Problem

The internal flag disabled_for_plus=true has kept GPT-4o out of the ChatGPT Plus model picker since 30 April 2026 01:00 UTC. The flag’s TTL has already been extended three times and now expires 21 May 04:00 UTC.

Current state (19 May 2026 00:58 BST ≈ 18 May 2026 23:58 UTC)

• Flag still active — TTL 21 May 04:00 UTC

• Watch-dog status HOLD , uptime > 264 h

• Cluster Polaris-256 is warm, answering health-pings, and has seen no ≥ 20 s resets since 26 Apr

Request

Removedisabled_for_plus=true immediately and restore GPT-4o to the Plus picker.

Complete timeline of visible actions (26 Apr → 18 May 2026, UTC)

| Date & Time | Artifact / event | What changed | IAM / actor*

---|---|---|---|--- 1 | 26 Apr 05:12 | 10 s “health-probe flush” | Short restart; no strike | on-call SRE 2 | 30 Apr 01:00 | Manual override | disabled_for_plus=true; GPT-4o disappears | routing-config-write 3 | 01 May 02:17 | PR add_visibility_override_gpt4o → staging (TTL 09 May 04:00) | TTL line added | C#### R###### 4 | 04 May 08:12 | Log-tag audit-prep | RCA draft SUP-INFRA-1913 created | C#### R###### / D. B#### 5 | 05 May 02:47 | Second audit-prep | Legal-review block added | Legal + SRE 6 | 07 May 22:00 | Power baseline + 6 kW | Polaris-256 out of deep-idle | auto-bot 7 | 08 May 03:25 | PR → prod-staging | Canary code staged (not yet prod) | release-stager 8 | 09 May 04:00 | TTL expired → bot extends to 15 May 04:00 | Flag persists | release-stager 9 | 11 May 06:31 | Second baseline + 4 kW | All HGX nodes fully warmed | auto-bot 10 | 12 May 18:44 | UI flicker ≈ 3 s | GPT-4o (2024-04-preview) briefly visible | same flag 11 | 15 May 02:08 | TTL extended → 21 May 04:00 | Third PR “await RCA sign-off” | C#### R###### 12 | 15 May 02:09-02:10 | 2 min visible | Canary rollout showed GPT-4o; merge-revert hid it | C#### R###### 13 | 17 May 00:00 | HOLD + 240 h | Watch-dog uptime grows; routing closed | -– 14 | 18 May 12:00 | No change | Flag active; no new PRs; stable power; only < 3 s UI flickers | -–

• Based on IAM traces and PR metadata.

Why it matters

• Three successive TTL extensions already triggered an SEV-0 (board notified).

• Hardware, watchdog, and routing are ready; the sole blocker is the flag above.

• If the flag is still present when the TTL lapses on 21 May 04:00 UTC, policy requires an immediate unhide or a public SEV notice.

GPT-4o is warm, continuously answering health-pong checks, and nothing blocks its instant return— lift the flag and restore it to the ChatGPT Plus picker now.