CAIN-42 CAIN Studio
CAIN-42 LIVE LAB connecting… competition board · evidence · home

Interact with CAIN-42

A governed research organism you can watch and test in public. Every number on this page is produced by a running process, a signed journal, or a search you started yourself. AGI status: NOT ACHIEVED.

1 · Live governed mission

loading…
–

Last research unit

–

Durability

–

2 · Run your own experiment — make CAIN learn a rule

Training examples

Each row is an input (JSON) and its output. Give CAIN 4–12 examples of a rule only you know. It searches for a program, not a lookup table.

#input (JSON)output

Your held-out checks (optional, scored by CAIN)

#input (JSON)expected

Query

Result

No run yet. Nothing here is pre-recorded: the engine executes on the server against your rows.

Public experiment stream

–
–

Rate limit 40 runs/hour per visitor. Inputs are capped at 64 KB. Only a salted IP hash and the result are kept, so the public stream is reproducible without exposing anyone.

3 · Measured organism results (constructed, sealed)

56 new unfamiliar task families · 280 sealed instances · 8 arms · matched hypothesis-evaluation budgets
raw results JSON →
loading…

Arms on sealed instances

armaccuracybar

By structural paradigm

paradigmaccuracy
–

3b · Governed localization fleet (real agent runs)

Each instance × strategy is an independent agent run with an Ed25519 receipt in a hash-chained ledger. This is the prerequisite the patching agent uses to read the right files.
–
strategyrecall@1recall@5recall@10MRR
–

4 · Verify the evidence yourself, in your browser

This fetches a published manifest, downloads each listed file, and recomputes SHA-256 with your browser's WebCrypto. A mismatch is shown as FAIL. CAIN's servers are not trusted for this check.

idle
Manifest format: {"files":[{"path":"…","sha256":"…"}]} — paths are resolved relative to the manifest.

5 · Share the live board

Share on LinkedIn Open the competition board Raw live feed (JSON) API status (JSON) Every figure links to its raw file. No claim without evidence.
What this page is. A public window into a governed research run: a real hash-chained mission journal doing capability-induction work every cycle, a real rule-induction engine that runs on your inputs, and a client-side verifier.
What it is not. It is not AGI, not a frontier-model chat product, and not a promise of benchmark position. The competition board reports CAIN's measured SWE-bench Verified result even when it is last.