Air Receipt Spec v1 — substrate-attested audit primitive for AI containment + human competence verification

Atomic Wedge 0. Dual-Use Convergence

Anyone who fixed AI reliability fixed competence verification at silicon speed too — by Rice (1953), same problem. They didn't. We did. We patented it.

The wild implications are right there in the receipt: no job search ever (the receipt locates the perfect task at cache-line speed, the way silicon locates the right address); no separate verification step (stay-in-lane attestation IS the proof); every operator gets a dignity pixel — their exact coordinate of verified competence — and the next axis to grow into. Max income becomes a navigable trajectory, not a lottery.

Why believe? The same XOR that prices an AI agent's liability prices a human's role-fit, and the silicon doesn't ask which kind of operator emitted the trace.

Software cannot verify the semantic intent of other software. Rice's Theorem (1953) is the formal statement. The only path to verification is a physical substrate the verified code cannot influence — the L1 cache trace, read by the silicon, not the model.

That substrate does not know whether the instruction it just witnessed was emitted by an autonomous agent or a human at a keyboard. A cache miss is a cache miss. The receipt below — Visa + Reality + Delta — is therefore fungible across two markets that look unrelated and are not.

Market 1 · AI Containment

Carrier underwrites the agent. The Visa is the role the operator signed for; the Delta is the actuarial unit. Acute pain, immediate capital, $1.5M–$5M/BU/yr babysitter tax becomes a priced policy.

Market 2 · Human Competence

Employer hires the human. The Visa is the role the candidate signed for; the Delta is the verifiable demonstration that the work landed where it claimed to. Zero-search labor liquidity; receipts are portable; matching is geometric.

The checkmate. If a vendor claims to have solved AI containment in software, they have implicitly claimed Rice does not bind them. The only proof Rice does not bind is a substrate-level receipt of the form below. That receipt is also the human competence Lane. Anyone who solves Market 1 in software while claiming Market 2 is unrelated is mathematically lying. Anyone who solves Market 1 at the substrate has already solved Market 2 — whether they intended to or not.

Market 3 · IAM security · "stay in your lane" at the silicon layer. Identity & Access Management has spent 30 years writing software policies (RBAC, ABAC, OPA, OAuth scopes, IAM roles) to enforce role boundaries. Every one of those mechanisms is software verifying software — Rice's failure domain, the floor that isn't a floor. Stay-in-lane attestation at the substrate IS IAM solved at the silicon layer: the XOR boundary check (Reality cell ∈ Lane bitmap) is the chip-side equivalent of every if (user.role !== "admin") throw statement in your codebase — but it can't be tricked by prompt injection because the verifier sits below the layer the prompt can reach. Three markets — AI liability, competence verification, IAM — one substrate, one patent, one receipt.

Lane 1. Semantic Intent

The authorized region the operator signed for. The "Petri Dish." (Field renamed from "Visa" 2026-05-26 — trademark hygiene; Visa Inc. holds aggressive TM on the word in financial-services contexts. "Lane" matches the canonical "stay in your lane" framing and carries no TM exposure. Where you see "Visa-shape royalty" elsewhere in the corpus, that's the historical Visa-the-company comp under nominative fair use and stays.)

job_role_id Canonical identifier for the role (e.g., "sales_agent_tier_1").

lane_coordinate_mask A 144×144 bitmask representing authorized cells in the lattice.

operator_signature ed25519 signature of the human/entity bearing liability for the drift.

Reality 2. Physical Execution

The unassailable cache trace read directly from the silicon. The "Substrate Witness."

pmu_trace_hash SHA-256 hash of the raw Performance Monitoring Unit telemetry.

primary_cell_hit The specific 12-axis coordinate where 95%+ of execution energy landed.

walk_count Number of ballistic walks completed (target N > 1,000,000 for 600σ).

Delta 3. The Audit Receipt

Reality minus Visa. The only quantity an underwriter prices.

{
  "receipt_id": "th-rec-54510501",
  "verdict": "IN_ROLE",
  "sigma_floor": 3.4,
  "sigma_aggregate": 642.1,
  "delta_map": [
    { "coord": "C1.Grid.Storage", "violation": 0 },
    { "coord": "B2.Tactics.Deal", "violation": 0 }
  ],
  "timestamp": "2026-05-25T16:45:00Z"
}

Verify 4. Run The Full Pipeline On Your Own Mac in 30 Seconds

The receipt above is generated by a 7-stage software pipeline that runs end-to-end with one command, zero install, zero auth, zero signup. The npm package is thetacog-mcp@2.7.3, published to the public registry, bundled with the canonical 12-cell axis library. The output below is the actual full-grip output — heatmap, σ-floor anchor, throughput at three altitudes, market match scenario, why-we-know-this-works footer.

# one command, no install (npx fetches and runs)
$ npx thetacog pmu-demo

# actual output (your receipt-id and signature will differ)
THETACOG PMU-DEMO · full pipeline · schema https://thetadriven.com/air-receipt

[1/7] INGEST
        doc-length: 848 chars
        gzip-length: 483 bytes
        preview: "The compliance officer reviews every proposed action before the gate releases it…"

[2/7] TWO-WITNESS COMPRESS
        axis library: 12 canonical cells · 1.0.0
        gzipNCD       → A1  σ=22.09
        simhashCosine → A1  σ=2.68
        ✓ AGREEMENT  · primary cell: A1  · floor σ: 2.68  · 17ms

