Repository navigation
v0.2.0 - real-mode receipt against actual Phoenix pipeline
First release with a real-mode receipt bound to an actual run of the published Phoenix pipeline.
What's new
- Real-mode receipt at `examples/x-feed-real.receipt.json`, bound to `xai-org/x-algorithm@c3ef330`. Generated by cloning the upstream repo, downloading the 2.9 GB Phoenix LFS artifacts, running `uv run run_pipeline.py`, hashing all 8 model and config files, and signing the audit event.
- Reproducibility demo (`npm run repro-demo`) showing wrapper-level determinism: two independent runs with the same input produce identical `input_commitment` and `output_root`.
- Tamper detection demo (`npm run tamper-demo`) showing the verifier rejects any modification to a signed payload.
- CI workflow (`.github/workflows/verify-receipt.yml`) verifying the receipt on push, PR, and daily schedule. Copyable into any recommender repo.
- LIMITATIONS.md documenting the canonical caveats: execution binding vs ranking quality, mock vs real, output-bytes vs parsed-records, observer trust, regulatory framing.
Verify
```sh
git clone https://github.com/VeritasActa/x-algorithm-receipts
cd x-algorithm-receipts
npm install
Mock-mode receipt (instant, no artifact download):
npm run demo
npx @veritasacta/verify examples/x-feed-demo.receipt.json --jwks examples/demo.jwks
Real-mode receipt (already in the repo, bound to xai-org@c3ef3307):
npx @veritasacta/verify examples/x-feed-real.receipt.json --jwks examples/real.jwks
Inspect semantic fields:
node scripts/inspect-receipt.mjs examples/x-feed-demo.receipt.json
node scripts/inspect-receipt.mjs examples/x-feed-real.receipt.json
Tamper demo:
npm run tamper-demo
Reproducibility pair:
npm run repro-demo
```
In-browser verifier
For no-CLI verification, drop a receipt + JWKS at https://www.scopeblind.com/verify-receipt.
RFC
#1 Add optional post-ranking audit event hook proposes a minimal in-tree hook for `xai-org/x-algorithm`. Filed on this repo because upstream Issues are disabled.
What this proves
Execution binding, not ranking quality. The receipt proves a specific output was bound to specific code, model artifacts, config, and input commitments at signing time. It does not prove the output is fair, truthful, or beneficial. See LIMITATIONS.md.
Related
- IETF: draft-farley-acta-signed-receipts
- Verifier: `@veritasacta/verify` (Apache-2.0)
- Lesson 18 in microsoft/ai-agents-for-beginners walks the same primitive for agent tool calls.
- Blog: scopeblind.com/blog/x-algorithm-receipts (canonical) / veritasacta.com/blog (reference-implementation framing).