nx_extvec_authentic_gate.nx

nx_extvec_authentic_gate.nx -- THE GUARD THAT WOULD HAVE CAUGHT IT ON DAY ONE, PLUS THE LEDGER. ★★★★★ THE LAW THIS GATE ENFORCES: A DIGEST PIN PROVES INTEGRITY-SINCE-FETCH, NOT AUTHENTICITY-AT-SOURCE. Every external-vector gate pins its source document to a SHA-256 computed in-process at the socket and treats "PIN OK" as if it established that the bytes ARE the authority's document. It does not. A pin proves only that the bytes have not changed since we hashed them. A corrupted, truncated, or -- as actually happened on 2026-07-31 -- a fetch that stored the raw HTTP WIRE TRANSCRIPT instead of the document, all pin exactly as cleanly as a good one. The pin is self-consistent BY CONSTRUCTION, so no amount of re-checking the pin can ever reveal the fault. ★INDEPENDENCE NEEDS A SECOND STACK, NOT A SECOND HASH. ⚠THE DEFECT THIS EXISTS TO PREVENT: knowledge/extvec/rfc2202.txt began HTTP/1.1 200 OK\r\nDate: ...\r\nTransfer-Encoding: chunked\r\n ... \r\n\r\n2ea6\r\n -- 707 bytes of response headers plus chunk framing -- because nx_vecfetch called the raw nx_https_get, which BY CONTRACT returns the entire response, and never stripped headers or de-chunked. All 14 pinned "authority documents" were transcripts. The shared decoder (nx_fetch_staged / fu_dechunk in nx_fetch_unit.nx) already existed and was not adopted. ★THE ADOPTION GAP, NOT A MISSING PRIMITIVE. TWO INDEPENDENT CHECKS, AND THEY FAIL FOR DIFFERENT REASONS -- THAT IS THE POINT: A. SHAPE -- is this file a document rather than a transcript? (content predicate, no external input) B. LEDGER -- does an independent stack agree on this exact file's digest? (corroboration, external) A catches the accident. B catches the substitution. A file can pass A and fail B (a DIFFERENT but well-formed document swapped in), or pass B and fail A (both stacks fetched the same transcript). ★A SINGLE CHECK THAT COULD HAVE CAUGHT BOTH WOULD HAVE CAUGHT NEITHER CLEANLY. ⚠HONEST LIMIT OF THE LEDGER, STATED SO NOBODY OVER-TRUSTS IT: CORROBORATION.tsv is a plain file this agent can write. It raises the bar -- swapping a document now requires forging a matching "independent" column, at which point that column is no longer independent and the forgery is recorded in-tree rather than invisible -- but it is NOT a transparency log. The July-2026 answer to that is Sigstore/Fulcio/Rekor or in-toto/DSSE: an append-only log held by a party we cannot edit. Nishi already has nx_coe_dsse / nx_coe_intoto / nx_coe_witness and this lane has NOT adopted them. That is the next rung, not this one. license_tier: ORIGINAL expect_exit: 0