# Atlas Field Notebook: two-hour comparison

## The question

How many source-backed, nonduplicate candidates can a focused human researcher prepare for governed intake in two hours?

Raw row count is not the result. The useful result is **field-ready candidates per hour** after the full-catalog identity preflight.

## Before the clock starts

1. Choose one narrow subject, place, or source family.
2. Write one inclusion rule that can stay unchanged for the entire session.
3. Download the CSV template or work directly in the Field Notebook.
4. Do not publish, contact anyone, or reinterpret a weak clue as verified evidence.

Example subject: `Independent guitar builders in Ontario`

Example rule: `Include only active guitar-focused businesses supported by an official website or a named discovery source.`

## Capture contract

- `name` — required.
- `website` — the candidate's official site when known.
- `source_url` — the exact page that supports the lead or finding.
- `country`, `city`, `category` — use the most specific facts the source supports.
- `notes` — one concise sentence explaining why the row belongs or what remains uncertain.
- Leave enrichment/task columns untouched unless the row came from an Atlas readiness packet.

One strong source is better than several search-result links. Record uncertainty; do not repair it from memory.

## Two-hour run

1. Enter the subject and fixed inclusion rule in the Field Notebook.
2. Start a new session immediately before research.
3. Work for 120 minutes.
4. Keep a row even when incomplete; its status makes the research gap measurable.
5. Run **Full-catalog preflight** after the final row is captured.
6. Export CSV, manifest, and benchmark JSON before starting another session.

## Compare these metrics

- captured per hour;
- source-backed per hour;
- field-ready per hour;
- source coverage percentage;
- field-ready yield percentage;
- possible-duplicate percentage;
- full-catalog identity scope and records scanned.

The benchmark is comparable only when the subject and inclusion rule are present, the 120-minute target is reached, at least one row exists, and the full-catalog preflight covers every row.

## Safety boundary

The Field Notebook stores the session in the current browser and exports local files. It performs zero database writes, zero publication writes, zero outreach, and zero model calls. Exported rows are research material, not published Atlas facts. Governed intake must still retain provenance, resolve canonical identity, review conflicts, and provide rollback before publication.