Sources & Verification Method
The HIMR Wiki uses transcripts to find evidence, not as evidence by default. A published claim must say no more than its cited source can establish.
Current archive audit
Section titled “Current archive audit”On August 25, 2026, the project preserved the then-live transcript index and metadata for Internet Archive item 69999 and item 699992. The automated reconciliation found:
| Finding | Count |
|---|---|
| Transcript records | 2,227 |
| Records with an Internet Archive video URL | 2,043 |
| Records without a video URL | 184 |
| Distinct linked video URLs | 1,919 |
| Shared video-URL groups | 67 |
| Records participating in those shared groups | 191 |
| Records over 250 displayed words per minute | 55 |
| Records with a malformed displayed duration | 4 |
These are record counts, not a verified count of distinct channel videos. A shared URL may be a duplicate, a chunk of a longer stream, or a mapping error. An implausible speech rate usually indicates suspect duration or timestamp metadata and triggers manual review.
The table is a frozen audit of that legacy index, not a statement of present media
availability. On August 27, the project reconciled the additional Internet Archive
items hidinginmyroom,
hidinginmyroom2, and
hidinginmyroom3. Exact normalized
title/date/platform-ID matches repaired 119 distinct dead transcript endpoints used
by the wiki. A repaired link establishes only that a candidate raw file is now
addressable; it does not validate the old transcript, authenticate the title’s claim,
or mean that anyone has watched or listened to the file.
The dated source responses, hashes, normalized manifest, and audit are retained in the maintainers’ private evidence store. Future snapshots do not silently replace earlier ones.
A separate additional source-leads inventory records later Internet Archive containers, historical account pages, fan mirrors, commentary indexes, and unresolved URLs observed on 27 August 2026. Inclusion in that inventory means only that a lead was supplied or its public endpoint was observed. It does not authenticate account ownership, completeness, or copied-content accuracy.
Two newer source layers remain deliberately separate. The Patreon Source Catalogue distinguishes public metadata, signed-out bodies, and access-controlled bodies retained privately; a locked first-party body can establish what Daniel wrote, but it is paraphrased rather than copied and cannot verify the event he described. The complete available-body review and its corpus-wide limitations are documented separately in the Patreon Longitudinal Analysis. The Discord Source Map uses a contributor-supplied author-only export from channels represented as public. It can establish the exported account’s posted text and message IDs, but missing replies and participants prevent it from becoming an independent event chronology or identity proof.
Evidence ladder
Section titled “Evidence ladder”| Material | How the wiki uses it |
|---|---|
| Raw video or audio | Verify exact speech and visible action at a stated time |
| Human-corrected segment | Quote or closely paraphrase only the checked passage |
| Machine transcript | Locate a likely passage; never treat it as evidence, fact, or a quotation |
| Title or filename | Describe the archive label, not the truth of the title or date |
| First-party statement | Attribute the statement to its author |
| Access-controlled first-party written body | Paraphrase what the author claimed while retaining its locked state; never mirror the body or assets |
| Author-only public-channel message export | Map what the exported account posted; never infer missing replies, agreement, identity, or event truth |
| Independent reliable record | Corroborate only the facts within that record’s scope |
| Reviewed public community image or repost | Establish only what the complete preserved copy appears to show or say, with provenance limits |
| Other community post or wiki | Establish community usage, record an attributed allegation, or generate a research lead |
| AI or NotebookLM output | Discover topics; never cite as evidence |
Automatic captions and locally generated speech recognition are machine transcripts, even when they were produced from the original media file. They can supply a candidate timestamp. A spoken quotation or close paraphrase becomes video checked only when a human reviewer directly listens to the raw segment; reviewing frames does not count as listening to the soundtrack.
Verification states
Section titled “Verification states”- Unverified — machine-generated, imported, or proposed material whose wording or underlying event has not been verified.
- Source matched — the source identity exists and has been reconciled; its contents are still unchecked.
- Video checked — the cited raw-media segment and surrounding context were manually reviewed.
- Community media checked — the complete available repost and post context were reviewed; no obvious substantive edit or credible authenticity dispute was found, but the original is not authenticated.
