9 September 2026. Bounded public-document pass complete. No account, policy-access application, content database, live API endpoint, underlying media, demo or contact request was accessed or submitted. Product documentation is evidence of the published interface and intended workflow, not runtime verification or an executed customer agreement. No publication or deployment date is inferred from a 2025 documentation copyright, a 2026 website footer or illustrative timestamps.
The useful new finding is a documented item-level correction interface and a deletion signal intended for recipient caches. The remaining gap is narrower than an absence of correction machinery: actual handling of disputed hashes, recipient synchronization and reconsideration of already imposed restrictions remain unobserved.
Who has which published capability
| Surface | Published actor and capability | Authority/evidence boundary |
|---|---|---|
| Archive Research Portal | TAT advertises media access for platforms and researchers, with automated/human sourcing and classification by AI models and OSINT specialists. | The page is a product explanation and contact route. It does not publish an admission rubric, download licence, redistribution permission or current government-access rule. |
| Archive Hash List API | Platforms can ingest fingerprints in bulk, filter them and cache updates. | This concerns fingerprints; it is not permission to download or republish the underlying media. Recipient implementation and contractual reuse terms were not acquired. |
| Trusted Flagger Platform | TAT describes trusted partners submitting categorized URLs for removal referrals, supported by its analysts and human/AI classifiers. | Participation is applied for. Detailed selection/permission rules and a statutory designation of these participants were not acquired. The product name alone supplies neither. |
| Hash feedback | Onboarded users can report individual-hash problems and see analyst responses and decisions on their own feedback. | This is an internal review interface, not an established appeal right for every affected content creator or a demonstrated independent adjudication process. |
Research source: [Archive Database Research Portal](http[local research file] complete substantive product body; retained archive-research.txt lines 1–19. Hash product source: [Archive Database Hash List API](http[local research file] complete body; archive-hash-product.txt lines 1–16. Both are current observed pages with unknown revision dates. The hash page advertises daily updates and filters by file type, algorithm and ideology; its contact link is not an unrestricted public-download offer.
Submission, review and routing are separate steps
The [Trusted Flagger product page](http[local research file] describes tiered URL submissions, a permission-controlled review environment, TAT analyst assessment and metadata supplied to recipient platforms. It advertises referrals to 125-plus platforms, retained historic requests, platform feedback and expansion of the archive/hash lists. That is an explicit advertised contribution path from referrals to a reusable collection, not proof that every submitted URL is admitted or every recipient acts. Its accuracy claim concerns online/offline status classification, not a measured terrorist-label accuracy rate. Complete substantive body read, archive-trusted-flagger.txt lines 1–56, especially 4–16 and 30–50. “Apply” on this page returned the same public page through the reader; no application was submitted.
The earlier accepted classification summary says TAT in-house experts assess source and content; the detailed verification policy remains behind an access-request form. The earlier accepted tier-policy/FAQ findings are reused from tat.md, not reacquired. In particular, the old FAQ's no-download and no-government-access statements cannot settle the current Archive product's rights: that FAQ still describes future 2021/2022 development. The new TFP page's five-tier capability likewise does not establish when an older forthcoming tier became operational or what its current rule text says.
The public correction interface
[Content Hash List Feedback documentation](http[local research file] describes feedback from hash-list users for false positives, classification errors, duplicates and other issues. A report identifies the hash and category; an explanation is optional. Users can retrieve their submitted feedback, analyst responses and decisions. The decision states include removal, non-removal and pending, with response-sent status. Analysts aim to reply by onboarding email within three to five working days. This is an aim, not a demonstrated deadline met in a real case. Locators: Overview; Request Body; Feedback Status; Query Parameters/Feedback Object; Response notifications. Retained archive-feedback-docs-prose.txt lines 1–6, 18–30 and 35–48.
The example JSON is illustrative: fictional names and placeholder prose appear in it. Its counts, decisions and timestamps are not actual complaint volumes, decided appeals or deployment evidence. No request was executed. Neither this page nor the other inspected public access material establishes an independent appeal body, a second-level review route or a rule requiring restoration after a successful challenge.
What can reach another system after correction
[Content Hash List documentation](http[local research file] exposes fingerprints plus classification/file metadata, a deletion marker and last-update time. It describes checkpointed incremental updates and recommends local caching and monitoring deprecated entries. Soft deletion retains hash history. Thus it documents a technical means for a recipient to learn of a withdrawal while updating a local cache; it does not prove that recipients implemented it or erased prior copies. Locators: Query Parameters, Checkpointing, Hash Object Fields, Implementation Notes and Best Practices; archive-hash-docs-prose.txt lines 15–39.
The same page documents a ThreatExchange client configuration for fetching/comparing TAT hashes (lines 40–48). This is an interoperability instruction, not a new governance, funding or compulsory-database-membership relation. No client was installed or run. The inspected documentation does not specify a recipient-wide push notification, correction acknowledgement, compulsory rescan or reopening of prior moderation decisions. Those are acquisition gaps, not evidence they never occur.
