AttestLayerAttestLayer

Automated Processing and Output Methodology

Version 2026-10-05 · Effective 2026-10-05Scope: Automated parsing, profile rules, response classification, source linking, customer review, signing, and verification, running in AttestLayer's own Google Cloud project. One narrow step uses an AI model (Claude, by Anthropic, in the United States): only the relevant record excerpts and the question text are sent to Anthropic's API, as listed under Subprocessors.Policy contact: support@attestlayer.comRelated: Terms of Service · Privacy Notice · Security and Data Handling

Buyer Review Pack uses bounded automated processing to transform customer-authorized records into draft client-review output.

AI points; it never writes. An AI model (Claude, by Anthropic) only selects sentences that already exist in your records. Every answer is a word-for-word quote, checked by fixed rules and by a second AI check, and nothing reaches your client until you approve it.

The workflow:

  1. parses the client request;
  2. identifies individual requirements;
  3. validates uploaded files;
  4. retrieves relevant excerpts from eligible records;
  5. asks Claude to select exact quotes and source locations;
  6. classifies each requirement;
  7. checks exact quotes with fixed rules and a second Claude model;
  8. generates package files;
  9. creates a canonical manifest;
  10. signs the issue receipt; and
  11. provides verification materials.

A requirement is classified Supported only where:

  • every asked fact has an exact quote in an eligible source record;
  • fixed checks and the independent verifier agree;
  • the exact source location is available;
  • authority and sharing mode are confirmed;
  • the answer contains only the required Yes/No prefix and verified quotes.

Customer confirmation required is used where records establish relevant facts but do not prove the client's exact requested representation.

After finalization, each answered row is marked from records (with its citation) or written by your organization (no citation). Rows you did not approve are left blank, exactly as your client sent them. Hidden, retired and header rows are never touched.

Not supported by submitted records is used where no submitted client-shareable record supports the requirement.

Review statuses

While you review

  • Supported. An approved record states the answer; the citation shows where.
  • Customer confirmation required. The records point to an answer but a detail is missing, out of date or contradictory; you confirm it.
  • Not supported by submitted records. No record states it; nothing is filled in.

In the final file your client receives

  • Each answered row is marked from records (with its citation) or written by your organization (no citation).
  • Rows you didn't approve are left blank, exactly as your client sent them. Hidden, retired and header rows are never touched.

The model never writes answer prose. The output must satisfy exact-quote, eligibility, fact-completeness, source-location, polarity, and independent-verification checks.

Customers must review every draft response before client submission.

Automated classifications are not legal, audit, regulatory, certification, insurance, employment, credit, medical, or compliance decisions.

See my price: free