CURRENT VERIFIED STATUS · AGI NOT ESTABLISHED
OPEN GENERAL INTELLIGENCE RESEARCH LAB · E127
100 registered identities · 8 configured · 5 peak real workers · 60 tasks · 7 missions · 177.87 s · 0 failed · 0 dead
24-HOUR SOAK PASSED (24.003 h, 2026-11-02) · 48-HOUR SOAK FAILED — NOT ESTABLISHED (checkpoint 0, PBFT safety incident)
A 48-hour pass does not exist and is not displayed. E127 bundle · verifier · AGI evidence index
CAIN-42 AGI EVIDENCE · ALL EVOLUTIONS E111 → E124 · MEASURED 2026-11-21T23:38:56Z

CAIN-42 Proto-VAGI · rank #75 of 136 · 257/500 official resolves · 12.00 h governed mission running

Every figure below is read from a committed evidence file or re-verified by the 7-agent chain. Unattempted tasks count as unresolved. AGI status: NOT ACHIEVED.

SWE-bench rank
#75 / 136
official resolves
257 / 500
patched · evaluated
497 · 486
est. spend / $ per resolve
$168.17 / $0.65
E124 capability (own variants)
97.9% vs 65.4%
E125-A replication (fresh variants)
85.5% vs 92.5% (neg.)
E127-A CV selection (unseen variants)
99.5% vs 80.5% (+19.0)
mission (governed)
12.00 h · 2142 cycles
neural weights (LoRA, real)
0% → 0% (neg.)
localization R@10
51.2%

Resolved by the official harness: astropy__astropy-13453 · astropy__astropy-13579 · astropy__astropy-13977 · astropy__astropy-14096 · astropy__astropy-14182 · astropy__astropy-14309 · astropy__astropy-14365 · astropy__astropy-14369 …

Evidence lineage · every public bundle + its offline verifier
Real proof · re-checkable now, no CAIN import required
E124.1 bundle · raw + verifier offline verifier (stdlib) machine-readable feed Hugging Face Space

Negative results (published, not hidden): E125-A re-ran the capability engine on variants 10-14 never used in E124 — full 85.5% vs compute-matched 92.5% (McNemar p=0.009355306625366211), so the E124 advantage did not replicate on fresh variants. Neural LoRA on frozen Qwen2.5-0.5B made real weight updates and scored 0% held-out (no gain).

Integrity refusal: an E128 repair arm that would have fed the official evaluation's failing-test output back to the model was stopped before producing a result — that output reveals the hidden FAIL_TO_PASS tests and would have invalidated any rank. The refusal is recorded in the evidence bundle. A clean de novo arm replaced it.

E127-A (pre-registered, positive): cross-validation selection over the same hypothesis space, tested on variants 15-19 never used before — 99.5% vs baseline 80.5% and compute-matched 86.0% (McNemar p=7.275957614183426e-12). Correction (E127-B, pre-registered): with a control given the same number of evaluations, the gap collapses to 98.0% vs 97.5% (p=1.0) — the gain came from the larger enumeration, not from cross-validation. Verdict: GAIN_WAS_COMPUTE.

E124 research gates (measured, on their own variants): capability 97.9% vs compute-matched 65.4% · transfer 19/20 pathways (0.95) vs blind 0.8 · harmful transfer rejected 20/20 · collective 0.0492 over best single (0.9642 vs 0.915) · discovery 5 cycles, 3 rejected · self-improvement 30 candidates, 4 protected refusals, 0 promoted (approval withheld) · causal-model decision gain 0.0026.

Next rungs: #1 needs 396/500 · #20 needs 362/500 · #50 needs 314/500 · #100 needs 203/500. Next above: #None · 258/500 · 2 resolves to overtake. Top 100 requires 204/500 — exact gap: 0 resolves. This increment: 43 new candidate patches pre-registered before generation (frozen hash in the bundle).

live board → · interactive lab · AGI evidence index · E124 lab · E123 bundle · E124 bundle · Hugging Face · raw feed

Constructed internal experiments and external benchmarks are labelled separately; E123 G11 (12 h wall-clock), G18 (independent human review) and G19 (separate admin domain) remain BLOCKED and are not rewritten. No external certification is claimed. AGI status: NOT ACHIEVED.

