# Compare findings across sources

Build a research brief that preserves source differences instead of treating every result as the same kind of evidence.

## Define a bounded research brief

```
Use Purrkey to investigate [question] using [requested sources]. Discover and inspect the tools first. Keep the total cost under [budget]. Compare the evidence in a table with source URLs and dates. Mark disagreements, missing coverage and your own inferences.
```

## Keep each source accountable

1. Break the question into capabilities: search, retrieve a specific record, or inspect comments. Discover each needed capability separately.
2. Check each candidate's source coverage, supported dates and pricing unit. Stop if a required source is unavailable rather than silently substituting another.
3. Execute bounded Runs and retrieve their original results. Pass validated output to the next step in your own client; Purrkey does not accept a multi-step workflow definition.
4. Compare observations using consistent time periods and measures. A discussion count is not the same thing as adoption or product quality.

## Resume without paying again

Save the request files and Run IDs for every step. If your research session stops halfway through, retrieve the available original results before deciding whether additional calls are needed.

```
purrkey get-run --id '<saved runId>' --wait 30s --output source-result.json
```

- [Runs and recovery](/docs/guides/runs): Resume uncertain or interrupted work using the original request.