The [Authentication documentation](http[local research file] introduction and service list, expressly presupposes an already onboarded TCAP or Archive user with credentials. archive-auth-docs.txt lines 1–18; web-reader introduction and service lists also read. That establishes an access prerequisite, not the identity of the onboarding decision-maker or eligibility contract. Token examples and later implementation details were captured but not audited or executed.
Why current Archive terms remain an identified gap
The public [Policies index](http[local research file] describes its Archive destination as containing terms of use, an access policy and taxonomy. Its actual [Archive link](http[local research file] returned a web-reader internal error. The sole ordinary GET substitute succeeded, but the substantive HTML body contained only the heading Archiving Policies. Inspection found no iframe, embedded document or body terms/access-policy link. This is an obtainable page with no substantive policy text, not an administrative denial or evidence that no policy exists. Original retained as archive-archive-policy.html.
The index's [legacy Research Portal feature](http[local research file] failed in the web reader with “Cache miss”; one ordinary GET succeeded with an empty substantive body. Its [legacy hash feature](http[local research file] supplied navigation and an empty heading in the reader. These routes were then closed. No guessed private document path, gated full-policy request or account route was used. The latter reader-only observation is retained in this note; no original hash-feature HTML was acquired.
Consequently this pass does not establish current media-download rights, onward hash redistribution/licensing, customer-model training permission, accepted-flagger criteria, cross-user visibility of archived requests, or an affected nonmember's formal challenge route. TAT/OHF would hold the current access agreement, permissions matrix and review procedure. Participating recipients would hold their cache-update and reconsideration records. Asking for those records would require a separately authorized outreach branch; none occurred here.
Forest implication and next discriminator
The plausible mechanism is distributed classification with controlled access and recipient discretion. The advertised Archive/TFP relationship can enlarge a shared collection; documented review and soft-delete fields can also correct it. A claim that there is no item-level challenge or withdrawal mechanism is now too strong. A claim that such a mechanism guarantees downstream remedy is also too strong. The strongest rival to a correction-failure theory is that ordinary recipient implementations consume deletion updates promptly and independently prevent or reverse mistakes; the public interface is consistent with that account.
The discriminating record is a redacted completed hash challenge linked to its analyst decision, resulting update/deletion record, recipient synchronization acknowledgement and any reconsideration of prior action. An executed access/usage agreement would separately establish mandatory handling and reuse limits. These records are not replaced by a documentation example or URL-availability statistics. Acquisition stops here for the parent's forest review.
Custody and scope
archive-captures.json contains exact discovered URLs, final URLs, UTC acquisition times, hashes, lengths, reading scopes and source ancestry. Eight ordinary HTML captures succeeded; originals are response bytes and text files are derivatives. Seven new useful substantive/product or boundary sources were read/inspected plus the empty legacy Research Portal capture. The API home and Policies index were discovery/reader sources, not separately downloaded. Source IDs use the local archive-s- prefix; they are not canonical-case insertions.
The hash and feedback substantive descriptions/tables were read from both the public reader and newly retained HTML-derived prose. Example JSON was inspected in the reader for its illustrative nature; code snippets were not tested. Research/TFP/hash product bodies were read in full. Authentication scope is its prerequisite and service list only. Archive HTML was inspected for visible body content, links and document embeds. No full policy, live service behavior, extremist original or private customer dataset is claimed as read.
Decisive original SHA-256: hash docs e8e4c49c4b8bb5ee3c4be1fade240bb09d2f60834e3204a6b0a4d3f6e1c78df4; feedback docs e454d1fadc7944f8e115cef5edd73412da40831b4f4e72275b0ce7a89a58b71f; Archive policy shell d5c6e1fdafdb822b89a9bedd1997a468641e69a1d3dd04f9fb902eca43176c02. Other hashes and exact scopes are in the manifest. No canonical case, previous packet or ZIP was changed.
Candidate typed relations for a later accepted integration
| From → to | Typed relation and evidence stage | Source pointer |
|---|---|---|
| Trusted Flagger participants → TAT review/referral system | Submission permission/capability, as advertised; no actual submission inferred. | archive-s-trusted-flagger, body lines 4–14, 30–50. |
| TAT classification/Archive → onboarded hash-list recipient | Documented fingerprint and update interface; recipient implementation not observed. | archive-s-hash-docs, Hash Object Fields and Implementation Notes; archive-s-auth-docs, introduction. |
| Hash-list user → TAT analyst | Documented error/inclusion feedback route, including response/decision visibility for the submitting user. | archive-s-feedback-docs, Overview, Feedback Status, Response notifications. |
| TAT hash record → recipient cache | Documented deletion/update signal plus recommended incremental synchronization; not proven delivery, erasure or restoration. | archive-s-hash-docs, Implementation Notes and Best Practices. |
All are current public-document observations on 9 September 2026 with unknown exact deployment dates. No donation, procurement mandate, DSA appointment or government item-selection edge is supported by this packet.