FORECAST METHOD
How the Codex reset forecast works
One outlook asks whether at least one global reset or new banked reset will occur in the next 24 or 48 hours. The baseline, model values and historical peak all use the same deduplicated history.
What the percentage means
Global reset or a new banked reset — one combined outlook.
| public signal | score | what it establishes |
|---|---|---|
| Tease | 50% / 75% | A hint, not a promise. Naming Thursday does not make it a commitment. |
| Same-post independent corroboration | 52% / 77% | A second reading of the same tease, not a second tease added to the first. |
| Explicit future commitment | 83% | A definite promise of a global reset or new banked grant; timing unspecified. |
| Explicit commitment with a valid window | 93% | Both a promise and a correctly parsed, unexpired time window are required. |
These are display bands, not calibrated probabilities or notification eligibility. A verified landing is a fact and ends its future outlook. An unverified completion claim is not proof of a landing.
The cadence model, honestly
The joint model uses up to the last four completed gaps plus the ongoing wait: (completed gaps + 0.5) / (exposure days + 1). Pace uses the same total time divided by the completed-gap count; the unfinished wait is time, not another reset. The 48-hour value uses the same joint daily hazard. Global and banked probabilities are never added.
One banked grant counts once across its announcement, landing and updates. Redeeming, spending or expiring an existing credit is not a new opportunity. Explicitly linked global/banked rounds count once; nearby timestamps alone do not establish a link.
Comparable model coverage starts September 29, 2026 with published banked monitoring. Earlier supported facts remain in the audit; missing banked observations are not assumed to be zero. Unverified landing times are excluded. Monitoring accounts do not establish that every account received the same grant.
The no-signal display cap remains 46% / 71%; the model sub-line quotes the uncapped joint model. Teases hold for 30 hours and fade toward the baseline until 60 hours. Dated commitments fall to 83% during a maximum 12-hour grace, then expire.
inference Experimental, low confidence. Daily UTC walk-forward scoring evaluates at least one event in each 24h/48h window, using only evidence known at prediction time. Live sample counts and both Brier scores are in /api/forecast → reset_outlook; historical coverage is incomplete.
Legacy signal calibration (separate historical record)
This is a historical score-v1 snapshot dated 2026-08-27; it does not define Alert v3. It reports the pre-v3 samples exactly as observed, with no smoothing, extrapolation, or hidden misses.
| band | resolved n | hits | hit rate | open |
|---|---|---|---|---|
| dated commitment | 2 | 1 | 50% | none |
| plain promise | 0 | — | insufficient history | none |
| legacy tease (historical) | 0 | — | insufficient history | The historical snapshot recorded 1 open sample at score 50; this is not a current v3 score claim. |
| legacy additional-evidence watch (historical) | 0 | — | insufficient history | none |
Retained as an append-only pre-v3 snapshot; it is not rewritten into the current decision contract.
The curated timeline is the best public landing-time proxy, not direct per-account telemetry. A hit means a verified reset landed inside the declared window plus the published 60-minute grace.
When resets happen
The histogram and homepage peak use the same deduplicated global-or-new-banked sample and coverage period. UTC confirmation/observation hours are historical patterns, not an arrival schedule.
Global reset or a new banked reset — one combined outlook.
Boundaries
These limits travel with the forecast so an excerpt does not overstate what Codex Reset knows.
- These are display bands, not calibrated probabilities or notification eligibility. A verified landing is a fact and ends its future outlook. An unverified completion claim is not proof of a landing.
- Comparable model coverage starts September 29, 2026 with published banked monitoring. Earlier supported facts remain in the audit; missing banked observations are not assumed to be zero. Unverified landing times are excluded. Monitoring accounts do not establish that every account received the same grant.