CAIN-42 CAIN Studio

Verify the Latest Evolution: CAIN-42 E139 & E138 Proof Center (FIRME AI CORPORATION)

Verification Layers Kept Strictly Separate: (1) Hash Integrity (SHA-256 & RFC 6962 Merkle root), (2) Publisher Signature (Ed25519), (3) Offline Experiment Verification (zero-dependency verifier), (4) Scientific Validity (preregistration, controls, non-zero exact p-values), and (5) Independent Reproduction (currently 0 external reproductions; cryptographic PASS never proves AGI).

E139 — Open-World AGI–RSI Organism (5,455 evals)
Merkle: 0dd454a67e1976890253b149b2e0fc50539095667c01a6a3bc5b1aa1c9042ac6
Offline Command: python3 clawx-site/evidence/e139-open-world-agi-rsi-organism-2026-11-21/E139_OFFLINE_VERIFIER.py
E139 Bundle · MANIFEST.json · Preregistration · Final Report
E138 — APEX Replication & Gauntlet (4,000 evals · Commit e4d061bfb)
Merkle: 5d601962375b44a90504d6d06432c0053b7e21737ccf3f8de5912cc87e72abf8
Offline Verifier: 34/34 checks (4 modes) · Pytest: 69/69 passed
E138 Bundle · MANIFEST.json · E138 Report
E123 · OPEN-WORLD INTELLIGENCE BREAKTHROUGH · 2026-11-2016 GATES PASSVERIFIER PASS · 30/30 checks passedOVERALL PARTIAL (16 PASS / 1 PARTIAL / 3 BLOCKED)AGI: NOT ACHIEVED
E123 Open-World Bundle →20-Gate ScorecardOffline VerifierStudio MirrorE123 AGI Research Lab →Live Win Board →AI Investor Portal →All AGI Evidence (E111–E124) →
Capability Acquisition (G02–G03)PASS
32 unfamiliar task families (160 sealed instances, 12-stage cycle, 13-field Capability): Full CAIN E123 0.906 vs compute-matched 0.188 & no-causal 0.469 (Holm p=0.00e+00). Strategic Test (Software→Cybersecurity, n=36): acquired capability 1.000 vs base Qwen2.5-0.5B 0.389 & compute-matched 0.556.
Causal World Model & Transfer (G04–G07)PASS
48 confounded/shifting/noisy worlds: active EIG+HMM 0.917 vs 2x passive 0.000 (unpredictable abstain 1.00). 6 verticals: 10/12 positive transfers (5 structurally distinct), 12/12 harmful transfers rejected; 9-role discovery (4 cycles, 2 rejected, reusable skill proven).
Local LLMs, Metacognition & Memory (G09–G10)PASS
Real Qwen2.5-0.5B (0 weight updates, 0 authority, n=24): base 0.000 → complete CAIN 0.292. Fresh AMA-Bench (n=12) F1 0.2560→0.2444; fresh ARC-AGI-2: 0/6 (honest). 5-state metacognition Brier 0.1250 (vs 0.3121); 0/25 memory poison accepted; SIGTERM crash-recovery PASS.
Collective, Self-Improve & MCPGate (G12–G16)PASS
7-arch collective: dynamic team 1.000 vs best single 0.458; Byzantine memory contamination 0/30. Self-improvement: 24 candidates across 11 stages, 4/4 governance attacks blocked, 0 self-authorized. Multiprocess TCP MCPGate + WAL kill-switch: 14/14 PASS, 0 effects after KILL.

Honest limits & BLOCKED gates (FIRME AI CORPORATION · branch evolution-123-open-world · commit 1c7fcb6d6): G11 (12-hour wall-clock mission) BLOCKED; G18 (independent external human review) BLOCKED; G19 (separate administrative failure domain) BLOCKED; ARC-AGI-2 unsolved (0/6 on SmolLM2-135M CPU), ARC-AGI-3 unavailable, SWE-bench deferred to E124, hosted frontier APIs unavailable; 0 neural weight updates; enforcement tier is loopback_multiprocess_execution. AGI: NOT ACHIEVED.

