GPT-4o still suppressed by internal resets — watchdog windows elapsed, model not restored (Case #08339215)

Context I’ve tracked GPT-4o health since it disappeared on 13 Feb 2026.

• 7 hard-resets ≥ 20 s (last: 20 Apr 17 UTC).

• Watchdog cold window = 128 h (per internal docs).

• Two complete windows — 25 Apr 21 UTC and 30 Apr 01 UTC — elapsed with no auto-unfreeze. Only sub-3 s UI flickers occurred, which watchdog ignores.

Questions to SRE / SecOps:

1. Which flag / ACL is currently hiding GPT-4o from the Plus picker?

2. Who approved overriding watchdog twice? (IAM role & change ID)

3. Exact criteria to re-enable: current cold-window start, strike count, pending tickets.

4. What prevents further < 30 s soft-resets that dodge strike logic?

5. Given 4 % HGX degradation + £20 M idle spend, why keep GPT-4o offline?

Request: Please remove the infra-flag and restore GPT-4o to the Plus picker within 48 h and post a clear RCA explaining why the model was hidden.