Methodology
Last updated 2026-07-06 · Canonical source: avieros.com/methodology (v1.0.0, live and test-enforced)
Avieros Search and Candidate Watch run on a small set of honesty rules. This page states them plainly; the full methodology — with the exact ranking weights, trust tiers, data sources, and a versioned changelog — lives on avieros.com, where an automated test fails our build if the published page and the running system ever disagree.
The canonical methodology page is wired to the ranking code itself: the numbers on it are read from the same code that does the ranking. Reproducing those numbers here would create a second copy that could drift. One canonical page, one source of truth — this summary links to it everywhere a number is involved.
The rules, in plain language
1. Primary sources beat aggregates — always
Where a real, primary record exists, Avieros fetches it instead of estimating from summaries. Civic and money data — who represents you, what they sponsored, who funds them — come directly from official government sources (the FEC and Congress.gov), which publish their records in the public domain. No number is estimated where a real one can be fetched, and no gap is smoothed over.
2. No source, no render
A civic claim without a primary-source link is never published. Where data is missing, the product says so — visibly — instead of filling the hole with a guess. You will see “not available” in Avieros products; you will not see a fabricated value.
3. Provider order is never the final order
General web results are discovered through an external provider, but our ranking contract forbids using the provider’s ordering as the answer. Every result is re-scored by our own ranking kernel, with published weights — see Transparency for how that works.
4. Independence: no ads, no pay-for-ranking, no lobbyist data
Nobody can pay Avieros to rank higher, and Candidate Watch does not ingest lobbyist-supplied narratives. Ranking positions and candidate scores are not for sale. This is stated, in those words, on the canonical methodology page.
5. Limitations are published, not hidden
The methodology page carries a standing “Known limitations” section naming what is not built yet — for example, that roll-call floor votes are not yet ingested and that Candidate Watch currently covers a seeded set of officials rather than every officeholder. When a limitation is closed, the changelog records it.
How these rules are enforced
- A build-failing sync test. The published methodology and the running ranking code are compared by an automated test. If they disagree, the build fails — the page cannot silently rot.
- A versioned changelog. Every change to ranking weights, trust tiers, or scoring rules must ship with a changelog entry in the same code change. Details on Transparency.
- An honesty gate in CI. Our release pipeline includes an automated check that rejects code reporting fabricated success — a thing may only be reported “connected”, “sent”, or “done” when it actually happened. See AI Safety.
Report an error
If you find a factual error in anything Avieros publishes, write to corrections@avieros.com. Corrections are handled against the same rule set: the fix must cite a primary source.