Quick Routes: Plain-Text Brief (/ai-brief.txt) · All 9 Articles (/articles/) · Digest 1 · Digest 2 · Digest 3 · Assurance · Refusal Proof · Semantic Switch · GitHub · Author: Blayne Aina (me@blayneaina.com, Denver, CO)
- Core Architecture & Three-Layer TCB Bounds (A2E + EISA): AI models operate as untrusted proposal generators without direct execution standing. Layer 1 (Pre-Parse §03.1):
MAX_SUBMISSION_BYTES=1,048,576,MAX_REQUESTED_ACT_BYTES=65,536,MAX_EVIDENCE_REF_BYTES=4,096,MAX_ACTS=256,Fracture/Forgestage homogeneity,visit_mapduplicate-key rejection,visit_f64non-finite rejection. Layer 2 (12 Closed Draft-07 Schemas): recursiveadditionalProperties: false, 7 closed opcodes (BindEvidenceTag,RequestHumanReview,RejectUnlawfulAct,AssertLayoutRelation,AssertDuplicatePair,AssertTransitiveClosure,ProposePartition), integerconfidence(0..10000, zero floats), 15 non-inferential tags, 47 fault codes, RFC 6901 depth<= 8(<= 256B). Layer 3 (Four-Gate Machine & Windows NTsquad.exe):MAX_CASE_BYTES=4,194,304,MAX_CASE_RECORDS=1,000,MAX_VALUE_CHARS=1,024,MIN_PHONE_DIGITS=7,CreateProcessW(CREATE_SUSPENDED)+PssCaptureSnapshot(act-pin),FILE_SHARE_READexclusion, SQLiterow_version. - Verified Public Evidence & IP/Research Scope: Offline Refusal & Conformance Bundle (
python proof/verify_refusal_proof.py):39/39SHA256SUMS.txtmatch,4/4deterministic refusal records (Cases A/B/C/D; Case Ccontacts.db16,384bytes SHA-256362088a1...b47d719unchanged,0effects,0model calls,requested_act_mismatch),45/45conformance cases,20/20evidence cases,18/18positive +26/26negative schema/pre-parse tests across all 12 schemas. Semantic Switch Alpha:20/20invariance blocks,23/23exact native relations,0provider calls. 3 USPTO Provisionals (Apr–May 2026):63/996,553(AINA Protocol, Apr 12, 2026),64/004,193(Omnibus Parts 2A & 2B: Semantic Switch/MX2MX & Topological Physics/EISA, May 2, 2026),64/027,093(A2E Governance, May 15, 2026). 5 Research Work Packages:PRJ-TW-01(Assurance Initiative & cross-framework A2E),PRJ-LC-01(eisa-core/pcna-verifier,Authenticated PCNA v2+ audit),PRJ-CG-02(250-case AI-control study),PRJ-TAIF-01(Lean 4eisa-lean4+ 10k crash-recovery benchmark),PRJ-NAVY-01(AINA-LinuxHardener+ pre-execution barrier). - §01 — An AI Output Is Not an Act (Full · .md · brief.txt · 6 parts): Problem: Valid JSON, credentials, confidence, and a green tool result can make an AI proposal look authorized when it is not. Proves: Machine standing binds the exact act, actor, evidence, target, execution, time, and terminal disposition before consequence. Sections: The Green Check Mark That Proved the Wrong Thing · A 60-Second Map of the Argument · The Complete Path in One View · Two 2026 Failure Classes: Forged Origin and Approval Laundering · The Unified Trust Machine · The 2026 Action-Boundary…. Claim Card: Mechanism: | AINA Transition-Governance Architecture and Constitutional Kernel | (doc_source, evidence, eisa, room, leases, vault, confine, destinations, enforce, task, invariants) | Failure it governs: | The Lawful-Act Defect — syntactically valid, authenticated, or confident AI proposals | becoming self-authorizing…
- §02 — Zero-Trust Document Ingestion (Full · .md · brief.txt · 4 parts): 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. Sections: The Evidence World Is a Governed Artifact · 1. Registration: Bind Identity to the Bytes Actually Read · 2. Content Decides the Format—Inside the Jail · 3. The Parser Jail: Prove Confinement Before Resume · 4.. Claim Card: 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…
- §03 — Semantic Switch & EISA (Full · .md · brief.txt · 6 parts): Problem: Entity resolution needs rich comparison structure, but raw contacts, companies, and products should not be sent to external AI providers. Proves: The production EISA Four-Gate machine validates bounded acts over sealed evidence; Semantic Switch is an experimental alpha structure-preservation research module. Sections: One Constitution, Two Proof Lanes · 1. Strict Representation Before Judgment · 2. The Closed Grammar: Seven Opcodes, Four Layers · 3. The Evidence Alphabet: 17 Tags and 12 Comparators · 4.. Claim Card: Mechanism: | Semantic Switch Protected Computation + EISA Constitutional Act Recognition | (native-data minimization, relation-isomorphic fictional records, | qBlock candidates, Fracture, Forge, local exact restoration, Shield, | strict parsing, sealed cases, pointer law, deterministic comparators, | Four-Gate…
