# Machine-reader notes for this evaluation hub

Canonical scope: https://evals.tayloramarel.com/

Respectful automated reading of these public research pages is welcome.

- Use GET or HEAD, one concurrent request, and at least three seconds between requests.
- Identify automated requests with a descriptive User-Agent. Cache responses when their headers permit, and reuse ETag validators when provided.
- Honor published robots.txt rules and Retry-After responses. Do not bypass access or rate controls, submit forms, probe private routes, or perform security testing.

## Evidence with exact source bindings

The [typed-decision correction](https://evals.tayloramarel.com/jev) has a [structured recount](https://evals.tayloramarel.com/jev/report.json) and [source register](https://evals.tayloramarel.com/jev/source-register.json). The register identifies exact public Git revisions and file digests.

The saved study has five scenarios. Each answered engine-scenario row has three checks. The numbers 14/15 and 9/15 describe checks, not fifteen independent trials. Ten Jev slots were unavailable; no Jev answer or provider call was recorded. An unavailable slot has no model-quality score.

The records distinguish configured model names, reported identities, local refusals, answered rows and provider calls. A policy hold is not a completed useful action.

The [research index](https://evals.tayloramarel.com/research.md) separates study records, publications and personal context. Baltor's product has a separate [agent reading guide](https://baltor.ai/agents.md).

Scope: this guide covers public reading of this evaluation hub, not private packets, application status, provider execution or publication authority. Other studies and older blog/PDF claims remain outside the Jev recount; a keyword match is not a legal or advice-quality assessment. Source terms govern reuse.