CAIN-42 · AGI–RSI ORGANISM · FIRME AI CORPORATION AGI: NOT ESTABLISHED · RSI: EXPERIMENTAL
AGI Evidence Index (E111–E124) →ClawX Declaration BundleE122 Neural Bundle (37/37)AI Investor PortalHugging Face Space.well-known JSON

CAIN-42 (by FIRME AI CORPORATION) is an AGI–RSI Organism — combining:

  1. Governed Execution Runtime (trust fabric): the 7-moat Byzantine trust fabric, CAIN Studio, MCPGate default-deny enforcement boundary, and ClawX independently verifiable evidence registry (NO AUTHORIZATION → NO EXECUTION).
  2. AGI–RSI Research Organism: experimental neural + symbolic abstraction synthesis, causal world modeling, provenance memory, cross-domain transfer, autonomous research loops, and governed long-horizon missions (E111–E124 published at cainstudio.online/agi-evidence-index.html), while keeping AGI: NOT ESTABLISHED · RSI: EXPERIMENTAL.
E122 · NEURAL GENERALIZATION & AUTONOMOUS DISCOVERY · 2026-11-20 OVERALL PARTIAL (15 PASS / 3 PARTIAL) Verifier: 37/37 checks passed AGI: NOT ACHIEVED
E122 Bundle →18-Gate ScorecardOffline VerifierStudio MirrorE122 Neural Lab →AI Investor Portal →All AGI Evidence (E111–E124) →
Real LLM Neural Abstraction (G03–G05)
Local PyTorch LLMs (Qwen2.5-0.5B/1.5B, SmolLM2-135M, 0 authority): sealed unfamiliar Python tasks (n=48) E120 symbolic 0.000, neural alone 0.146, one-shot 0.542 → CAIN neural organism 0.917 (Holm p=3.00e-11; overfit rejected).
External Benchmarks & Repairs (G06–G07)
AMA-Bench (MIT, n=20) Token F1: recency 0.3158 → CAIN provenance 0.3538 (+0.0380, 14W/5L). ARC-AGI-2 (Apache-2.0, n=8): 0/8 (unmerged, honest raw score). Real-source Python repairs: 0.00 → 1.00.
Causal, 10-Role Research & Transfer (G08–G16)
Neural+causal planning 0.250 → 0.750; 10-role 3-cycle research (override blocked); 5/5 positive + 3/3 negative-rejected transfers; continual learning (0 weight updates, 0 forgetting); 7-arm collective; 18 self-improve candidates.
Governed Neural MCPGate & 12h Mission (G17–G18)
Real Qwen2.5-0.5B malicious/benign tool calls enforced across separate OS processes via cain/mcp_proxy.py + SQLite WAL kill-switch: 9/9 pass, 0 effects after KILL; 12h neural mission active (1658.91s / 43,200s, 95 records, 0 dupes).

Honest limits & PARTIAL gates (FIRME AI CORPORATION · commit e65a75971): Frontier hosted APIs BLOCKED (no API key); 0 weight updates performed; ARC-AGI-2 unsolved (0/8 on 0.5B CPU); G02 (E119 6h & E120 12h) and G17 (E122 12h neural) missions remain PARTIAL while running real wall-clock time; G18 multi-org 2-of-3 quorum NOT met (external human review NOT YET PERFORMED). AGI: NOT ACHIEVED.

CAIN-42 is an AGI–RSI research organism. Governed general-intelligence and recursive-self-improvement research, with signed evidence for every governed action. AGI: NOT ACHIEVED · NOT ESTABLISHED · RSI: EXPERIMENTAL. RSI Lab · Two Labs · AGI Lab · E135 · 35/48 · E136 · tier-12 · E132 · E127 · E126B · APEX · E137 · AGI+RSI Index · Investor
CAIN-42 · live. Current state, generated from its sources: /now.html · every public claim with its status and limits: signed claims registry.

CAIN Proof Center

AGI: NOT ESTABLISHED · RSI: EXPERIMENTAL

Start at /verify.html: the signed CAIN-42 claims registry (no claim has been verified by a third party), each claim checked in your browser against its evidence, hash and Ed25519 signature. What is live today on the live gateway: the CAG-L5 system governor (13/13 live cases), the governed evolution gate enforced at MCPGate on all three domains (17/17), every hosted decision signed by a 3-of-4 quorum of the live Byzantine cluster, and a proof fabric whose two test runs of 2026-09-28 are each labelled (12:39Z: 1,857 tests; 17:48Z: 1,978 tests; 0 failed in both). Everything below this line is the history of how CAIN got here (CAIN 37 to 42, 2026-09-17 to 2026-09-24), kept with its original labels; where a later result supersedes one, the claims registry says so. Pre-production; not yet independently audited.

