What is public
Original fictional packets, the business briefs, answer keys, reference workbooks, evaluation code, website, API execution records, delivered artifacts and author reviews are published under the repository’s MIT license. Trademark rights are excluded. Customer documents, credentials, private account context and Lev internal prompts are excluded.
What the scores establish
This is a six-case diagnostic release with one fresh attempt per system and case. Sources and keys were frozen before execution, then the outputs received machine-assisted author review. Independent CRE practitioner validation, repeated trials and broader authentic-document qualification remain outstanding. Check counts are not independent deals; no confidence interval or industry-wide superiority claim is made.
Corrections with a visible history
- Uniform exclusion of the ambiguous insurance-evidence criterion
- Presentation normalization without financial answer repair
- Source-location review interpretation and rationale
- Exact-context resumption after gateway credit exhaustion
Original records remain available. A correction applies to every affected result, including results unfavorable to Lev.
Contributors and provenance
The initial scope was informed by a methodology engagement with The AI Consulting Network. This release uses an original implementation and original synthetic cases. It does not redistribute the consultant’s private materials or imply his endorsement. Case authoring and review used AI assistance; no independent practitioner review is claimed.
What comes next
Independent CRE review of the keys and memo rubric, a larger held-out corpus with permitted authentic layouts, repeated trials, separate Index ingestion evaluation, offering-memorandum tasks, complete cost attribution and a fully public artifact-generation adapter. These are future work, not completed coverage.