What Evidence Lab does

Engagement boundaries, working outcomes, and the researcher's role.

Result

You can decide whether Evidence Lab fits your need and can name the research workflow that the engagement will configure. The result is a personal, inspectable environment for a bounded set of tasks, not a universal autonomous researcher.

What the engagement covers

Evidence Lab maps one current workflow, agrees observable outcomes, selects a platform and a limited set of components, configures the project, works through a representative task, verifies the result, and checks that the researcher can repeat the process independently.

The public documentation explains that route. The private consultation workspace holds case-specific evidence, access details, diagnostics, and decisions; none of that belongs in this repository.

Actions

Bring a recent real case whose expected result and quality problems you understand.
Describe the trigger, inputs, human decisions, output artifact, review process, and known failure points.
Agree one to three outcomes that can be demonstrated on that case.
Choose Codex or Claude Code as the first working layer after checking account and data restrictions.
Add only the instructions, skills, plugins, and connections required by the accepted route.
Complete the task, retain evidence, test recovery from a representative failure, and repeat the cycle independently.

Evidence Lab is responsible for diagnosis, configuration, instruction, and a verifiable working process. The researcher remains responsible for the scientific question, source eligibility, domain judgments, interpretation, and final conclusions. Organisational data owners remain responsible for access policy and permitted use.

Check

Complete this sentence with observable nouns and verbs: “When _ happens, the environment uses _ to produce _; I accept it only after _.” A second person should be able to identify the input, output, and acceptance gate from your answer.

If it fails

If the sentence covers an entire research project, contains no inspectable output, or ends with “the AI checks it,” narrow it to one repeatable scenario and name the human or mechanical check.

Next: How the engagement works.