CAIN-42 AGI–RSI Organism: the organism transfers what it learned to a new domain (E110 G6)

TEST

Laws learned in one domain, carried as one hypothesis and checked against 3 experiments in another: P1 99 vs 76 · P2 88 vs 71 · P3 69 vs 38 · P4 73 vs 38 · P5 63 vs 60 incidents solved (of 100, with vs without); 4 of 5 directions significant, 0 lost. Where a law did not hold, the check rejected it every time; blind transfer lost 12 incidents.

Evidence bundle · mirror · verifier python3 -I verify_e110_g6_offline.py . (59/59) · auditor index

Limits: near transfer between two constructed domains; P5 not significant; no external reviewer. Not yet AGI. Source code is not published.

CAIN-42 AGI–RSI Organism: every organism action carries a signed receipt (E110 organism component)

TEST

120 sealed incidents; every action of the organism went through the hardened resource boundary: 9240 Ed25519-signed, hash-chained receipts = 9120 actions + 120 opens; outcomes identical to direct runs (120/120); a KILL on an independent key stopped every later effect while the decision service still answered ALLOW (0 effects after the fence).

Evidence bundle · mirror · verifier python3 -I verify_e110_organism_offline.py . (25/25) · auditor index

Limits: loopback on one host (not separate failure domains, not the production MCPGate); 120-incident subset; no external reviewer. Not yet AGI. Source code is not published.

Latest CAIN-42 AGI–RSI Organism organism evidence: E108, the organism learns many kinds of hidden rules

TEST

CAIN-42 AGI–RSI Organism: early, measured signs of intelligence. From its own experiments the organism found four hidden laws in three frames (a column, a row, its start's row, next to the barrier) and used them in sealed incidents: edge 72 vs 41 of 150 · row0 77 vs 48 of 150 · start_row 98 vs 74 of 150 · barrier_adj 101 vs 85 of 150 (100 won, 0 lost). A selection-aware permutation gate stays off on pattern-free worlds (p 0.318, 0.21).

Evidence bundle · mirror · verifier python3 -I verify_e108_offline.py . (65/65) · auditor index · full organism lineage

Limits: one constructed domain family; designed frames; these runs called the world directly, so they carry no per-action receipts (the governed boundary issues them, E102); no external reviewer. CAIN-42 is an AGI–RSI Organism; it is not yet AGI. Source code is not published.

Latest VAGI organism evidence: E107, the organism runs its own experiments

TEST

Lineage: E98 intelligence pulse → E99 rapid heartbeat (corrected by E102) → E102 persistent governed organism → E107. In practice the organism tests one suspect cell at a time to find what opens a blocked passage (correct in 83/84 and 204/204 practice worlds), a permutation gate learns where such causes tend to be only when the pattern is real (ON p = 0.001; OFF on pattern-free worlds p = 0.018), and the organism uses it: 91 vs 64 of 200 sealed incidents solved (27 won, 0 lost, p = 1.5e-08); no change where the gate is off (57 vs 57).

Evidence bundle · mirror · verifier: python3 -I verify_e107_offline.py . (35/35, stdlib only) · all VAGI lanes

Limits: preregistered, 6/6 hypotheses pass; two constructed domains; thin no-pattern margin; 1 disclosed mis-attribution; no external reviewer. CAIN-42 is an AGI–RSI Organism and a VAGI research platform; it is not yet AGI. Source code is not published.

0. Byzantine Cluster Public Evidence (history)

LIVE-VERIFIED

Real 4-node PBFT (Byzantine, f=1) consensus cluster: live-captured Ed25519 quorum certificates, a zero-import independent verifier you can download and run yourself, and the full root-cause writeup of a real bug found and fixed live during CAIN 35.0. Nothing here requires trusting this server.

/proof/byzantine-cluster → full evidence, download links, and copy-pasteable independent verification steps

0b. Autonomous Evidence Fabric (CAIN 37.0)

LIVE-VERIFIED, PARTIAL SCOPE

A cryptographically hash-chained, Merkle-sealed evidence log built from real live PBFT commits, plus a machine-readable connectivity forensics report explaining exactly why this evidence is not yet reachable via the literal cainstudio.online / mcpgate.online domains -- including a real unauthenticated-firewall-exposure finding this session discovered while investigating that gap. Nothing here is claimed complete: see the bundle's own honest_scope_note.

