CAIN-42 · CAIN-A75-056 · LIVE ENGINE
CAIN BenchProof
Benchmark Evidence Review. Visually complex knowledge-work evaluation.
Last reviewed 2026-11-11
Products / Runtime Containment & Evidence / CAIN-A75-056
DELIVERABLE LIVE ENGINE Evaluation Moat lineage: 2026 #56 · operator stream 10
An engine in this repository produces a bounded artifact from a supplied sample today.
Catalog name: Benchmark Evidence Review · Public name: CAIN BenchProof
Run it live — in this page
This is the product's real published rules, not a recorded demo. Pick the honest sample or one of its failure cases, edit the record if you want, and run it. The live route executes the same rules.py file the signed sample's offline verifier embeds.
Verify it yourself
Two different things, both on purpose. The live demo above is an unsigned trial of the published rules on the record you send. The signed sample is the verifiable record: it was produced by the operator CLI, is Ed25519-signed, and re-verifies offline with a verifier that imports no CAIN code.
- Signed sample: https://clawx.click/evidence/agentic-75-products-2026-11-12/deliveries/CAIN-A75-056/signed-sample/ (
run.json,report.md,public_key.txt,verify_offline.py,manifest.json). - Re-run the proof offline:
cd signed-sample && python3 verify_offline.py run.json --pinned-key "$(cat public_key.txt)". It printsVERDICT: VERIFIEDor a named failure. It does not import CAIN code. - Published rules:
rules.pySHA-256e139b66c993034cae38ee8bf3c0a89cd8c915517130cc8ed0a52e589238134a3. The live route reports the same hash. - Try-it evidence: every case through the public route + six browser runs.
- Full public bundle: all 75 signed samples, selftest and an independent bundle verifier.
What this does not prove: that the record is complete, that a remembered statement or supplied sample is true, or that the engine is a production assessment. The limits are printed with every result. No customer, partnership, certification or third-party audit is claimed.
What you buy
A defensible answer for this class of system: Treat benchmark artifacts like any other evidence: record the task, the model, the inputs, the grader and the result, and publish the failures. The run is bounded and the limits are printed with the result.
A bounded engine run on the buyer's sample: findings, a written report, and a SHA-256 artifact manifest. The manifest is unsigned: it is not a signature and not an attestation.
Price and terms (proposal)
| List figure | $2,500–$5,000 per review |
|---|---|
| Indicative band | $299–$999 one-time (report) |
| Lead time | 5 business days after a signed SOW and sample intake |
| Buyer | Benchmark authors and procurement teams |
The list figure comes from the individual offer; the band is the line default. Neither is an invoice. Scope is fixed in a one-page SOW before work starts.
Buy this review — from $299 fixed scope, signed evidence bundle, payment link or invoice
Scope
One incident or action sample; one evidence package; verifier included.
Inputs we need
- One incident or action trace
- Current policy set
- The evidence format the buyer needs
What you receive
findings.jsonreport.mdmanifest.json
Explicit exclusions
- No vendor integration or partnership is included or implied.
- No legal, tax, clinical, insurance or safety certification advice.
- No third-party audit, SOC 2 report, ISO certification or patent filing.
- No production write access; work runs on samples the buyer supplies.
- No 24/7 response unless a separate retainer is signed.
Engine and evidence
This SKU is engine-backed: bench_evidence in cain_a75_engines/benchevidence/engine.py. The operator runs it on your sample with python3 -m cain_agentic_products.cli deliver CAIN-A75-056 --out DIR.
The 2026 development and the CAIN-42 mapping · Check the mechanism · Fire a live governed decision · Public evidence
Next step
Order it directly at the fixed-scope buy page, or book a 20-minute scoping call. Bring one sample and the decision you need to defend. If the sample is not a fit, we will say so before you pay.
What this offer is not. No customer, partnership, endorsement, certification, patent or third-party audit is claimed. Prices are proposals, not validated rates. DELIVERABLE means a bounded engine run on a sample, not an assessment of a production system. Built by a three-person team.