What a source can tell us.
And what it can’t.
Every brief keeps the documented claim separate from the outcome someone might hope to get.
Start with the original.
We link to the document behind a claim and record when it was inspected. A source review can establish what the vendor documents. It cannot establish a performance improvement on another team’s work.
Five articles repeating one announcement are still one piece of underlying evidence. We count independent work separately.
The labels have a limited meaning.
- Source note
- A review of what a document says. This is the format of the current collection.
- Unverified
- The scoped outcome has not been independently established or reproduced in the brief. It does not mean the claim is false.
- Corroborated
- Independent evidence supports the stated claim. Its methods and limits should be included.
- Reproduced
- A result has been repeated under specified conditions. Those conditions travel with the result.
- Disputed / Inconclusive
- Evidence conflicts, or there is not enough to resolve the question. Neither should be hidden.
The first three notes include no independent study or reproduction. We do not assign numerical confidence scores.
Recommend a next step, not an answer for everyone.
Try means a limited experiment may be worthwhile. Watch means requirements or evidence are still missing. Skip means the tool does not currently fit the stated use.
A useful recommendation names the scope, permissions, acceptance criteria and stopping conditions. The briefs express these as practical next steps rather than product ratings.
Review the analysis too.
Claude-assisted claim extraction and source comparison are planned, not integrated into this site. Model output will not count as evidence. Published work still needs source checks and human review.
If a document changes or a note misses something, send a correction with the claim and a supporting source. Reviews are snapshots, not continuous monitoring.