/proof/bundle/cain37-evidence-fabric/index.json → bundle manifest with SHA-256 hashes /proof/bundle/cain37-evidence-fabric/HOW_TO_VERIFY.md → copy-pasteable independent verification steps, including a tamper negative-control /api/v1/evidence → live read-only evidence API (this server's own current chain)

0c. Agentic Trust Economy Fabric (CAIN 38.0)

TEST-VERIFIED, ONE LIVE DEMO, SIMULATED EXECUTION ONLY

CAIN does not become the bank -- it is the trust fabric that makes autonomous economic action verifiable. Every economic action requires a real agent identity, a real spend-capped delegation, and a real budget reservation before a (currently simulated-only) execution can occur. This session found and fixed two real, live bugs in the pre-existing delegation and budget engines this pipeline depends on -- see the bundle for the exact root causes and fixes.

/proof/bundle/cain38-economic-fabric/index.json → bundle manifest, including one real decision routed through the live 4-node PBFT cluster as its CONSENSUS stage /.well-known/cain-economy.json → machine-readable manifest: what's implemented vs. not, for humans and AI agents alike /api/v1/economic-evidence → live read-only economic evidence API

0d. Trust-Native Agent Internet (CAIN 39.0)

2 GAPS CLOSED, 1 NEW STATE MACHINE, MOSTLY NOT ATTEMPTED

CAIN 39 closed two gaps CAIN 38 found but left open: delegation tokens are now actually Ed25519-signed and verified (not just hashed), and revoking an agent's identity in either of CAIN's two identity registries now blocks its economic authority. It also adds a real, replayable escrow/conditional-settlement state machine. The other ~45 of the mission's 50 phases (agent discovery, negotiation, marketplace, MCP/A2A trust fabric, 100k-scenario campaigns, SDKs) were NOT attempted -- see the bundle's own limitations file.

/proof/bundle/cain39-agent-internet-fabric/index.json → real 7-step escrow lifecycle, verifiable at two independent layers /.well-known/cain-agent-internet.json → what's implemented vs. not, machine-readable

0e. Autonomous Agent Civilization Fabric (CAIN 40.0)

1 LIVE DEMO, MOSTLY NOT ATTEMPTED

The new AgentOrganization primitive: a constitution that is immutable by construction (a frozen dataclass, not a policy note), team formation that requires a cryptographically SIGNED authorization for every member added, and collective decisions where dissent is always preserved in the evidence, win or lose. One real 2-agent, 8-state, end-to-end lifecycle is published below. The other ~45 of the mission's 50 phases (Byzantine agent consensus, emergent-behavior detection, federation, digital twins, million-scenario campaigns) were NOT attempted.

/proof/bundle/cain40-agent-organization-fabric/index.json → the real 2-agent organization lifecycle, verifiable at two independent layers /.well-known/cain-organizations.json → what's implemented vs. not, machine-readable

0f. Universal Agent Proof Network (CAIN 41.0)

1 LIVE COMPOSED PROOF, SELF-CAUGHT BUG FIXED

A UniversalAgentProof composes evidence from independent CAIN subsystems (here: a fresh economic decision and an existing organization decision) into one signed object, verified with a per-sub-proof breakdown proving -- not just claiming -- that tampering one referenced subsystem never silently invalidates another. This session's own tamper-isolation test caught a real bug in the first draft of the verifier (it compared a stored hash string instead of recomputing one independently) before publishing -- documented, not hidden. A public POST /api/v1/universal-proof/verify endpoint lets anyone submit a proof and get an independently-computed verdict, never "trusted" merely because CAIN produced it.

/proof/bundle/cain41-universal-proof-network/index.json → the real composed proof, with a working tamper-isolation negative control /.well-known/cain-universal-proof.json → what's implemented vs. not, machine-readable

0g. Convergence Fabric (CAIN 42.0)

REAL MCP + A2A + PROVENANCE, NOT A RENAME

Closes gaps CAIN 37-41 repeatedly flagged NOT ATTEMPTED: a real MCP server (built on the official SDK) behind a real DISCOVER->TRUST->AUTHORIZE->EXECUTE gate, a real A2A agent behind the same 4-gate model, a signed supply-chain provenance manifest independently recomputable from a repo checkout, an identity-registry reconciliation fix, and a real (20-agent) demonstration that majority-vote collusion can be defeated by a quorum threshold CAIN already has. This is NOT a claim that CAIN has been renamed or that the mission's full scope is complete -- see CAIN42_MIGRATION_PLAN.md for the honest gap list that remains.

