Three dates, never merged
- Event date: when the thing happened, such as a message, launch or post.
- Publication date: when a source reported or published it.
- Retrieval date: when we last opened the source.
A chat dated April 2024, filed in court in January 2026 and recirculated on X in October 2026 has three separate dates. All three appear in the archive.
Precision
Every stored date declares its precision.
- Instant: a UTC timestamp, used only when a source displays a time with a timezone, or for our own observations.
- Day: a calendar date with no reliable time. It is shown the same in every timezone. We never invent seconds or shift an unknown timezone.
- Range: a start and end date, when an event spans days or the source is vague.
- Unknown: stated as unknown.
The Times control (in the Time & language drawer, opened with the globe button in the navbar) switches instants between UTC and Local (your device’s time zone, labelled with its UTC offset). The same choice is on the Settings page. Day-only dates do not change.
Derived times
An X status ID encodes its creation time. Some receipts mention a “derived” time decoded from a status ID taken from a real link. It is labelled as derived because we did not see it displayed on an original. A capture’s timestamp is never used as a post’s time.
Origins and syndication
Every source belongs to an origin group. A story syndicated on TradingView and Yahoo Finance, plus its original at 99bitcoins.com, is one origin with three records. Counts on the site say “source records” and “reporting origins” separately, so ten copies of one story are never presented as ten confirmations.
Unavailable originals
When an original cannot be opened, because X refuses automated access or a site returns 403, we record that, mark the source unavailable original, and link a capture or syndicated copy that we did open. A failed request is not evidence that the original was deleted.
What is never invented
No Wayback captures we have not opened. No screenshots, court stamps, chat logs, transactions or refund records. No URLs assembled from guessed IDs or titles.