# Evidence record

Date:

## Question and prediction

- Question:
- Prediction before searching or running:

## Source

- Work and URL:
- Version or file hash:

## Method

- Commands, notebook, or tool, with its version:
- Criteria or settings you chose, and why:

## Result

- Unit counted, and the rows or passages included:
- Result:

## Passages checked

| Location | Expression | Labels | Evidence |
|---|---|---|---|
|  |  |  |  |
|  |  |  |  |

## Limits and open judgments

- Missed or misleading example:
- Unresolved rows and their reasons:
- What the result cannot establish:

## To repeat this

- Steps another person would follow:
- Next change to try:

## When comparing methods

- Input: text version, passage boundaries, transformations or chunking, language
- Instructions and settings: rules, labels, model identifier and revision, settings, LLM instructions
- Results and errors: agreement with the shared reference annotations; missed, extra, boundary, and label errors
- Practical costs: hardware, runtime, first download compared with later runs, any charges
- Repeat runs: procedure, date, changed conditions, and differences on a repeat run