How this archive works

Announcements are not individual account resets

We archive public announcements and the quota actions they describe. We do not access your account, know whether your quota changed, or infer the time you use a banked reset.

• Announcement: the source post or observation. A single announcement can describe two actions.
• Action: a stable event ID with its type, evidence and revision. Different source copies are evidence, not automatically new actions.
• Verified global reset: scope and action have supporting verification. The old regular label alone is insufficient.
• Banked grant: issuance of a saved reset. The user’s later use is a separate account action. Imported classifications remain labelled unverified until reviewed.

How banked resets work · Personal reset timer

Which time does each field describe?

• announced_at — publication/announcement time used in historical intervals, shown in UTC.
• effective_at — when the quota action took effect, only when known. Unknown stays null; publication time is not substituted.
• verified_at — when a verification was recorded. It is not a reset timestamp.
• coverage.checked_at — coverage check time, not proof of a new event.
• meta.generated_at — first creation of this frozen snapshot. Page refresh does not make old source data fresh.

Time precision is explicit. An unknown hour never becomes midnight or a data point in an hourly distribution. The personal calculator uses your selected IANA time zone and handles ambiguous daylight-saving times separately.

Today’s answer and missing coverage

The tracker distinguishes a recorded public action, an announcement awaiting completion, no record under established current coverage, and unknown/expired coverage. An announcement about a future window is not a completed action.

A successful HTTP request only confirms transport. It does not prove recent content, complete scope or that the event happened. Historical sources leave required coverage only when an equivalent current replacement has supporting evidence.

If any required scope is missing or stale, we keep the last known records and say coverage is unknown. We never turn a timeout or empty feed into “No reset”.

Current snapshot unavailable

The definitions remain readable. No statistics are fabricated while the data endpoint is unavailable.

Interval formulas and denominators

For a defined collection, sort its eligible actions by announcement time. For n records there are n−1 adjacent completed intervals: (tᵢ − tᵢ₋₁) / 86,400 seconds. Each collection has its own range, count and definition.

We show mean, median, minimum/maximum and percentile ranges. A mean is not a conditional probability. The legacy combined collection mixes regular records and banked grants and must not be advertised as a global reset cycle.

Two actions from the same post may have a zero announcement gap. We retain the relationship rather than deleting an action or claiming instantaneous consecutive global resets.

No claim of a meaningful weekday pattern or a useful next-reset probability is made from these small, selected announcement samples. A prediction experiment would first require cleaned event definitions, saved forward predictions and comparison with a baseline.

Recent Tibo signals

Recent captured author posts can describe a future reset, a reset hint, subscription reopening, a usage policy change or a launch. We display the strongest matching category per post with its original link, publication time and source check status. Policy and credit context is a reason to watch for updates, not evidence of a completed reset.

The mirror capture retains up to 4,096 characters of author text and excludes embedded quoted posts. Classification is based on this captured text and may miss replies, ambiguous wording or omitted text. We keep the author’s “today” or “tomorrow” without inventing a UTC execution time. Signals expire from the active display 72 hours after publication; this is a display window, not a forecast deadline. The posts remain in the timeline.

This reading aid assigns no reset probability and does not insert archive actions or send reset notifications. Later captured completion messages resolve earlier reset plans and hints. Source failures keep dated evidence visible with an unavailable or overdue label.

Corrections, merges and revisions

Stable IDs and source evidence remain traceable. A source-text or classification change adds revision evidence. A merged record remains visible as a relationship rather than disappearing without explanation. “Needs review”, “disputed” and “retracted” are distinct from verified.

Unresolved legacy classifications are disclosed. A lack of revision records does not certify earlier data. Frozen snapshots and exports allow a reader to see what an earlier published version contained.

Attribution and reuse

You may reuse WhenResets-created metadata, classifications and correction notes with attribution to WhenResets and a link to the snapshot/version you used. Preserve the coverage, verification and time-basis fields.

Original posts and source excerpts belong to their authors. Public access to this archive does not grant a new license to republish third-party text, images or datasets. Follow the original source’s terms and keep source links.

Do not label source-classified or stale data as official confirmation. This is an independent archive with possible omissions; it is not an OpenAI account API.