How RepairCostly is built
Updated
The rule about numbers
No number on this site is typed from memory or estimated. Every figure, whether it is a dollar amount, a number of days, a distance or a threshold, comes from a public page that we read and saved with its address and the date. Figures live in typed data files, not in the page text, and each one names its source. A test fails the build if a figure's source is unknown or if the words behind it are not in the saved page.
When we cannot find a public source for a figure, the page does not make one up. It says that no source was found, or leaves the page unpublished. That is why this site has a short list of cost tables: repair prices vary by region, season and scope, and public, dateable price sources are scarce.
What counts as a source
Government and regulator pages first: the Bureau of Labor Statistics for wages, the EPA and CDC for water, mold and safety, the Federal Trade Commission for contractor advice, state insurance departments and the National Association of Insurance Commissioners for how claims work. Where a source is not government text, we paraphrase briefly and name it. We do not copy price tables from commercial cost sites.
State rules differ. When a page uses one state's guide, such as Texas, California or North Dakota, it says which state. A deadline in one state is not a deadline in yours.
How pages are tested
Checklists, question paths and letters are data, not hand-built pages. Tests check that every question path can be reached and ends in an urgency level and at least one next step, that every letter field is used, that every checklist item has a unique id, and that every cost row has a source and a check date and runs low to high. Other tests check that page text contains no loose numbers and that every source on a page is one we saved.
The saved pages
The pages we read are saved under docs/data in the project, with their address and the date. Most are saved as the original page exactly as fetched, HTML or PDF, with a checksum recorded in a manifest, and the tests read figures from that raw text (for a PDF, from its text as extracted by a standard tool, which is saved beside it). Federal wage data is saved as an excerpt of the Bureau of Labor Statistics file with its address and checksum. A few agencies refuse automated downloads, so for those pages the saved file holds a text extraction made by a reading tool, labeled that way, until a raw copy can be saved. A person checking a figure should open the source link on the page.
Updates
Public guidance moves, so every page has a re-check date. Wage data is re-checked yearly and guidance pages twice a year. When something changes, the change is logged on the changes page with the date.