Collection and editorial methods
How material enters the record, what a capture establishes, how prominence is decided, and where coverage remains incomplete.
Purpose and unit of record
This project collects public material about the July 2026 COLDCARD predictable-RNG incident. A registered source is normally one stable URL, social post, or discussion thread. Its record page identifies the original publisher, publication time where known, every state held, the capture method, and reviewed changes between states. Registration means the item is part of the record. It does not mean the project endorses it or that it must appear in narrative prose.
Discovery and selection
Known URLs are polled on schedules recorded in the source registry. Separate discovery lanes inspect selected community listings, watched X accounts and the signed-in X home timeline. Candidates are hydrated by the driver before an intake agent reads them. Agents do not browse for their own evidence and cannot write the archive.
Intake considers several questions separately:
- whether the item is a primary statement, first-hand account, technical analysis, reporting, or materially significant reaction;
- whether it adds evidence or context not already held;
- how close it was to the event and how likely the original is to change or disappear;
- whether its provenance is clear enough to preserve responsibly; and
- whether a crosspost or repeated account adds its own discussion or distribution context.
Selection for preservation is deliberately broader than selection for prominence. The register holds useful incident material that may never be quoted elsewhere. Landing, timeline and explainer placement require a stronger contribution to orientation, chronology, a disputed question, or an evidenced explanation.
Capture and change history
A capture records what the collector received from a URL at a stated time. Text extraction is deterministic and the capture metadata records hashes, response details and provenance. Every poll appends an index entry whether the normalised text changed or not. A changed state creates a new snapshot and difference; earlier states remain in place. The append-only rule applies from 6 August 2026, with the earlier Reddit migration exception disclosed at /corrections/ and in the citation guidance.
Differences are classified as source content, capture noise, or capture correction. Classification says what kind of change the collector saw; it does not verify claims in the new text. Public source pages show short excerpts and differences. Full third-party captures remain local rather than being published as mirrors.
Editorial treatment
The record reports and attributes. When parties disagree, it presents each account with its date and assumptions and does not choose a winner. Material claims use the evidence bases defined on the About page. The project does not publish independent firmware findings, comparative rankings, legal conclusions, or guidance authored by the archive itself. Its current editorial boundary is recorded in the public repository.
Coverage limits
- Discovery began after the story broke, so early material may be recovered later and its first archive capture may post-date publication.
- Community discovery samples bounded listing and search surfaces. Deleted, private, unindexed, paywalled or never-linked material may be absent.
- X discovery depends on a signed-in read-only session and the platform's rendered results. A watched-account list is not a complete index of X.
- Nostr discovery is probationary, and podcasts or video are normally represented by publisher metadata, transcripts or show notes rather than broad media mirroring.
- Publication time and capture time are distinct. A recovered Wayback state is labelled with Internet Archive provenance.
Current evidence gaps are gathered at Known limits. Suggestions of missed public material and corrections are welcome through the contact route on the About page.
Reproducibility and citation
The source registry, reviewed change ledger, capture tooling and site are public. Machine-readable source and change records are published at /record/sources.json and /record/changes.json. Each build names its repository commit at /version.json so a reader can identify the state used. A held capture establishes publication, not truth; the citation guide explains how to carry that distinction into reporting or research.