Live evaluation reports.
Public reports from ZeroIndex projects, generated by @zeroindex-ai/eval-pack. The evidence behind “tell the truth about AI.”
-
00 / ask-zeroindex
RAG chat widget on zeroindex.ai
LLM-as-judge over 30 hand-labeled questions across positive, negative, adversarial, and multi-part categories.
-
01 / contract-lens
Structured contract extraction, verified
Deterministic checks over the PDF -> JSON pipeline: field-fact accuracy, party recall, and per-field citation verification — a hallucinated or mis-paginated quote is flagged against the source.
-
02 / intake-zero
Prospect-intake classify + triage
Programmatic checks over the Claude classify/draft steps: engagement-type and fit-score calibration, draft rules, and prompt-injection resistance.
-
03 / repo-xray
Repository onboarding reports, cited
Golden repos through the agentic analyzer: citation-resolution (every claim re-resolved to its exact lines), kept findings, and an onboarding-quality judge — with the synthesis model as an A/B dimension.