B
HomeLibrariesCode AISearchProjectsBookmarksAddress
FeedbackHelp Desk
Libraries
Code LibraryDiagramsAssembliesProductsCalculatorsChecklistsCSI MasterFormat SpecificationsPermitsAddress ReportDOT SpecsModel CodesTransportationRailroadFire SafetyAccessibilityStructuralTransit SpecsUSACE
Other Jurisdictions
Transportation
MUTCDFHWA FP-14
Accessibility
ADA StandardsPROWAG
PricingLog in

Open

The law should be readable — by people, and by the machines they ask.

Building codes are law. Once a jurisdiction adopts a code, that text governs what you may build, what you must fix, and whether your project is legal. You are required to comply with it. It follows that you should be able to read it.

That principle is not controversial anymore. In April 2026 the Third Circuit held that publishing standards that have been incorporated into law is likely fair use. Courts have been moving this direction for years.

But a rule that says the public may read the law says nothing about how. And in 2026 a growing share of people don’t read the law directly — they ask an assistant, and the assistant reads it for them. Which means the practical question has quietly changed. It is no longer “is the code online?” It is “can the tools people actually use reach it?”

Most code sites have answered that question by saying no.

What we do differently

We let AI assistants read us.

Our robots.txt explicitly welcomes GPTBot, ClaudeBot, PerplexityBot and the rest, across the whole site except account and billing routes. We publish an llms.txt describing what our corpus covers and how we’d like to be cited. Every code section, fee schedule and DOT spec lives at a stable, server-rendered URL that returns its content to an anonymous request — no login, no JavaScript requirement, no cloaking.

This is a deliberate choice with an obvious cost: a person who gets their answer from an assistant may never see our site. We think that’s the correct trade. If someone needs to know the minimum egress width and an assistant tells them, citing the section, that is the system working. We would rather be the source that gets cited than the site that got visited.

The leading alternative takes the opposite view. As of 19 August 2026, up.codes’ robots.txtdisallows all crawlers by default and names GPTBot, ChatGPT-User, ClaudeBot, Claude-Web, anthropic-ai, Google-Extended, PerplexityBot, Amazonbot and Bytespider for explicit exclusion; Googlebot, Bingbot and DuckDuckBot are permitted. We’re not claiming that’s improper — a business with licensed content has real reasons to fence it. We’re pointing out that the fence exists, and that when you ask an assistant a code question, the answer is being assembled from somewhere else.

We check the text against the official source.

Every permit rule and DOT spec traces to a source document, and the link is on the page. Our AI answers retrieve from verified content and cite the section they came from, inline and numbered, so the citation is checkable rather than decorative. We are still extending “last verified” stamps across the code corpus — where a stamp isn’t there yet, it isn’t there, and we won’t imply otherwise. The methodology page says exactly what the word means in each case, and what the retrieval gate refuses to serve.

We publish what we got wrong.

Our corrections log is public and reverse-chronological. It currently records 5 corrections, the most recent on August 17, 2026 — among them the placeholder rows we had published as code sections, which inflated our own coverage claims until we caught it; the DOT spec sections whose summaries had been generated rather than extracted; and wrong ampacity values in a wire-size calculator.

Those are bad errors. Publishing them costs us something. We do it because a code reference that never admits error is asking for a kind of trust it hasn’t earned, and because the alternative — quietly editing and moving on — is how a reference source decays without anyone noticing. If you find something wrong, tell us and it goes on that page.

What this is not

This is not a claim to be more accurate than everyone else. We are a small operation with a corpus we are still extending, and the corrections log is the evidence for both halves of that sentence. It is a claim about method: check against the source, show the source, timestamp the check, publish the misses, and don’t lock the result behind a robot fence.

Nothing on this site is legal advice, and no code reference — ours included — substitutes for the authority having jurisdiction. Verify against the adopted text before you build.

How we verify →What we got wrong →How we compare →