CAIN-42 evidence library
VAGI Behavior Suite v1 — evidence bundle
Evidence bundle: 0 claims, 0 of 0 invariants held.
Last reviewed 2026-11-19
Browse the raw bundle · SHA-256 manifest
Verify it in your browser
Your browser downloads the bundle's SHA256SUMS manifest and the file(s) behind this page, hashes them with SHA-256 locally (WebCrypto), and compares. A match shows the record you are reading is the published one; it does not by itself prove who published it (see the signed claims registry and the bundle verifier for that).
The bundle's own README
VAGI Behavior Suite v1 — evidence bundle#
Generated: 2026-11-18T13:59:49.980105+00:00 Seed: 9402
What ran#
- 10 domains x 50 held-out tasks = 500 tasks.
- Arms: none (floor), fixed (one procedure), library (hand-written procedure set), governed (library + authority gate).
- In-distribution: none 72/500,
fixed 112/500, library 500/500, governed 500/500.
- Distribution shift: 10 domains x 10 unfamiliar variants = 100 tasks;
library 5/100.
What this is not#
- Not a public benchmark (tasks are constructed).
- Not a frontier-model result (the library is hand-written procedures).
- Not independent (no external verifier, no witness).
- Not AGI/VAGI/production/certification.
The 500/500 is a procedure ceiling on the families the library was written for. The 5/100 shift result is the more important number: it is the generalization gap.
Verify#
python3 verify_offline.py . python3 verify_shift.py .
Both verifiers import no CAIN code. They recompute the counts from the raw rows and check the commitment hash against the published task bodies. Integrity is checkable; independent witnessing is not claimed.
Try CAIN-42 on your own agents
Create a free account and every new account starts with a 30-day trial of the full platform. Or try the sandbox first, with no account at all.
Create a free account → · Try the sandbox · See the whole ecosystem