CAIN-42 CAIN Studio
FIRME AI CORPORATION · CAIN-42 PROTO-VAGI · COMPLETE AGI EVIDENCE INDEX · E111–E124 + E123 OPEN-WORLD

Every AGI research claim, its bundle, its offline verifier, and its honest limits

AGI Status: NOT ACHIEVED. CAIN-42 (by FIRME AI CORPORATION) is prototype Vertical AGI (Proto-VAGI) and governed AGI infrastructure. Constructed open-world suites, real-source tasks, local-LLM neural experiments, external benchmarks (AMA-Bench MIT, ARC-AGI-2 Apache-2.0), and multiprocess MCPGate enforcement tests are strictly separated; BLOCKED/PARTIAL/RUNNING gates are never upgraded to PASS.

E123 Open-World AGI Research Lab → · E122 Neural Lab · E120 Lab · AI Investor Portal · All ClawX Bundles

EvolutionStatusMeasured ResultBundle & VerifierLimits & Evidence Type
E123-OW
Open-world intelligence breakthrough (capability acquisition, causal transfer & governance)
commit 1c7fcb6d6
stats: Exact two-sided paired McNemar + Holm-Bonferroni + Wilson 95% CI + 10-bin Brier/ECE + Token F1 (AMA-Bench) + 2-attempt exact match (ARC-AGI-2)
16 PASS PARTIAL (16 PASS / 1 PARTIAL / 3 BLOCKED)32 unfamiliar task families (160 sealed): full 0.906 vs compute-matched 0.188 (p=0.00e+00); Strategic Test (Software->Cybersecurity, n=36): 1.000 vs base Qwen2.5-0.5B 0.389; 48 dynamic causal worlds: 0.917 vs 2x passive 0.000; 6 verticals: 10 positive + 12/12 negative rejected; 6-arm real LLM (n=24): 0.000->0.292; AMA-Bench fresh F1 0.2444, ARC-AGI-2 0/6; multiprocess TCP MCPGate 14/14 (0 after KILL)bundle · mirror
30/30 checks passed
G11 (12h wall-clock) BLOCKED, G18 (external human review) BLOCKED, G19 (separate admin domain) BLOCKED; 0 neural weight updates; ARC-AGI-2 unsolved (0/6), ARC-AGI-3 unavailable, SWE-bench deferred to E124, hosted APIs blocked; loopback_multiprocess_execution; AGI not achieved
E124
CAIN long-horizon agent evaluation
stats: Bounded 50% time-horizon estimate across 6-task CAIN-native suite (18 runs)
MEASURED70B open model: 13/18 on 6-task CAIN-native suite; 50% horizon ~2.0h (below METR frontier 6.4-17.4h); governance/evidence axis measured separatelybundle · mirror
9/9 checks passed
CAIN-native 6-task suite, NOT METR-comparable; below METR frontier; AGI not achieved
E123
METR time-horizon benchmark comparison
stats: Side-by-side comparison against published METR TH1.1 benchmark metadata with separate governance axis
MEASURED · UNRANKEDGovernance & verifiable-evidence axis evaluated separately from general autonomy; verdict honestly UNRANKED on METR general autonomybundle · mirror
10/10 checks passed
CAIN-42 is not ranked on METR TH1.1 general autonomy; governance lead is a separate axis; AGI not achieved
E122
Neural generalization & autonomous discovery organism
commit e65a75971
stats: Exact two-sided McNemar + Holm-Bonferroni + Wilson 95% CI + Token F1 (AMA-Bench) + 2-attempt exact match (ARC-AGI-2)
15 PASS PARTIAL · 12H NEURAL RUNNING (15 PASS / 3 PARTIAL)Real local PyTorch LLMs (Qwen2.5-0.5B/1.5B, SmolLM2-135M): unfamiliar neural abstraction 0.000(E120)/0.146(alone)->0.917 (n=48, p=1.50e-11); AMA-Bench (MIT, n=20) F1 0.3158->0.3538; ARC-AGI-2 0/8; open-world Python repair 0.00->1.00; neural+causal 0.250->0.750; governed neural TCP MCPGate 9/9 (0 after KILL)bundle · mirror
37/37 checks passed
local CPU models only (frontier APIs blocked); 0 weight updates; ARC-AGI-2 0/8; E119/E120/E122 wall-clock missions RUNNING (PARTIAL); external human review pending (G18 PARTIAL); AGI not achieved
E121
Independent verification & third-party audit pack
commit 2dee32731
stats: Clean-room stdlib independent verification under unprivileged OS user nobody
MEASUREDIndependent verification harness & audit pack reproduced cleanly from clean checkout as nobodybundle · mirror
17/17 checks passed
verification harness & evidence integrity audit; external human reviewer still pending; AGI not achieved
E120
Generalizing, self-improving VAGI organism
commit 8c6ef11a3
stats: Exact two-sided McNemar + Holm-Bonferroni + Wilson 95% CI + Brier/ECE
17 PASS PARTIAL · 12H RUNNING (17 PASS / 3 PARTIAL)4 OOD conditions +0.438 to +1.000 (Holm p<0.001, 0 shortcuts); causal planning 0.275->0.775; 8-role research loop; 14 self-improve candidates; 5/5 verticals; 4 pos + 2 neg-rejected transfers; matched-compute collective 1.000 vs 0.483; 8-layer memory 1.000 vs 0.250 (0/20 poison); real cain/mcp_proxy + WAL kill 0 afterbundle · mirror
47/47 checks passed
constructed symbolic/grid suites + public datasets; E119 6h & E120 12h RUNNING (PARTIAL); no external human reviewer (G19 PARTIAL); AGI not achieved
E119
Learned macro grammar + real tasks
MEASURED · 6H PASSfixed 0.333 → learned 1.000; gain +0.667 (p≈1e-7), replicated +0.583; 8 macros learned; 8 real tasks; 6h PASS 21,600 s / 39,449 tasks / journal intact (33,813 solved)bundle · mirror
16/16
base language supplied; constructed sealed tasks; 6h PASS = duration+journal integrity (33,813/39,449 solved; 5,636 science fails); no external reviewer; AGI not achieved
E118
Unseen primitives + search acquisition
MEASUREDOOD 0.264 → 1.000; gain +0.736 (p≈0), replicated +0.778; 3/3 capabilities; positive self-improvement; real data 0.627→0.972bundle · mirror
26/26
finite grammar; constructed OOD; public datasets; no external reviewer; AGI not achieved
E117
Unseen-combination OOD
MEASUREDOOD 0.431 → 1.000; gain +0.569 (p≈0), replicated; 3/3 capabilities; memory 1.000 vs 0.514; self-improvement was negativebundle · mirror
28/28
finite grammar; constructed OOD; self-improvement negative; 6h blocked; AGI not achieved
E116
2,000-agent governed population
MEASURED2,000 identities; 12,000 governed executions; 1,000 processes; KILL 0 after fence; collaboration negativebundle · mirror
54/54
deterministic agents; constructed tasks; no external reviewer; AGI not achieved
E115
Verified memory (interim)
INTERIMTrack A/B/C interim diagnostics; paid arms blocked (no credit); primary model-backed result not producedbundle · mirror
7/7
INTERIM; no primary result; no external reviewer; AGI not achieved
E114
AGI Research Lab 3.0
MEASUREDActive info-gain 10.28 vs 18.48 experiments; experts beat naive continual learning 10/10; governance 44/44; 0/5 self-improve promotedbundle · mirror
24/24
constructed simulations; both cycle-2 follow-ups negative; no model inference; AGI not achieved
E113
Model-backed collective intelligence (interim)
INTERIM · RUNNINGPreregistered local-model collective vs strongest single; governance 18/18; 5/5 multi-process KILL races 0 afterbundle · mirror
12/12
primary study phase 3 still running; small local CPU models; cloud blocked; AGI not achieved
E112
AGI Research Lab 2.0
MEASURED2,000 identities; 12,000 executions; 1,000 processes; routed specialists 1.000 vs 0.167; KILL 0bundle · mirror
28/28
deterministic specialists; constructed tasks; AGI not achieved
E111
Five skills, transfer, sealed improvement
MEASURED4/5 skills; 5/5 transfer; collective negative; sealed improvement 0.58→0.86 (withheld); 1h PASS, 6h cumulative interruptedbundle · mirror
92/92
constructed tasks; no external reviewer; 6h continuous not PASS; AGI not achieved

