Manifest · artifact integrity
Phase 7 — Manifest
Readable view of the Phase 7 manifest. The hashes let a reviewer confirm that the report matches the executed inputs.
Run identity
| Benchmark | phase7-real-world-v1-end-to-end-rag |
|---|---|
| Run ID | 45e479aa-3689-48ad-b555-1fb3452b9b3b |
| Created (UTC) | 2026-10-31T15:58:14Z |
| Report | reports/phase7_end_to_end_rag.json |
| Fixed data | 51 passages · 40 cases · 200 candidates |
SHA-256 fingerprints
| Artifact | SHA-256 |
|---|---|
| Report | 17d270958719939d55f4ecf41de00c1879a52776a8209f6ea64f4e6674ffdd69 |
| Answers | 97beb4d96f47b0abf41a02313b10f23d8376b634a40d382bebbc8ca6a48c40b7 |
| Judge outputs | 926b805b818ce3140b6c09f9df5325d7a069ba32ca33fc21a41a8aa98b6ca0fa |
Models and inputs
Models executed: gpt-6.1-sol (OpenAI) and deepseek-v4-pro (DeepSeek). The manifest also pins 25 input fingerprints (scripts, data, prompts) so post-run drift is detectable.
This page presents the manifest substance. The complete file remains
reports/phase7_end_to_end_rag_manifest.json in the repository.