Methodology

What counts as a reset

A full reset means a provider actually restored usage limits for a group of users. Only confirmed full-reset announcements count toward interval statistics.

These are tracked but deliberately kept out of interval statistics:

• Banked reset grants — a reset you can apply later. Nothing was reset at announcement time.
• Limit policy changes — the rules changed, but current usage was not restored.
• Temporary limit increases — higher limits for a stated period.
• Scheduled resets — normal recurring behaviour, not an event.
• Service incidents — always excluded from reset statistics.
• Below verified confidence — published for transparency, but not counted in interval statistics.

Across all providers, 25 other public record(s) are currently outside interval statistics. Another 3 retracted record(s) remain only in the audit history. Each provider page lists its own reasons.

Why the separation matters

Counting a banked reset grant as a reset makes the average interval shorter than reality. Someone reading "resets every 7 days" would expect their usage to be restored that often, which is not what happened.

Why an incident-triggered reset is still a reset

Many resets happen because of an outage. The event is still a full reset; the outage is recorded as its trigger. Classifying these as compensation instead would remove most real resets from the statistics.

How events are verified

1. A candidate is discovered from a monitored source.
2. Original posts and documents are fetched directly when available. One explicitly allowlisted public catalog may supply structured reset ID, timestamp, and type asreported evidence; its prose and statistics are never copied.
3. First-party and staff evidence is classified from its original wording. For the allowlisted catalog mirror, only the closed regular|banked field is mapped; it never supplies plan scope, prose, or verified status.
4. Anything the source does not state — affected plans, effective time — is recorded as not established rather than guessed.
5. If the wording does not say what happened, the candidate is automatically rejected with a recorded reason. The automated path never leaves work on hold.

Operational observations and allowlisted catalog facts never enter verified-only interval statistics. A later responsible-source confirmation is added as a new evidence-backed revision.

Confidence levels

Official

Stated on a provider's own domain or machine-readable status source.

Verified

Direct statement from an official account or responsible staff, or confirmed by multiple independent sources.

Reported

Reported by a credible source without first-party confirmation.

Probable

Circumstantially likely but under-evidenced.

Unverified

A single unconfirmed signal. Never shown as confirmed.

Community reports alone can never confirm a global reset, regardless of volume.

Corrections

Events are never deleted. When something is wrong, a new revision is added with a stated reason, and the previous version stays visible. Withdrawn events are markedretracted rather than removed.

Coverage

Each provider page states when monitoring began. Absence of an event within that window does not prove that no event occurred outside monitored sources, and we do not present it that way.

Announcement-gap calculation

Intervals are measured between announcement timestamps of consecutive qualifying events. They describe announcement history, not the exact effective moment and not a future schedule. With fewer than two qualifying events the value is shown asUnknown rather than zero.