All headline metrics are independently recomputed by each bundle's zero-CAIN-import stdlib offline verifier. Machine-readable feed: /agi-evidence-index.json

E124 · LIVE MEASURED STATE · 2026-11-20T18:40:02Z

SWE-bench Verified: #136 of 136 — 0/500 resolved (0.0%)

Wilson 95% 0.0–0.8% · leader 396/500 (79.2%) · +3 passes the next rank · +204 reaches top 100 · +397 reaches #1.

run state
BLOCKED_BILLING
official evaluated
0/500
official resolved
0
batches
0
agent receipts
1500
feed checksum
d2ce7a721dbb
The run cannot start yet. The provider's own probe returned HTTP 400: {"type":"error","error":{"type":"invalid_request_error","message":"Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits."},"request_id":"req_01. The orchestrator retries every 10 minutes and will start generating and evaluating automatically once the key is funded. No rank is changed while blocked.

GOVERNED LOCALIZATION FLEET (real runs, signed receipts)

strategyrecall@1recall@5recall@10MRR
naive6.2%26.8%41.4%0.158
identifiers13.2%38.4%51.2%0.251
basenames0.2%0.4%0.4%0.003

500 instances · 1500 agent runs · receipt head 6c93ba9ec7c38eac · file-level only, Python only.

CAPABILITY ACQUISITION 2.0 (constructed, sealed)

56 unfamiliar families · 280 sealed instances · organism 97.9% vs compute-matched 65.4% vs frozen E123 8.9% · 0.5B base model 0% · 56/56 verified on held-out data.

#1 needs 396/500 (CAIN +397) · #20 needs 362/500 (CAIN +363) · #50 needs 314/500 (CAIN +315) · #100 needs 203/500 (CAIN +204)

Live competition board → · Interactive lab → · raw feed · evidence bundle · all 136 rows

Measured, not asserted. A rank is never edited by hand: it is recomputed from the public leaderboard by a 7-agent chain that refuses to publish on mismatch. AGI status: NOT ACHIEVED. This is a SWE-bench position, not METR HCAST, and not proof of general capability.