Methods & Corrections
Salt Tab is factual, evidence-first, transparent, auditable, and non-prescriptive — not “unbiased.” We disclose editorial choices rather than claim none exist. Every label below results from stored evidence and rules, never from an AI judgment alone.
Demonstration data. Everything currently published on this site is clearly fictional seed data (Riverbend, Cedar County, and the named people do not exist). It exists to exercise the evidence system and must not be read as real political information.
Claim states
Verified
Supported by sufficient identifiable evidence.
Direct statement
Accurately attributed to a speaker, but the truth of the statement is not implied.
Corroborated
Supported by at least two genuinely independent evidence chains.
Single source
Rests on one identifiable evidence chain.
Disputed
Credible evidence conflicts.
Unverified
Reported but not adequately substantiated.
Developing
Material facts may still change.
Incorrect
Contradicted by decisive available evidence; an explanation is required.
Outdated
Previously accurate but superseded.
Unknown
Evidence is insufficient to determine the answer.
Source & evidence labels
Primary source
An original record, recording, dataset, filing, or document.
Secondary source
Synthesis or repetition of other reporting or primary evidence.
Direct quote
A speaker's exact words, preserved with surrounding context.
Video verified
Matched to original video with a timestamp.
Audio verified
Matched to original audio with a timestamp.
Transcript verified
Matched to an official or independently verified transcript, with verifier attribution.
Independently corroborated
Confirmed by at least two genuinely independent evidence chains.
Circular reporting
Many outlets repeat the claim, but they derive from a small number of original reports.
Single source
Rests on one identifiable evidence chain.
Anonymous source
Rests on one or more sources whose identity is not disclosed.
Disputed
Credible evidence conflicts.
Unverified
Reported but not adequately substantiated.
Source removed
The original source is no longer available; the research record is retained.
Corrected
A material change was made; prior states remain in the history.
Developing
Material facts may still change.
Source classification
Credibility is evaluated at the claim/evidence level, not assigned permanently to a media brand.
Primary
Original government records, legislation, filings, datasets, recordings, or direct observation.
Direct reporting
Independently gathered reporting, interviews, or observation.
Secondary reporting
Synthesis or repetition of other reporting or primary evidence.
Commentary / discovery
Opinion, social posts, or aggregators — useful for discovery, insufficient alone for verification.
How candidate comparisons work
Comparisons are structured, sourced records — never scorecards. Salt Tab publishes no ratings, endorsements, or predictions for any candidate.
- One issue set per race, fixed before positions are gathered, and applied identically to every candidate.
- Every cell is documented or explicitly empty — a sourced position, or “Not yet documented.” Positions are never inferred.
- Neutral candidate order — ballot order when known, otherwise alphabetical by surname. Order is not a ranking.
- All US regions are configuration (all 50 states + DC seeded); the next November general election is prioritized, nearest race first.
Full method: docs/comparison-methodology.md. Browse races at /elections.
Corrections policy
When we get something wrong, we fix it in public. Every material correction is listed on the page it affects and on the homepage, with what it said before, what it says now, and why. Each correction also creates a new version of the page’s research record, so the earlier version stays on file.
How we choose what leads
Items are placed by stored editorial reasons — important, upcoming, developing, viral, local — never by clicks or traffic. The reason is shown on each item as a badge.
Proof of research record
At publication, each page produces an immutable manifest — its claims, evidence, sources, quotes, provenance, and known unknowns — and a SHA-256 hash of a canonical serialization. A correction creates a new version chained to the previous hash; history is never overwritten. This is a notarization layer today and is designed for future public anchoring.