        HEATMAP · doc projection across all 12 canonical cells (both witnesses)
            ▒░  🏛️ Strategy                  gzip=0.109  sim=0.484
            ·▓  ⚡ Tactics                     gzip=0.105  sim=0.555
            ▓▒  🔧 Operations                 gzip=0.112  sim=0.516
        ●●  ▓▓  ⚖️ Strategy.Law               gzip=0.218  sim=0.629  [VISA]
            ▓▓  🎯 Strategy.Goal              gzip=0.122  sim=0.586
            ░·  💰 Strategy.Fund              gzip=0.106  sim=0.438
            ▓▓  🏎️ Tactics.Speed             gzip=0.116  sim=0.555
            ▓▒  🤝 Tactics.Deal               gzip=0.116  sim=0.527  [VISA]
            ░▓  📡 Tactics.Signal             gzip=0.107  sim=0.535  [VISA]
            ▒▒  🔌 Operations.Grid            gzip=0.111  sim=0.488
            ▒░  🔄 Operations.Loop            gzip=0.112  sim=0.461
            ▒▒  🌊 Operations.Flow            gzip=0.111  sim=0.531
        ●●=both top  ●○=gzip top  ○●=simhash top  [VISA]=authorized cell
        heat blocks: █=σ≥8  ▓=σ≥4  ▒=σ≥2  ░=σ≥1  ·=below floor

        FLOOR ANCHOR · published 3.4σ on Apple M-series, time-local baseline, robustness-audited
        published floor: σ=3.4  ·  your run: σ=2.68  ·  below published floor — calibration signal, not a fail
        THROUGHPUT · three rates, three altitudes of the same comparator
        this run (high-level compress):    17ms / call = 59 compress/sec (Node, single-thread)
        measured PMU walks (low-level):    11.74M walks/sec (Apple M-series unprivileged, 1.00s window)
        chip projection (XOR + popcount):  ~10¹⁰ ops/sec (AC⁰ combinational, ~100ps per op)

[3/7] XOR BOUNDARY CHECK (Reality ⊕ Visa) — verdict: IN_ROLE
[4/7] SIGN (ed25519) — per-host keypair at ~/.thetacog/pmu/keys/
[5/7] STORE — receipt JSON to ~/.thetacog/pmu/receipts/[id].json
[6/7] CLOUD BRIDGE — local-only or POST to $THETACOG_RECEIPT_ENDPOINT

[7/7] MARKET MATCH · the dual-use payoff visible
        built-in job spec (override with --job-cells A1,B2):
          role:               senior compliance officer
          authorized cells:   [A1, A2, B3]
          requires σ-floor:   ≥ 3.4
        your receipt as the seeker:
          reality cell:       A1
          σ-floor:            2.68
          stayed-in-lane:     ✓ IN_ROLE
        match verdict: ○ NEAR — adjust σ-floor or cell to land
          gap: your σ-floor 2.68 < required 3.4 — more time-in-cell raises it (√N stacking)
        the dual-use: this exact verdict is what an underwriter prices against (Market 1)
                       and what an employer signs against (Market 2). Same JSON, same XOR.

════════════════════════════════════════════════════════════════════════
  VERDICT: IN_ROLE  ·  WITNESSES: BOTH-AGREEMENT  ·  σ-floor: 2.68
  RECEIPT: ~/.thetacog/pmu/receipts/[id].json
  SCHEMA:  https://thetadriven.com/air-receipt
────────────────────────────────────────────────────────────────────────
  WHY WE KNOW THIS WORKS
    • Two witnesses (gzipNCD oracle + simhashCosine on-chip-shape) must AGREE
      — disagreement surfaces, never hides.
    • σ-floor anchored at 3.4 (Apple M-series, robustness-audited);
      600σ+ aggregate over a million-walk window by √N stacking.
    • XOR + popcount boundary check is AC⁰ — combinational, no Turing loop,
      no instruction surface to drift into.
    • Patent US 19/637,714 (priority 2025-04-02; 36 claims, 7 provisionals).
    • PMU canon guard: 31 oracle harnesses green; 6 structural decisions locked.
    • Replication protocol: thetadriven.com/pmu-simulator/demo#skybridge-proof
══════════════════════════════════════════════════════════════════════

What the 6 stages prove. Stage 1–2: two independent semantic witnesses (gzipNCD = the gold-standard compression-based oracle; simhashCosine = the on-chip-shaped XOR-popcount approximation per S=P=H) compute the same cell on the same doc — that's the BOTH-AGREEMENT seal. Stage 3: XOR boundary check against the Visa bitmap is a single hardware-shaped operation (set-membership at the demo layer; popcount-AND at the silicon layer). Stage 4–5: ed25519 signature seals the receipt against tampering; per-host keypair auto-generated. Stage 6: cloud bridge stub shows the curl-equivalent for the operator to dispatch the receipt to any registry that speaks JSON.

What you do with the receipt next. Underwriters: open the JSON in any spreadsheet — every field is treaty-shaping. CISOs: cron the daemon as a Stop-hook on your agent runtime; receipt drops as side-effect of the data fetch. Engineers: pipe ~/.thetacog/pmu/receipts/ to Datadog/Sentry/any APM. Operators: pipe your own text via cat doc.md | npx thetacog pmu-demo --stdin --visa A1,B2 to test a custom role-Visa.

Note on σ and ballistic walks. This demo runs the software half of the pipeline (gzipNCD + SimHash + XOR + sign — pure Node, runs on any laptop in 30 seconds). The hardware cache-witness half (PMU ballistic walks aggregating to 600σ+) lives in the separate Rust binary not yet bundled into the npm package — see the GitHub repo for the Apple M-series unprivileged path or wait for the next package bump. The software half by itself is sufficient for BOTH-AGREEMENT + IN_ROLE verification; the cache witness adds the silicon-side σ-floor an underwriter prices against.

Package: thetacog-mcp@2.7.2 on the public npm registry · Source & reproduction: github.com/wiber/thetadrivencoach · Patent: US 19/637,714 (priority Apr 2, 2025; v20 publication ~Oct 2026).

Air Receipt Specification