The latest articles on DEV Community by US Directory (@usd_org). https://dev.to/usd_org https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4216061%2Fb74d4e49-02cf-451c-8fe4-5cbf545b9fb9.png https://dev.to/usd_org en US Directory Thu, 29 Oct 2026 21:01:57 +0000 https://dev.to/usd_org/designing-a-scoring-method-you-can-audit-notes-from-us-directory-2h0k https://dev.to/usd_org/designing-a-scoring-method-you-can-audit-notes-from-us-directory-2h0k <p>US Directory describes local service businesses using public records, and it publishes the method behind every listing. A few design choices make that method something a reader, or a developer, can audit instead of trust blindly.</p> <h2> Missing data is not a zero </h2> <p>When a fact cannot be checked, the factor stays unscored and its weight is redistributed to the factors that could be checked. A company is neither punished nor rewarded for a gap in the record. Positive-practice factors with nothing to read get a neutral value capped at 0.5, so silence never looks like excellence.</p> <h2> Evidence has a strength </h2> <p>Every fact carries the strength of its origin: 1.0 for an agency record, 0.8 for a file published on the company's site, 0.6 for a precise statement, 0.35 for a loose claim. Values that rest on site statements alone are capped at 0.6.</p> <h2> Additive contributions </h2> <p>The score is a weighted average over checked factors, and each factor's contribution is reported separately. Because contributions sum exactly to the score, a reader can see which factor moved it and by how much.</p> <h2> Score and confidence are separate numbers </h2> <p>Confidence measures how much of the picture was checked and how strong that evidence is, with a bonus for independent source families. A weak license match lowers confidence without changing the score.</p> <h2> Ranks must survive noise </h2> <p>Before a list shows rank numbers, the weights are randomly redrawn 500 times. If the leading places shift too much, the list falls back to tiers without numbers.</p> <h2> Versioned and machine-readable </h2> <p>Factors have stable identifiers; a factor whose meaning changes gets a new one. The full method is published as a machine-readable file in a public repository, alongside the human-readable text.</p> <p>Read the methodology at <a href="https://usd.org/methodology/" rel="noopener noreferrer">usd.org</a>.</p> <p><a href="https://usd.org/" rel="noopener noreferrer">US Directory</a></p> datascience opendata transparency algorithms