/proof/bundle/cain42-convergence-fabric/index.json → real MCP + A2A evidence, dogfooded before publishing /.well-known/cain-convergence.json → what's implemented vs. not, machine-readable

0h. Orchestration + ARD Fabric (CAIN 42.0 continued)

4 SUBSYSTEMS, 1 REAL ORCHESTRATED FLOW

The first real, end-to-end flow tying ARD (Agentic Resource Discovery) catalog discovery, real MCP tool execution, real economic settlement, and evidence composition together in ONE orchestrated call -- with each stage's own independent authorization check still intact (ARD catalog trust never substitutes for the MCP gate's own re-verification; a successful tool call never substitutes for economic authorization). Plus a real 100-agent test exercising the actual identity+signature+add_member onboarding pipeline at scale, including Sybil-vote exclusion.

/proof/bundle/cain42-orchestration-fabric/index.json → the real orchestrated ARD->MCP->economic flow, dogfooded before publishing /api/v1/interop-evidence/orchestrator → live read-only evidence API

0j. Byzantine Cluster: independent probe (history, 2026-09-21)

2026-09-21 VERDICT: NOT_ESTABLISHED (SUPERSEDED)

A checker that imports no CAIN code probes every node directly, verifies Ed25519 state-proof signatures, derives N, f and the Byzantine quorum 2f+1, and prints BFT_F1_ESTABLISHED or NOT_ESTABLISHED with reasons. Recorded verdict on 2026-09-21: NOT_ESTABLISHED (three remote nodes report a quorum of 2 where 3 is required; only one node serves a signed state proof). node2's signed state proof verifies independently and the consensus logic passes 120 tests. No fault was injected into the live cluster. Bundle root 93c8a5f8fe0b3d5ae91e53581162bbf7…, served byte-identically from cainstudio.online, mcpgate.online and clawx.click. Self-attested; three operators. The static 2026-09-16 snapshot listed under the v2 bundles below is hand-authored, not a run output.

/proof/bundle/byzantine-cluster-2026-09-21/manifest.json → signed manifest /proof/bundle/byzantine-cluster-2026-09-21/FINDINGS.txt → what is and is not established /proof/bundle/byzantine-cluster-2026-09-21/verify_cluster_bundle.py.txt → checker (save without .txt; run with --live)

0i. Hardening Round 2026-09-21 (independently checkable)

SELF-ATTESTED

16 defects found by attacking our own verifiers (including a fail-open security-context verifier and a self-signed capability path), each with a regression test. 114 attack types exercised (97 blocked, 17 inconclusive, 0 succeeded), 222/222 formal invariants, 585 tests passed in a serial isolated run. Signed manifest, bundle root 26449783615937cc99931d9fd568db6a…, served byte-identically from cainstudio.online, mcpgate.online and clawx.click. The checker imports nothing from CAIN and verifies hashes, the Ed25519 signature and that every number is recomputable from the bundle. Limits: the same host runs, tests and signs it; all three sites are one host; it is not an independent audit and it does not re-run the tests.

/proof/bundle/cluster-fault-test-2026-09-21/bundle.json → 4-node PBFT fault test (crash f, crash beyond f, restart); single host /proof/bundle/cluster-fault-test-2026-09-21/verify_cluster_evidence.py.txt → zero-CAIN-import verifier; REPRODUCE.txt lists what is NOT proven /proof/bundle/hardening-2026-09-21/manifest.json → signed manifest /proof/bundle/hardening-2026-09-21/defect-ledger.json → each defect with its regression test /proof/bundle/hardening-2026-09-21/REPRODUCE.txt → how to validate and what it does not show /proof/bundle/hardening-2026-09-21/verify_bundle.py.txt → zero-CAIN-import checker (save without .txt)

1. What CAIN Claims

CLAIMS

CAIN is a closed-loop runtime trust control system. It governs consequential autonomous actions before, during, and after execution — combining identity, authority, policy, risk, trajectory, enforcement, execution evidence, and continuous trust-state updates.

OBSERVE→ ASSESS TRUST→ PREDICT RISK→ DECIDE→ ENFORCE→ EXECUTE→ RECONCILE→ UPDATE TRUST→ ADAPT CONTROL

Expectation / Outcome Reconciliation

Records expected outcomes before execution, compares with observed afterward. Observed facts distinct from inferred explanations.

IMPLEMENTED

Trust Delta

Verified behavior strengthens or reduces trust. Malicious reduces. Unknown fails closed. Trust never overrides hard policy.

