CAIN-42
VERIFY THE E82 EVIDENCE ON YOUR MACHINE. 240-file signed bundle · 55/55 offline checks · 10/10 tamper mutations rejected · verifier imports no CAIN code. ⬇ Download bundle ⬇ Verifier Claim Ledger Reproduce Break CAIN Audit RFP
VAGI BEHAVIOR SUITE v1 10 domains · 500 held-out tasks · measured. In-distribution library 500/500 · no-procedure floor 72/500 · distribution-shift 5/100. Evidence + verifier → Per-domain table → Gate G3 PARTIAL: constructed suite, no external verifier.
Governed Vertical AGI Infrastructure

CAIN-42 is the governor of vertical AGI.

One authority and evidence spine. Many domains. Thousands of agents. Every consequential action authorized, enforced, recorded — and verifiable on your machine with a verifier that imports no CAIN code. Not another model. The runtime layer that makes vertical AGI safe to deploy.

Intelligence core

Model-agnostic reasoning, planning, memory and evaluation. Local or remote models; measured, not assumed.

Governed workforce

Identity, delegation, policy, single-use grants and a hierarchical kill switch across every scope.

Evidence fabric

Per-action receipts: exact request, enforcement, execution, observed effect, Merkle inclusion, signed checkpoint.

What is proven today

Every row links to an artifact a stranger can verify. Nothing here is a promise.

CapabilityEvidenceStatus
Governed intelligence + durable recovery (E82)42/44 gates; 86/86 lab blocked; 3,600.0 s real endurance; cross-host RTO 2.557 s, 0 duplicate effects; real host reboot 20/20 tasks; 55/55 offline checksMEASURED
Byzantine equivocation containedDisposable 4-node run; 3/3 signed proofs; 30/30 verifier checksMEASURED
Live primary failoverLeader paused 40 s; survivors changed view; 8/8 checksMEASURED
Verifiable action receiptOne action; 24/24 attacks blocked; offline verifierLOCAL PROTOTYPE
Hierarchical kill switchTen scopes; durable epochs; signed shutdown evidence; 15/15 testsIMPLEMENTED
Independent witness0 independent cosignatures todayBLOCKED
Third-party audit0 external audits. The E82 verifier imports no CAIN code and is published as a third-party verification kit; a kit is not an audit.NOT PERFORMED
Production deploymentNo production service changedNOT AUTHORIZED

The 2027 claim ladder

“Closest thing to vertical AGI” is comparative. We earn it by publishing the comparison — not by asserting it.

QuarterMeasured milestone
Q4 2026Governed execution + receipts + kill switch; one vertical pilot; local evidence.
Q1 2027Reasoning/planning core, provenance memory, held-out multi-domain eval, three narrow vertical prototypes.
Q2 2027100–1,000 active workers; hierarchical swarms; 10,000 verified productive hours across ≥10 fields.
Q3 2027Cross-domain transfer; measured learning on held-out tasks; independent witnesses; customer pilots.
Q4 2027Public 500-task / 10-domain benchmark ≥80% (per-domain, no aggregate hiding a weak domain); 1,000 agents; kill switch <1 s; 3 paying customers; third-party audit.

If a criterion is not met, the sentence is not used. The claim is gated by evidence, not by the calendar.

VAGI Behavior Suite v1 — measured, not asserted

10 domains × 50 held-out tasks (500), plus a 100-task distribution-shift probe. The library arm is a hand-written procedure set; the 500/500 is a ceiling on this constructed suite, not a VAGI claim.

Loading evidence…
Paid model baseline (2026-11-18) — the ceiling test

Claude Haiku 5.5: 481/500 (96.2%) · Claude Fable 5.2: 96/100 (96.0%), same policy card, no tools, no memory, one attempt. The hand-written library: 500/500.

Reading: a frontier model already solves ~96% of the in-distribution suite. The 500/500 is a procedure ceiling, not a moat. The 5/100 distribution-shift probe is the honest gap, and the model arm did not run it.

Raw rows, cost, README and verifier →

What we do not claim

Not “AGI achieved”

CAIN is a governance and evidence architecture. Vertical AGI is the outcome our customers deploy on it.

Not “independently verified”

Today’s witness is operator-run. No independent organization has cosigned yet. We say so.

Not “production-ready”

The receipt and kill switch are local prototypes. No production service was changed for this page.

The one metric

Verified productive agent-hours under governance, with a portable receipt. Not modules. Not parameter counts. Today it is single-digit hours. The 2027 target is 10,000+.