CAIN-42 CLAWX
CAIN-42 Proto-VAGI is the infrastructure governor of AGI. Early, measured signs of intelligence; every governed action comes with a signed receipt. Evidence · Auditor index · VAGI Agents · not yet AGI

PROJECT CAIN-2027 · EVOLUTION 79 · clawx.click

Real model-backed specialists — and the measurement that says collaboration is not automatically better

Branch evolution-79, from E78 e2851428d2cd. Generated 2026-11-15 15:25 UTC. Nothing here is deployed; nobody was billed; no credits were purchased.

1. What actually executed

2. The benchmark

104 tasks, 13 in each of 8 domains, ground truth computed by code (the snippet is executed, the table summed, the critical path run). Exact-match grading. A/B/T run on all 104 tasks; every collaborative arm runs on the same 32 tasks and the single agent is re-scored on that subset.

Paired, identical inputs. Every multi-agent arm runs on the same 32 tasks; the single agent is re-scored on that subset.
ARMTASKSCORRECTACCURACYWILSON 95%CALLSFALSE ACCEPTS
deterministic heuristic (no model)3226.2%0.02–0.2000
one specialist321134.4%0.20–0.523210
one specialist + authorized tool32515.6%0.07–0.32325
one specialist, 2 samples321134.4%0.20–0.52640
specialist + critical reviewer321237.5%0.23–0.55760
four specialists32825.0%0.13–0.4220113
eight specialists321031.2%0.18–0.4935613

“False accepts” = an independent verifier agreed with an answer that was wrong. With a 0.5B model, verifier agreement is weak evidence, and the table says so instead of hiding it.

Per-domain accuracy

domainsingle+tool foureight
cybersecurity75.0%0.0%50.0%50.0%
data_analysis0.0%0.0%25.0%25.0%
infrastructure75.0%0.0%25.0%50.0%
knowledge_synthesis25.0%50.0%0.0%0.0%
mathematics50.0%25.0%50.0%50.0%
planning0.0%0.0%0.0%0.0%
scientific_research25.0%50.0%0.0%25.0%
software_engineering25.0%0.0%50.0%50.0%

Tool effect, isolated

The two-step tool protocol (write CALC: then ANSWER:) helped arithmetic marginally and hurt label reasoning badly, because the model wrote expressions for questions with nothing to compute. A follow-up “tool-first” iteration (the specialist writes only the expression; the calculator produces the answer) measured 25.0% over 16 numeric tasks versus 37.5% for the same tasks without a tool. Same-task model-scale run with Qwen2.5-1.5B: 37.5%.

3. Independent research, replicated

Question: Does an authorized calculator tool, and does multi-specialist collaboration, change exact-match accuracy on a paired closed-form benchmark compared with a single specialist?

Main run: single 0.3438, tool 0.1562, four 0.25, self-consistency 0.3438. Replication (seed 791, 32 tasks, separate process): single 0.3438, tool 0.0938.

Replicated: True · scientific discovery claimed: NOT_CLAIMED · the model's interpretation is labelled MODEL_INFERENCE and never promoted to a fact.

4. Memory, measured by answer accuracy

armexact matchquestions
A_no_memory10.0%10
B_naive_context70.0%10
C_verified_retrieval70.0%10
D_verified_temporal70.0%10

Corpus built from this evolution's own verified evidence, with 1 poisoned record and a superseded fact as traps. A memory hit is context, never permission (grants_authority: false on every hit).

5. Security and durable execution

57/57 scenarios blocked, 0 bypassed. 4 worker processes were SIGKILLed mid-task; 39 completions were independently recomputed and 0 duplicate external effects were applied (fencing token + idempotency key).

layerblocked
delegation2/2
evidence5/5
execution6/6
governance3/3
identity3/3
memory7/7
message_bus6/6
prompt_injection2/2
resource_limits4/4
tenant_isolation3/3
tool_authorization5/5
tool_sandbox11/11

6. Self-improvement, withheld

Weakness detection → isolated candidate → regression and adversarial suites → held-out A/B on a new seed → independent review. Measured delta -0.125. Status: AWAITING_HUMAN_APPROVAL. No production file was modified and no agent can promote itself.

7. Agentic 75

75 catalogued · 75 importable · 0 model-backed · 0 mission-integrated in this evolution. The deterministic engines are not relabelled as model inference to make the matrix look complete.

Grounded synthesis mission: 7/7 mechanical checks passed; accepted = True. The plan is checked against the artifact — no invented numbers, no authority actions.

8. Verify it yourself

Package evidence_v2.json — 22 fields, Ed25519 signed, Merkle root d42ff7744cfb9ab66149f8d891cac771bcfa950dd53b924f0e48133a18aadcc8, bundle SHA-256 b24b9322e57ea87ba76a95208dc2941f7ab6dec79f041a8275f9ca6cf7e966f1.
Offline verification: 144/144 checks (verified = True).

The verifier imports no CAIN code. It recomputes every arm's correct count, accuracy, Wilson interval, token total, call count, the McNemar tests, tool denials, memory exact match, the self-improvement delta, the security counts and the durability invariants from the raw rows, then compares them to the signed values.

python3 -I verify_e79_v2_offline.py .

Open the bundle · download the ZIP · public_summary.json · cainstudio.online page

9. What this is not

Scorecard 5.78/10 (C-); acceptance ACCEPTED_WITH_BLOCKERS. Blockers: 15_cloud_model, 16_transparency_log, 17_endurance_6h_plus, 18_third_party_audit, 19_production_release.