DEMI Locate: evidence that a registry code finds the physical file across chats, projects and AI systems

PropositionsT1: A UPL code indexes file name, SHA-256 hash, registration time, and owner location note in the Registry | T2: The AI retrieves the record, searches its own chats by name then date, fingerprints the candidate file, and calls verify_fingerprint requiring all 64 hex characters to match | T3: One registration carries up to 50 other files (Worker V5.9.4) | T4: A 128-file folder scan distinguished 52 registered files, 27 proposed, 11 private, 15 missing, and resolved two files believed unregistered as other-file entries of older records | T5: Look-alike copies with names ending '(1)' were confirmed byte-identical via fingerprint | T6: The system does not establish authorship, ownership, or legal right

ConstraintsSHA-256 all-64-character match is the sole proof of identity; name matching is not proof | Chat search operates on words only, not on fingerprints or dates | Each Claude project is sealed; Claude cannot read ChatGPT chats and vice versa | A Word file stored in project Context has not been byte-checked | Location fails if no location note was written | GPTs-based Portal Builder assistant retires 11 Dec 2026 | Worker version V5.9.4 enforces the 50 other-file limit per record | The AI never registers, pays, or holds keys

MechanismsT1: Verify by reading the Registry record for a known UPL code and confirming file name, hash, and timestamp fields are present and consistent | T2: Run the six-step method on UPL-TECH-26-741263-EN; confirm chat found by name, fingerprint returns afb7261e…fb7e2485 matching block 26216980, result delivered in ~1 min 6 tool calls | T3: Load a record with 28 other files and confirm Worker V5.9.4 accepts up to 50 | T4: Re-run folder scan on the 128-file set; assert counts 52/27/11/15 and the two reclassified other-file entries | T5: Fingerprint both a main file and its '(1)' copy; assert hashes are identical | T6: Confirm no ownership or legal claim appears in any Registry output field

BoundariesOut of scope: authorship establishment, ownership claims, legal rights, payment, key management | Out of scope: cross-vendor chat reading (Claude cannot access ChatGPT history and vice versa) | Out of scope: fingerprint-based or date-based chat search | Not yet built: byte-check of Word files held in project Context | Not yet built: ChatGPT run of the same locate method | Not yet built: project or collection label on Registry records | Not yet built: AI-written session passage auto-registered at session end with session main file as Parent | OPEN: Will Word files in Context pass byte-check? Can ChatGPT replicate the method end-to-end? What collection-label schema should be added to records? How should auto-session registration be triggered and scoped?