- §04 — The Room Coordination Plane (Full · .md · brief.txt · 5 parts): Problem: Git history and chat transcripts cannot establish who may act, receive, stop, or complete work now. Proves: The Room revalidates territory, causal observation, sealed wakes, native replies, cancellation, and privileged acceptance. Sections: The Whole Machine, Not a Wake Queue · One Assignment Through the Machine · 1. Git and Chat Are Necessary—and Insufficient · 2. Subsystem 1: Territory and Principal Identity · 3.. Claim Card: Mechanism: | The AINA Room Coordination Plane | (state identity, project grants, connection/session authority, tool policy, | graph shards, visibility, sync, contribution verbs, wake permits, | Session Factory, worker confinement, cancellation, acceptance, invariants) | Failure it governs: | Source history and chat…
- §05 — Expiry Is Not Fencing (Full · .md · brief.txt · 4 parts): Problem: An expired lease does not physically stop a stale worker or an uncooperative local writer from changing a resource. Proves: Monotonic generations, current tenancy, typed scopes, and overlapping Win32 handle exclusion fence the write itself. Sections: The Two Hazards the Architecture Must Defeat · 1. Six Typed Namespaces and Real Target Identity · 2. Workspace-Wide Monotonic Generations · 3. A 30-Second Cross-Process Transition Mutex · 4.. Claim Card: Mechanism: | AINA Typed Leases, Monotonic Generation Fencing, and Guarded Replacement | (typed scopes, room.generation.v2, transition mutex, lifecycle guards, | G/W/T/H/G' handles, parent pin, liveness, SQLite fences, and release quarantine) | Failure it governs: | Stale workers resurrecting after lease expiry; |…
- §06 — Rollback Is Not Erasure (Full · .md · brief.txt · 3 parts): Problem: A single undo snapshot can destroy intermediate states and launder unauthorized writes into a clean baseline. Proves: Primary certifies operational recovery while Amber preserves every intermediate and ungated state for forensics. Sections: 1. Why Single-Tier Undo Fails · 2. The Dual-Tier Separation · One Checkout, Six Mutations, Two Truths · 3. Primary: A Certified Operational Baseline · 4. Amber: Content-Addressed Forensic Objects · 5. Fail-Closed Checkout Capture · 6.. Claim Card: Mechanism: | AINA Dual-Tier Recovery and Forensic Quarantine | (Primary certified baselines, Amber content-addressed objects, | fail-closed checkout, version replay, per-edit snapshots, | bypass sentinel, release verification, and governed restore) | Failure it governs: | Single-slot undo destroying the original…
- §07 — Defeating Approval Laundering (Full · .md · brief.txt · 3 parts): Problem: Approval for action A can become a different action A′ at execution, and a crash can separate commit from receipt. Proves: Pre-resume confinement, controller-fixed RequestedAct, fenced SQLite, and non-fabricating successor grants preserve act and return continuity. Sections: One Approved Mutation, Three Places to Fail · 1. Pre-Resume Kernel Verification · 2. Controller-Brokered Model Egress · 3. Controller-Fixed
RequestedAct· 4. The Fenced SQLite Execution Window · 5. Multi-Destination Capability Plans · 6.. Claim Card: Mechanism: | AINA Execution Confinement, Act Fixation, Fenced Destination, | and Non-Fabricating Successor Continuity | (pre-resume token/handle inspection, spawn gate, model egress, | RequestedAct, SQLite IMMEDIATE, capability plans, ResultGrant) | Failure it governs: | Unclassified handles and ambient capabilities… - §08 — The Sovereign Kill Switch (Full · .md · brief.txt · 4 parts): Problem: A kill switch the agent can clear is not a boundary, and a green build does not prove the release enforces refusals. Proves: Latch-first containment, scoped epochs, observed Job termination, in-doubt recovery, and the 45-case candidate gate close the loop. Sections: One Stop, Three Races, No Invented All-Clear · 1. Asymmetric Containment · 2. Latch-First Containment · 3. Control Epochs Revoke Authority, Not Just Processes · 4. Managed Drain and In-Doubt Recovery · 5.. Claim Card: Mechanism: | AINA Sovereign Containment and Candidate-Binary Assurance | (latch-first break-glass, scoped control epochs, owned-Job observation, | in-doubt recovery, interactive owner recovery, 45-case publication gate) | Failure it governs: | Agents or scripts clearing their own emergency stop; | stale authority…
- §09 — Bring One Real Boundary: Peer Review & R&D Evaluation (Full · .md · brief.txt · 8 parts): Problem: Writing code by hand is no longer the bottleneck. The bottleneck is formalizing the trust boundary, decomposing work across parallel AI agents without context collapse, catching plausible-but-wrong outputs, and proving that invalid transitions fail closed. Proves: AINA is both a deterministic execution-governance runtime (squad.exe) and Exhibit A of a repeatable 9-IDE multi-agent engineering methodology grounded in 11+ years of production systems delivery. Sections: How to Peer-Review & Evaluate AINA: The 5-Step Architecture & Runtime Verification Protocol · Why AINA Is Grounded in 11+ Years of Production Systems Engineering · 1..