# Research index and evidence status

Research pages and source records at https://evals.tayloramarel.com.

## Typed-decision evidence correction

The [September 25 study correction](https://evals.tayloramarel.com/jev) is tied to [saved run records](https://evals.tayloramarel.com/jev/source-register.json). Its [machine-readable recount](https://evals.tayloramarel.com/jev/report.json) retains all 40 recorded engine-scenario slots: 25 answered and 15 unanswered, with 10 recorded provider model calls.

Jev was not measured: ten slots were unavailable and no provider call was recorded. The two answered Tactical conditions each contain five scenarios and fifteen policy checks. The correction does not establish legal correctness, a benign-twin false-positive rate, quantization performance or a general model-quality ranking. A [short PDF](https://evals.tayloramarel.com/jev/correction-2026-10-18.pdf) presents the same correction.

## Other study pages

- [Evaluation-execution divergence](https://evals.tayloramarel.com/divergence): Research on differences between stated judgments and actions in task settings.
- [DueCare](https://evals.tayloramarel.com/duecare): Research on exploitation indicators and worker-oriented responses.
- [Camouflaged Alpha](https://evals.tayloramarel.com/alpha): Financial research and cross-document hypothesis exploration.

## Distinct public purposes

The [personal homepage](https://tayloramarel.com/) gives biography and prior-work context. The [publication index](https://papers.tayloramarel.com/) lists papers and working notes. The [blog](https://blog.tayloramarel.com/) contains essays. Baltor's product documentation and account service are separate from this personal research index.

Machine-readable entry points: [llms.txt](https://evals.tayloramarel.com/llms.txt) and [agents.md](https://evals.tayloramarel.com/agents.md).

Scope: the Jev recount is source-pinned; the other linked studies and older PDFs are not revalidated here. The blog still contains conflicting Jev and quantization claims. Posting does not establish peer review or predictive performance. Fellowship/application materials are separate from measured research, and this guide makes no acceptance or submission-status claim.