Methodology

This page describes how the tools on this site are built, what they are allowed to do, and what has to be true before one of them is published. It is transcribed from the engineering rules this repository runs on, not written for marketing.

The four gates

A tool ships here as a free tool only if all four of these hold. If any one fails, it is not a free tool.

1. No language model in the answer. Every result is deterministic arithmetic or deterministic logic. No tool asks a model what the answer is.
2. No call to another service while you use it. Reference data is baked into the site when it is built, so using a tool sends nothing to a third party and depends on nothing being up.
3. Published reference data only. The tools read published standards and public reference data. Licensed content is not ingested from any source.
4. A reference only scope note on every tool. Each tool states plainly that it informs, and that it does not advise, certify, or replace the primary source.

How a tool goes live

Publishing a tool takes two separate switches, and both are the founder's to throw.

The first is the tool's row in the catalog, which carries a status of either live or soon. The second is an independent per tool flag: a tool listed there as false returns a 404 whatever the catalog says. A tool is only reachable when both agree, so one switch flipped by accident publishes nothing.

Before either switch moves, every regulatory number the tool renders is verified against its primary source, meaning the CFR, EUR-Lex, or the standard itself. Secondary write ups and recalled figures are not acceptable evidence for a number on a page here. That verification is performed by a person, not by an agent, and the release is not made without it.

How data is versioned

The reference datasets live as files in the repository, one committed copy per dataset, and they are the source of truth. Application code never reads them at runtime: each one is embedded into a code module when the site is built, so a change to a dataset is not live until that embed is regenerated.

Each dataset carries its own provenance record naming the publisher, the instrument, the edition or release marker the publisher itself uses, the date it was retrieved, and how many entries it holds. Those records are what the data vintage line on each tool page and the table on Sources are built from, so the same fact has exactly one home and cannot say two things in two places.

Identifier columns are normalised for whitespace and format. Name columns are kept verbatim to the source, including typography a reader cannot type, and the matching is done in the code rather than by editing the data.

What reference only means

Every tool on this site carries the same line: reference only, not legal or regulatory advice, verify against the primary source before you act. Here is what that means in practice.

Each tool works from published reference data baked into the site when it is built. No tool calls out to another service while you use it, and none of them ask a language model what the answer is. The same inputs give the same result every time.

Every result names the list or provision it checked, and the scope note tells you what the tool did not look at. A clear result on one list is not a clearance opinion. Read the source before you act on it.

Corrections