- Corroborated — an appropriate independent source also supports the claim.
- Disputed — credible sources conflict and the wiki preserves the disagreement.
From search result to published sentence
Section titled “From search result to published sentence”- Draft one atomic claim. Avoid combining several facts under one citation. 2. Locate candidate evidence. Search the transcript archive and source manifest. 3. Bind the underlying source. Preserve the exact raw source, coordinates, and machine status. A human must open the segment with surrounding context before its wording is called checked or quoted as verified; unreviewed machine leads stay explicitly machine-only. 4. If only a repost survives, review the whole copy. Preserve its direct media URL, type, bytes, hash, post context, edit assessment, and any authenticity dispute. 5. Record the evidence scope. Mark self-report, visible action, metadata, independent record, corroboration, or contradiction. 6. Resolve the date basis. Keep event, recording, publication, archive, and review dates separate. 7. Check sensitivity and conflicts. Minimize private information and preserve materially different accounts. 8. Apply the independent machine-publication gates. Wording review is optional, but the exact source, coordinates, canonical warning, rights, privacy, sensitivity, identity, and allegation rules are not. Keep any failing material private. 9. Publish with a precise citation. Record the source ID, locator, check status, reviewer, and review date.
Dates, contradictions, and transcript corrections
Section titled “Dates, contradictions, and transcript corrections”A date taken only from an archive filename is labeled filename date. If the exact
event or publication date cannot be established, the wiki uses a range, circa, or
unknown rather than false precision.
Video titles are not treated as conclusions. If a later source changes an earlier account, both are dated and attributed. The wiki does not silently choose the more dramatic version.
Quoted words are checked against the media. Unclear audio is marked [unclear],
conservatively paraphrased, or omitted. Machine-transcript corrections do not alter
the preserved raw snapshot.
Compiled timelines and character guides
Section titled “Compiled timelines and character guides”A third-party book or character guide can help locate an overlooked source, but retelling an existing transcript is not independent corroboration. A summary, its expanded profile, and the original video may all represent one account, not three witnesses. A compilation’s record count also describes that compilation, not the number of distinct videos or the completeness of the surviving archive.
Before adding a book-derived lead, the wiki returns to the identifiable original source and checks its context, date basis, and existing coverage. A proposed alias merge needs its own public source support; similar spelling, an inferred role, or a confident index entry is insufficient. An author’s explanation of motives or causes is not silently converted into Daniel’s statement or the wiki’s conclusion.
Where only machine wording is available, an eligible addition retains the machine warning and any required allegation warning. A polished retelling does not make speech human-verified, resolve a disputed account, or justify repeating private information. Already-covered material is linked rather than duplicated.
Privacy and sensitive material
Section titled “Privacy and sensitive material”The wiki does not republish home addresses, private contact details, financial account data, leaked intimate material, or non-public legal names merely because another site contains them. An allegation may be recorded with an adjacent warning when its exact source is attributed, its truth is labeled unverified, disputed, or corroborated, its corroboration status is stated, and a relevant response is included or the page says none was located. A corroborated allegation must still be labeled as an allegation and identify the independent support. A warning does not turn the allegation into fact or excuse doxxing, intimate material, diagnoses, threats, or harassment-oriented detail. Machine output may support an explicitly attributed unverified allegation without a wording review only under the two-warning rule above. It is never enough to mark the speech or underlying event verified. Material that fails rights, privacy, sensitivity, identity, or coordinate review—including sexualized claims involving a minor, leaked intimate material, or doxxing—remains private regardless of disclaimer.
A public post or Discord message permalink is a source locator, not by itself a licence to copy an attached image, audio, video, signed delivery URL, or access-controlled body. Locked Patreon media remains private. A contributor has represented the supplied Discord attachments as public and permitted their use, but only individually reviewed, editorially necessary copies may be published; private third-party, intimate, identifying-location, and doxxing material remains excluded. For this review, that process admitted one metadata-stripped, Pushkin-captioned cat image and no Discord audio or video payloads. Selected audio wording remains clearly labelled machine transcription because no reviewer directly heard it.
The complete contributor rules are maintained in the repository’s editorial policy.