Evidence first. Certainty earned.

1. Capture with permission

A replay transcript, live-stream audio window, uploaded clip, microphone segment, or pasted statement becomes text. The visible transcript helps you catch obvious transcription errors before trusting a finding. Live audio is processed in short windows, so the timestamp identifies the transcript segment rather than guaranteeing word-level timing.

Notes stay traceable

Notes Studio keeps your written notes separate from the transcript. When requested, the model creates an overview, key points, decisions, action items, open questions, and claims worth checking. It is instructed not to invent owners, deadlines, decisions, or timestamps. A claim with a reliable transcript marker can return to that moment; unmarked text is never assigned an estimated time.

2. Separate “said it” from “is it true?”

The exact transcript quote is checked separately from the truth of the underlying proposition. Except when authorship is itself disputed, a result may not turn a statement into “the speaker said X” and call it supported merely because X appears in the transcript. The verdict applies to the substance of X.

3. Interpret fairly, then test hard

The checker states the strongest reasonable, good-faith, testable meaning of the speaker’s words while preserving important qualifiers. If one sentence bundles claims that could receive different findings, it separates them when practical. Opinions, predictions, jokes, rhetorical questions, and value judgments are not forced into factual verdicts.

4. Search both directions

Research starts with one focused web search and may use up to three additional searches to locate controlling primary evidence, test the strongest counterevidence, resolve a key date or definition, or examine a credible conflict. It stops when the direct evidence is sufficient. Finding a page that repeats the assertion is not enough.

5. Prefer the closest evidence

Official statistics, legislation, court records, regulatory documents, original studies, filings, and direct institutional records come first. Reputable secondary reporting is used when a primary source is unavailable or the interpretation itself is disputed. Material legal, scientific, statistical, and government claims should include the closest available primary authority or receive lower confidence.

6. Treat legal claims as legal claims

For legal and regulatory statements, the analysis separates the governing text, binding holdings, questions a court expressly left open, lower-court decisions, agency rules or guidance, and policy arguments. It identifies the jurisdiction and current status as of the research date; a dissent, agency position, purpose argument, or predicted consequence is not presented as controlling law.

7. Match the claim precisely

A source should match the claim's geography, time period, population, definition, and comparison. Every detailed result exposes evidence supporting the claim, evidence challenging or limiting it, a direct comparison, and the most important uncertainty. A general source that does not establish the actual proposition should not support a confident finding.

8. Hold public power to a higher accountability standard

Official House and Senate ethics materials describe public office as a public trust. When the evidence saved with a receipt establishes an accurate version and does not support the materially stronger version stated by a president, legislator, candidate, or other public official, the checker calls that an avoidable factual failure and explains how it can distort public judgment. Professional staff or access to government records increases the reasonable expectation of verification, but office alone is not proof of subjective knowledge. A briefing, warning, prior correction, acknowledgment, or repetition after notice can support the stronger conclusion that a false statement was knowing or reckless. This standard is applied identically regardless of party or ideology. House ethical standards · Senate public-trust standard

Finding labels

Supported

Direct evidence substantially supports the statement as expressed.

Contradicted

Direct evidence substantially conflicts with the statement.

Misleading

Some part may be grounded, but framing or omitted context materially changes the impression.

Needs context

The statement cannot be responsibly judged without an important definition, date, comparison, or surrounding fact.

Unverified

Available credible evidence is insufficient, too new, or materially conflicting.

Not checkable

The statement is an opinion, value judgment, prediction, or otherwise not objectively verifiable.

Known limitations

Fast checks trade some depth for speed. Sources can be wrong. Search ranking can omit relevant evidence. A short transcript can remove context, and a transcription mistake can change the claim. Live replay depends on the platform retaining DVR history; when DVR is unavailable, the receipt can return only to the transcript segment. Breaking news may not yet have reliable primary records. Confidence labels describe the evidence available to this run, not an objective probability that the statement is true.

Corrections standard

A launch-ready corrections process should preserve the claim, finding, source set, reason for correction, and time of change without retaining the underlying private recording. That public correction workflow is a release gate, not yet an active feature.