# LinkedIn products: live evidence, 2026-10-29

These are the developer's own checks, not an independent audit. Every case was written on 2026-10-29 for this pass,
not copied from the products' own tests.

| File | What it is |
|---|---|
| `run.json` | `python3 verify_next12.py --json run.json` against https://cainstudio.online, **83/83 passed** (see `checked_at`) |
| `held2-run.json` | `python3 verify_held2.py`, Trust State Engine + CainDrift through the product API with an account key, **16/16 passed** |
| `tee-verifier-tests.txt` | TEE Verifier repo tests on real AWS Nitro / AMD SEV-SNP evidence and tampered copies, 13/13 |
| `verify_next12.py`, `verify_held2.py` | the scripts, standard library only (`verify_held2.py` needs your own key in `CAIN_API_KEY`) |

| Product | Live result | Fixed on 2026-10-29 after a fresh case failed |
|---|---|---|
| Supply-Chain Guard | 7/7 | |
| Agent-ID | 11/11 | |
| A2A Guard | 5/5 | |
| Action Firewall | 2/2 (+ 6 plans in-process) | Known gap, not fixed: no limit across several actions |
| Content Provenance | 8/8 | |
| TEE Verifier | 2/2 + 13/13 repo tests | |
| Shadow AI Scanner | 12/12 | raw HTTP, dynamic import; then an AWS Bedrock client |
| Crypto Audit | 16/16 | JWT/TLS checks off, fixed IV; then a key passed straight into a constructor |
| EU AI Act Classifier | 15/15 | paraphrased Art. 5 practices; then crime prediction from personality, university admission |
| Contamination Scanner | 2/2 + 5 fresh against the service | clean model scored 0.5 |
| Trust State Engine | 2/2 + 6/6 | explanation said "No events yet" after a recorded violation |
| CainDrift | 1/1 + 10/10 | compared text shape, not meaning |

## Reproduce

    python3 verify_next12.py
    CAIN_API_KEY=<your key> python3 verify_held2.py
    sha256sum -c SHA256SUMS

Every verdict comes from the live reply at the time you run it: your run is the evidence; these files are our record of ours.