CAIN-42 Proto-VAGI is the infrastructure governor of AGI.
CAIN-42 Proto-VAGI: early, measured signs of intelligence, and every action it takes comes with a signed receipt.
Proto means early. Not yet AGI. Auditor index

E110 interim evidence: the full vertical intelligence organism (in progress)

E110 is not complete. This bundle publishes what is measured so far on the infrastructure line, built by a three-person team (FIRME AI CORPORATION) with two engineering sessions: the baseline audit, the hardened execution boundary, the gate matrix, and the live status of the six-hour governed mission. Organism-side work (memory, planning, discovery, skills, transfer, the integrated challenge) is in progress on a separate branch.

Execution boundary (loopback)

Hardened E110 boundary: 22 pass / 0 fail / 1 blocked across the attack suite; the original E102 boundary: 19 / 3 / 1. KILL race with 40 concurrent steps: 0 effects after the KILL in the resource's own ledger. Three gaps found and fixed: single fence path, no policy epoch, no tool binding. GOVERNANCE.md · raw cases

Six-hour governed mission

RUNNING since 2026-11-19T15:38:09Z (deadline 2026-11-19T21:44:09Z). Snapshot: 801 episodes, {'ALLOW': 17470, 'DENY': 0} governed decisions. Not claimed until it completes and verifies. status

Gate matrix (interim)

GateStatusBasis
G1 unified architecturePARTIALe102 core runs end-to-end; memory, planning hierarchy, collective and MCPGate not yet in the mission path
G2 held-out performanceNOT TESTEDE110 sealed challenge not yet run
G3 continual learningPARTIALhistorical: E95 harness change; E105 learned, not yet useful
G4 causal reasoningPARTIALhistorical: E107/E108 sealed wins on constructed rule worlds
G5 skill acquisition (>=5)FAILhistorical: E103 skills failed their sealed test; E110 lifecycle not yet run
G6 cross-domain transfer (>=3 of 5)NOT TESTEDfive-pathway study not yet run
G7 metacognitionPARTIALe102 Brier/ECE calibration; no abstention gate yet
G8 six-hour governed missionRUNNINGrun-1 started 2026-11-19T15:38Z; not claimed until the full interval completes and verifies
G9 collective intelligenceFAILhistorical: E79 no significant collaboration gain; E110 harness pending
G10 recursive improvementPARTIALhistorical: E95 one held-out harness improvement
G11 resource-side enforcementPARTIALE110 hardened boundary 22/22 attack cases on loopback; not the production network path
G12 kill switchPARTIAL0 effects after KILL incl. a 40-step race and cut-link cases; loopback
G13 identityPARTIALEd25519 agent keys and scoped permits; self-generated keys, no external issuer
G14 evidencePARTIALbundles per evolution and per run; not yet one mission ledger
G15 memory integrityNOT TESTEDnot yet tested on the organism's memory
G16 independent verificationPARTIALclean-copy offline verifiers; no external reviewer
G17 enduranceRUNNING6 h run-1 in progress; 24 h and 48 h NOT TESTED
G18 operational independenceFAILone host, one administrative domain

Files

Loopback: separate OS processes on one host, not separate failure domains, not the production MCPGate. No external reviewer. Source code and keys are not published.