Canonical URLs: Full §02 · Raw Markdown · Plain-Text Brief · ← §01 Brief · §03 Brief → · All §01–§09 Master Brief · All Articles · Raw Markdown Executive Synthesis: Problem: Hostile documents can attack the parser, while models can fabricate citations and make unknown coverage disappear. — Proves: LPAC parser confinement, immutable views, explicit omissions, and server-cut srv_ and ev_ receipts establish served evidence. — Evaluation Scope: Use the Zero-Trust Evidence & Citation Audit Challenge on one bounded document question. Lossless Parts (4 total, 100% unabridged): Part 1/4 (.txt) · Part 2/4 (.txt) · Part 3/4 (.txt) · Part 4/4 (.txt)

  • Introduction & Thesis [Part 1/4]: # Zero-Trust Document Ingestion and Server-Cut Evidence Hostile bytes need a parser jail. Enterprise judgments need a server-computed evidence chain. Neither a RAG chunk nor a model-written citation can supply one. **By Blayne Aina · Principal Systems Architect and creator of…
  • The Evidence World Is a Governed Artifact [Part 1/4]: A conventional retrieval pipeline often looks like this: That pipeline can be useful for search. It does not establish an execution-grade evidence world. A chunk does not prove which source revision produced it. A flattened string does not reveal what the parser dropped. — Spec/Table: The path ends in .pdf, or the response declares a PDF media type.: The bytes are a PDF, or the correct parser was selected. | The parser returned text without crashing.: Every text-bearing part was recovered, or every warning was benign. | The extraction is complete for its declared profile.: The source is complete, true, visually understood, or free of excluded content.
  • 1. Registration: Bind Identity to the Bytes Actually Read [Part 1/4]: Subtopics: Single-open capture; Amber retention is a separate observation; Drift is not a new version of the old citation — The evidence chain begins before parsing. doc_source registration resolves the caller through the Room's actor rules, requires an owner-opened project, canonicalizes the target through the same file-identity machinery used by governed writes, and refuses controller-state and… — Key Rules/Claims: The extraction profile recorded in the parser identity is text/v1. | The task's governing-source reading policy is complete_text_v1, backed by the required source-reading coverage proof.
  • 2. Content Decides the Format—Inside the Jail [Part 1/4]: Subtopics: Fourteen budgets govern expansion — Format identification is itself parsing. A ZIP central directory must be inspected before the system can know whether a package is a DOCX, an XLSX, or an ordinary archive. — Spec/Table: Pdf; Content or structural identity: %PDF- signature; Extraction or typed outcome: Structural PDF extraction, layout, annotations, and admitted local OCR paths | Json; Content or structural identity: .json or .jsonld tie-break after higher-priority signatures; Extraction or typed outcome: UTF-8 JSON document with bounded RFC 6901 source locators | Doc; Content or structural identity: OLE2…
  • 3. The Parser Jail: Prove Confinement Before Resume [Part 2/4]: Subtopics: Private staging; Six-step launch contract — A sandbox named in a configuration file is not a sandbox observed in a child token. AINA's document worker is built around pre-resume verification. Each parse receives an ephemeral scratch tree with three distinct grants: The child recomputes the input digest before parsing. — Key Rules/Claims: bin/: a private read-execute copy of squad-doc-extract.exe, so the jail need not read the installation directory. | input/: read-only source.bin and a bounded doc_source.worker/1 request containing the expected input digest and extraction limits.
  • 4. Extraction Is Not Complete If the Gaps Disappear [Part 2/4]: Subtopics: The 35-reason omission ledger; Parser warnings become evidence; Gaps are addressable nodes; Exhibit 1 — A synthetic ViewManifest with an addressable omission — A parser that returns a string has answered only one question: what text did this adapter emit? It has not established whether the document contained more text, whether a library recovered by discarding structure, or whether the caller later received everything extracted. — Spec/Table: Integrity; Values: verified, failed; Question answered: Did the captured bytes match the registered identity and integrity checks? | Inventory; Values: complete_for_profile, partial, unknown; Question answered: Could the adapter enumerate the document's parts and establish a denominator? | Extraction; Values: complete_for_profile, partial, unsupported, failed; Question answered:…
  • 5. Immutable Views: Identity Includes the Reader [Part 2/4]: Subtopics: Commit children first, then seal the parent — Two systems can read identical bytes and produce different evidence because their parsers, profiles, feature sets, or limits differ. AINA therefore binds a view to all of them: RFC 8785 JSON Canonicalization Scheme (JCS) gives the manifest one deterministic serialization.
  • 6. Server-Cut Evidence: srv_ Before ev_ [Part 3/4]: Subtopics: Event 1: the Room serves content and issues srv_; Event 2: the Room computes the citation and issues ev_; Exhibit 2 — A valid synthetic citation issuance; Refusal is part of the citation grammar; Exhibit 3 — quote_ambiguous; Exhibit 4 — drifted — A server-cut citation is not a model emitting a prettier reference format. It is a two-event protocol. The squad.evidence.receipt/1 record binds: The receipt identifier is srv_ plus the first 32 hexadecimal digits of the receipt's canonical SHA-256. — Key Rules/Claims: the Room, source revision, and view digest; | the recipient and recipient session; | the serving tool; | each node and optional character interval served; | delivery counts and complete, truncated, or chunked status; | admission, optional via-handle or wake identity, time, and nonce.
