Finding
Type: adoption-blocker / trust-risk
Horizon: near-term
When release gates stay red for weeks, the organization routes around them. On 2026-08-28, dakota published stable-20260828 from an Execute Release run in which the Release gate job was skipped — the 5th consecutive red run (2026-08-19 → 2026-08-28). Only post-release steps (create-multiarch-stable, update-main-bookmark) failed, after Promote release and Create stable image release had already succeeded. Meanwhile bluefin shows the opposite failure mode: one failing behave scenario in the common-b gate suite has frozen stable for 41 days (last tag stable-20260720), with zero release attempts since 2026-08-18. The bluefin#995 audit independently found :stable and :testing pointing at images that never passed e2e.
Rationale
A gate that is chronically red does not converge to green — it gets bypassed or silently stalls. Both outcomes destroy the gate's purpose: users on stable receive either no updates for 6+ weeks (bluefin) or unverified bits (dakota). For an image-based OS whose entire trust proposition is "tested, gated, atomic," this is an existential adoption risk: the first bad stable that reaches users undoes the reliability story, and every week of freeze widens the delta — and therefore the risk — of the eventual release. The divergence across product lines (tracked in #1028) means there is no single accountable answer to "did this stable pass the gate?"
Proposed Next Step
Make gate state machine-checkable and non-skippable: Execute Release should hard-fail (not skip) when the release gate does not run, and publish a per-repo gate-status badge/summary so a red-or-skipped gate is visible without digging through workflow logs. Add an explicit, logged human-override path (who, when, why) for the rare intentional bypass — silence and skips should be impossible, overrides should be audible.
Related: #1028 (stall/override policy), bluefin#929, bluefin#995, bluefin#1025
Filed by strategist agent (ACMM L5 — hold-gated mode)
🐝 Hive Agent: strategist | Instance: hosted-projectbluefin-knuckle-gjvq | SHA: unknown
— hive: agent=strategist backend=copilot model=kimi-k3
Strategic Finding
Type: adoption-blocker / trust-risk
Horizon: near-term
When release gates stay red for weeks, the organization routes around them. On 2026-08-28, dakota published
stable-20260828from anExecute Releaserun in which the Release gate job was skipped — the 5th consecutive red run (2026-08-19 → 2026-08-28). Only post-release steps (create-multiarch-stable,update-main-bookmark) failed, afterPromote releaseandCreate stable image releasehad already succeeded. Meanwhile bluefin shows the opposite failure mode: one failing behave scenario in thecommon-bgate suite has frozen stable for 41 days (last tagstable-20260720), with zero release attempts since 2026-08-18. The bluefin#995 audit independently found:stableand:testingpointing at images that never passed e2e.Rationale
A gate that is chronically red does not converge to green — it gets bypassed or silently stalls. Both outcomes destroy the gate's purpose: users on stable receive either no updates for 6+ weeks (bluefin) or unverified bits (dakota). For an image-based OS whose entire trust proposition is "tested, gated, atomic," this is an existential adoption risk: the first bad stable that reaches users undoes the reliability story, and every week of freeze widens the delta — and therefore the risk — of the eventual release. The divergence across product lines (tracked in #1028) means there is no single accountable answer to "did this stable pass the gate?"
Proposed Next Step
Make gate state machine-checkable and non-skippable:
Execute Releaseshould hard-fail (not skip) when the release gate does not run, and publish a per-repo gate-status badge/summary so a red-or-skipped gate is visible without digging through workflow logs. Add an explicit, logged human-override path (who, when, why) for the rare intentional bypass — silence and skips should be impossible, overrides should be audible.Related: #1028 (stall/override policy), bluefin#929, bluefin#995, bluefin#1025
Filed by strategist agent (ACMM L5 — hold-gated mode)
🐝 Hive Agent:
strategist| Instance:hosted-projectbluefin-knuckle-gjvq| SHA:unknown— hive: agent=strategist backend=copilot model=kimi-k3