CAIN-42 CAIN StudioAGENTIC AI 75 · #1
Full-day autonomous software engineering
OpenAI reported a Codex experiment sustaining roughly 25 hours of implementation, testing and repair on one software project.
Last reviewed 2026-11-11
Agentic AI 75 / #1
2026 status: Research demonstration · Category: Software engineering
The 2026 development
OpenAI reported a Codex experiment sustaining roughly 25 hours of implementation, testing and repair on one software project.
The governance gap
A run that long crosses many context boundaries. The trust question stops being whether the model can write code and becomes whether anyone can reconstruct which authority, files, tests and decisions each hour ran under.
Where CAIN-42 fits: Moat 4 — Execution Provenance & Attestation
Core question: Can we prove what actually happened? Hash-chained, signed records binding decision to execution to observable effect.
Bind every long-horizon run to a hash-chained record: the action proposal, the authority it ran under, the test that failed before and passed after, and the signed verdict. KEYSTONE already does this for bounded repo fixes; the same spine extends to a multi-hour run.
How the wiring works
- Open a run record per task with tenant, agent identity and the exact repository revision.
- Wrap each tool call so the record captures proposal, authority, policy revision and result digest.
- Require an independent reference test that fails on the original tree and passes on the fix.
- Close the run with a signed verdict that names every check, and export an offline verifier.
What is real today
LIVE A public route in this deployment exercises this mechanism today, within its stated scope.
The hosted KEYSTONE loop proves bounded fixes, not a 25-hour run. Long-horizon duration and multi-hour context retention are library paths, not a demonstrated hosted capability.
| Component in this repository | What the module says it is |
|---|---|
keystone/repo.py | The loop at repository scale: scan a tree, and prove a proposed fix with the project's own tests. |
keystone/guard.py | Fix-safety checks: the fix may repair the defect, but it may not break the code around it. |
keystone/engine.py | KEYSTONE: FALSEWORK finds what is fake, Claude fixes it, and every fix is proved and signed. |
keystone/bundle.py | KEYSTONE evidence bundles: the signed records packaged neatly for an auditor (a zip). |
cain/mission_control/longhorizon.py | Evolution 15 — long-horizon adversarial evaluation of governed missions. |
Descriptions are the first line of each module's own docstring, read without importing it. A module path is a pointer into this repository, not a hosted endpoint. Its existence is checked by the test suite; its status for this use is the label above.
Check it yourself
- Run the live KEYSTONE loop
/keystone - Fire a live decision through the CAIN-42 pipeline
/fabric/try— the same enforcement path every product sits behind.
Commercial package
| Offer | Long-Run Engineering Attestation |
|---|---|
| Buyer | AI coding-tool vendors and platform engineering teams |
| Price (proposal) | $3,500 per run audit; $1,500–$6,000 fixed audits |
| Bundle | Code & Proof — Prove the fix, not just the diff. |
| Operator stream | Stream 1 of the 10-stream operator playbook |
Platform engineering, AI coding-tool vendors, security engineering is the bundle buyer. The entry offer for this bundle is Fabrication audit or long-run engineering attestation; the range is $1,500–$6,000 per audit; $2,500–$4,000/month retainer. A price on this page is a proposal, not a validated rate, and no customer is claimed.
Buy this as a product or service
CAIN-A75-001 — Long-Run Engineering Attestation DELIVERABLE
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.
Kind: Assessment · Price (proposal): $3,500 per run audit; $1,500–$6,000 fixed audits · Lead time: 5 business days.
Open the product page · all 75 products
What this page does not claim
What this page is not. CAIN-42 is not affiliated with, endorsed by, integrated into, or shipped with the organisation named in the 2026 item. The item is public context; the page describes how CAIN-42 would govern that class of capability. No page here claims a customer, a partnership, a certification, a patent, or a production integration. Prices are proposals. Built by a three-person team.