  • 7. Network Capture: Authorize Every Hop, Not Just the First URL [Part 3/4]: Subtopics: Captured, supplied, and governing are different roles — Documents also arrive through retrieval. A fetch tool that authorizes only the initial URL can be redirected into a loopback service, private network, or cloud metadata endpoint. — Spec/Table: room_captured + doc_…; What the Room observed: Registered bytes captured from admitted project territory; May anchor a governing requirement?: Yes, after first-hand serving and the other requirement checks | room_captured + cap_…; What the Room observed: A response body fetched through the Room's egress-gated path; May anchor a governing requirement?: No; research capture is not a registered…
  • 8. Findings: Claims the Server Can Re-Evaluate [Part 3/4]: Subtopics: Absence is a coverage claim before it is a search claim; Exhibit 5 — absence_unsupported — A citation establishes a passage. A finding states something checkable about it. The evaluator supports five typed assertions: The interpreter remains separate from the machine-checked assertion. — Key Rules/Claims: equals — exact text or typed-value equality. | in_range — a typed numeric value against minimum/maximum bounds, with an optional unit check when the view knows the unit. | matches_regex_bounded — a pattern capped at 512 characters, compiled with 1 MiB size limits and nesting depth 32.
  • 9. Revision, Reassessment, and Correspondence Without Retargeting [Part 3/4]: Evidence is allowed to change. History is not allowed to move underneath its citations. When a new view is committed for a source, dependent findings, evaluations, and requirements receive deterministic rsa_ reassessment flags. — Key Rules/Claims: correspond proposes candidate locations in a later view. | The candidate search has four stages, with a separate same-kind rule for textless structural nodes: same_node_same_text, same_node_same_kind, exact_text, normalized_text, then the weak same_node_changed_text fallback.
  • 10. Echo Collapse: Count First-Hand Origins, Not Repetitions [Part 3/4]: Subtopics: Exhibit 6 — Three citation records, one first-hand origin — Multi-agent systems are particularly good at making repetition look like corroboration. Agent A reads a passage. Agent B cites A's finding. Agent C cites B's summary. The graph now contains three records and still contains only one first-hand observer.
  • 11. The 2026 Evidence Landscape: What External Work Establishes [Part 4/4]: The 2026 literature and vulnerability record reinforce the need for explicit evidence boundaries. They do not validate AINA, and their measurements should not be stretched beyond their actual scope. The composition matters. Sandboxing addresses parser blast radius. — Spec/Table: Zhao, Pandey, and Misra, What the Author Wrote, What the Agent Reads, October 2026 preprint; What it actually reports: Across 76 real policy documents, formatting-blind ingestion reduced requirement-recovery micro-F1 from 0.690 to 0.150 for Docling and 0.075 for plain text; six answering models replicated the loss.
  • 12. The Twenty-Case Evidence Conformance Gate [Part 4/4]: The epistemic subsystem has its own 20-case evidence battery inside the 45-case candidate-binary publication gate. These are author-controlled engineering checks, not external certification. The gate demonstrates named path behavior on the candidate binary. — Spec/Table: evidence_jcs_rfc8785_vectors: Canonical serializations match the RFC's vectors | evidence_corpus_through_mcp: Corpus documents read through the confined worker with expected nodes and outcomes | evidence_governing_pdf_six_pages: The governing PDF yields its six text pages through squad_doc_source read | evidence_repeated_text_resolves_each_node: Repeated text is ambiguous until occurrence…
  • Bring One Document Boundary [Part 4/4]: The Zero-Trust Evidence & Citation Audit Challenge is the direct application of this layer. Bring one document-heavy AI workflow and one bounded evidence question.
  • The Article 02 Claim Card [Part 4/4]: Key Rules/Claims: Mechanism: | AINA Epistemic Trust Layer | (doc_source, confined extraction, evidence views, receipts, handles, | findings, requirements, reassessment, correspondence, and lineage) | Failure it governs: | Hostile document parsing inside the controller; stat/read/hash substitution; | silent extraction loss; fabricated or unserved citations; stale evidence; | unbounded absence claims; and multi-agent repetition masquerading as corroboration.
  • Before EISA Judges the Act [Part 4/4]: Return to the renewal contract. The conventional pipeline returned a clean answer: “Annual increases are capped at 3.0%, and there is no change-of-control clause.” Its citation looked plausible. Its confidence was high. Its evidence world was never established.
  • Frequently Asked Questions [Part 4/4]: Subtopics: What is a server-cut citation?; Does LPAC confinement make document parsing safe?; Does complete_for_profile mean the whole document was read?; Can a server-cut citation prove that a source is true?; How does bounded absence work?; Do three agents citing one passage provide three independent origins? — A server-cut citation is an ev_ handle issued by the Room from a caller-bound srv_ served-content receipt. The caller supplies an exact quote or served node; the server verifies coverage, computes Unicode-scalar coordinates, selects the narrowest structural node, and binds…
  • Continue the Series [Part 4/4]: Key Rules/Claims: Series index: From Proposal to Consequence: The AINA Transition-Governance Series | Previous: Article 01 — An AI Output Is Not an Act | Next: [Article 03 — Semantic Switch and EISA: Protected Computation and Entity…
  • Primary 2026 Sources [Part 4/4]: Key Rules/Claims: Zhao, Pandey, and Misra — What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion, arXiv:2610.30026 | [Wang et al.
  • Publication Metadata [Part 4/4]: Title tag: Zero-Trust Document Ingestion & Server-Cut Evidence (srv_ → ev_) Meta description: Why RAG chunks, file paths, and agent-written citations are not evidence: inside AINA's LPAC-confined parser, omission ledger, canonical views, and server-cut srv_ and…