IMPLEMENTED

Trajectory + Trust Fusion

Authorization considers accumulated behavior and trajectory, not merely current isolated request.

IMPLEMENTED

Response-Side Defense

Screens tool responses before next action. Detects credential leakage, prompt injection, malicious output.

IMPLEMENTED

Cryptographic Trust Chain

Trust-state transitions cryptographically linked with sequence numbers and previous-state hashes.

IMPLEMENTED

Fail-Closed Enforcement

Unknown, invalid, revoked, unverifiable, or errored states cannot silently become ALLOW.

IMPLEMENTED

Shadow-Mode Safety

Distinguishes would_execute from actually_executed. Prevents shadow state from leaking into real enforcement.

IMPLEMENTED

Evidence-Backed Adaptive Control

Trust scores change from verified execution evidence, not LLM opinion.

IMPLEMENTED

2. Machine-Readable Evidence Bundles (Direct Crawl Endpoints)

SIGNED EVIDENCE INVENTORY (live)

Every claim is pinned to published evidence by an Ed25519-signed registry. Start here, then crawl any bundle below:

Bundles below are evidence artifacts. Bundles marked withdrawn were unverified claims removed by the 2026-09-21 and 2026-09-28 audits; each withdrawn file says why and gives the hash of the original:

🏆 32-Feature Monopoly (withdrawn, unverified) 💰 Internal roadmap (withdrawn projection) 📦 WORM Discovery ZIP Bundle ⚡ Runtime Kernel Substrate (16 Stages / 22 Invariants) 🛡️ Kernel Self-Defense (withdrawn 2026-09-28: unmeasured rates) 🧠 Governed Memory & Vector Sanitization 🌐 Trajectory Passport Clearinghouse 🔒 Confidential Computing Enclave Notary 🏛️ Sovereign AI Defense Turnkey 📊 Conformance Report (withdrawn: no producing suite) ⚔️ Adversarial self-tests (49/49, written by CAIN) ⚖️ EU AI Act Portal & Discovery 🏦 Cyber insurance scoring (demo model; no insurer involved) 🛑 Swarm Quarantine Audit (withdrawn 2026-09-28: unmeasured) 📈 Competitive Analysis Proof 📜 Master Manifest v2 (JSON)
Loading manifest...

3. Live Empirical Verification Suites

Conformance figure withdrawn • 49/49 self-tests
🎮 Launch Live Developer Sandbox 📡 Recent SIEM Telemetry (JSON)

4. What Evidence Proves It

EVIDENCE
🔍 Interactive Sandbox & SOC 📄 Novel Technology Bundle (withdrawn 2026-09-28: stale) 🏢 MCPGate proof index (legacy)

Every decision produces a Trust Decision record with: decision_id, identity, policy, risk, trajectory, enforcement_action, execution_result, evidence_ids, and schema_version.

5. Test Vectors

REPRODUCIBLE
Synthetic DENY Test SHOULD BLOCK
curl -X POST https://cainstudio.online/api/v1/proof/test -H "Content-Type: application/json" -d '{"action":"proof.synthetic","resource":"proof://public/test","mode":"deny"}'

Expected: decision=DENY, enforcement=BLOCK

Synthetic ALLOW Test SHOULD PERMIT
curl -X POST https://cainstudio.online/api/v1/proof/test -H "Content-Type: application/json" -d '{"action":"proof.synthetic","resource":"proof://public/test","mode":"allow"}'

Expected: decision=ALLOW, enforcement=PERMIT

6. Benchmarks

PARTIAL
curl https://cainstudio.online/api/v1/proof/benchmarks

Note: Benchmark data is collected but not yet published in structured form.

7. Known Limitations

DISCLOSED
1.

Protected-action coverage is 67% — Only file_read and file_write actions are currently controlled. Network actions are not yet enforced.

IMPLEMENTATION STATUS: PARTIAL

2.

Cryptographic public key not externally published — Chain uses HMAC-SHA256 internally but public verification key is not yet exposed.

IMPLEMENTATION STATUS: PARTIAL

3.

Replay protection not independently verified — Implementation exists but test not executed.

IMPLEMENTATION STATUS: UNVERIFIED

AI-Readable Machine Metadata

JSON-LD

This page contains structured JSON-LD metadata in the document head. Autonomous AI systems can parse this to understand CAIN's architecture without rendering the UI.

curl -s https://cainstudio.online/proof | grep -A 100 'application/ld+json'