The Rule Was Never the Hard Part

Law's oldest machine, and the new reader it wasn't built for

by Sam Rogers
9 min read
analysis
governance
legal
regulated-industries
strategy
The Rule Was Never the Hard Part

We think of law as rules. It is the thing on the page, the statute, the code section, the regulation with its numbered paragraphs. Yes, and a rule has never decided a single case.

Because rules do not apply themselves. Someone has to say what this rule means for a specific set of facts. That act is where law actually happens, not the rule. It's also where law has quietly stowed its disagreements for some time now.

Almost no one outside the legal profession thinks about the interpretation layer, which I think is the most interesting machine humans ever built for dealing with the fact that reasonable parties do disagree.

The quietest institution

This wouldn't make a good courtroom drama, but it's dramatically important.

A company writes to a regulator and asks for guidance with the fancy version of "does this rule apply to us, this way, on these specific facts?"

Then the regulator writes back. This could be called a no-action letter or a private letter ruling or an advisory opinion or something similar. The cleanest version is the Joint Interpretation Agreement, where two parties write down together how a rule applies to a specific situation and both put their names to it.

Whatever it's called, this is what happened:

  1. A question was posed
  2. A position was taken
  3. An authority responded
  4. A resolution was recorded, with the facts it depended on and the names of who decided, and then preserved so that others could rely on it.

That is an arbitrated decision record. Law has been producing them at scale, in every jurisdiction, for a long, long time. Longer than any of us has been alive.

And it's not just for regulators. HOAs keep a decision log. Co-ops record their governance rulings. A flying club writes down who agreed to what about the shared aircraft. The same machine runs at every scale of human coordination. People don't do this because we like it, most of us find it painfully boring and really would rather not. Yet we do it because the same problem exists at every scale: the rule is never enough by itself, and somebody has to keep the record of how we agreed that it applies. Otherwise next year no one will remember and the argument starts up all over again. Which is universally much worse than "boring."

Law even protects the losing side. A dissenting opinion doesn't just get deleted, it's published right next to the majority, filed where the future can read it, because a decision whose objections have vanished is a decision no one can audit. Also, oftentimes it's the dissent that the next generation builds on. The rule is the conceptual structure, and the interpretation is how it actually lands. And the interpretation, with all its facts and authorship and preserved disagreement, is what the law remembers.

The memory we can't read

One problem the legal system has never solved for itself: memory is real, yet almost impossible to query directly.

The all-important interpretations are scattered about. A no-action letter is in one agency's archive, a private ruling in another's, a state advisory opinion lives as a PDF on a government server, an agreement sits in a filing cabinet. And all of these were inevitably written in incompatible formats by authorities that never coordinated, and there is no way to lay them out side by side and ask the obvious question: how have different authorities interpreted this same obligation?

The institution preserved the record and left it illegible. It has the memory, but there's no way to search it.

When the only readers were lawyers & paralegals getting paid to hunt around, this gap didn't matter so much. But it matters enormously now, because a new kind of participant is starting to use these rules at machine speed, and it can't hunt through any filing cabinets.

Making the record legible

This is the work at the center of the PAICE legal graph, and PubLedge is the piece that handles the interpretation layer directly.

PubLedge is an open protocol for recording fact-specific interpretations in a portable, comparable form. Each record is plain markdown with structured frontmatter, hash-pinned with a SHA-256 fingerprint so it can't be quietly altered after the fact. It's also bound to a shared upper ontology so that records issued by authorities who never spoke to each other can finally be queried all together.

It's not a regulator, it issues no rulings. It's not a law firm, and it gives no legal advice. It's not a blockchain, the integrity layer is a plain hash over a plain file. What it is is a way to write the record down so the record can be read, by a person or by an AI agent, across authorities, without scraping. It's deliberately unglamorous.

Critically, PubLedge does not stand alone, it's part of a larger whole. ObligationFirst supplies the shared schema that lets laws, agreements, and cases be represented as the same kind of normative content. EveryAILaw.com holds the corpus of the rules themselves. AI Incident Law holds the record of what happened when rules were broken, the public incidents and the legal and regulatory fallout. Put together, these four layers are a machine-readable map of what the law requires, how it has been interpreted, and what it cost when someone got it wrong. And the whole thing is open by charter, code and content licensed for reuse, because a map of the law that one company owns is not a public record. It's a private toll road with the law painted on it, and we've got enough of that already. (Yes, I'm looking at you Westlaw.)

The hash-pinning is the linchpin, and it's the same instinct that runs through everything I build. It preserves not just that an interpretation existed but its exact content. So the argument can't be smoothed or rewritten later to fit a more convenient story. It's what makes the integrity of the record, rather than just its survival.

Why a machine needs the interpretation layer

Give an AI system a statute and it will run with it. Confidently. Quickly! But correctly? Well, it will tell you what the rule says alright. What it will not know, unless someone made the layer legible, is how that rule actually lands: the no-action letter that carved out the exact case in front of it, the advisory opinion that read an ambiguous term the narrow way, the interpretation that three authorities share and a fourth pointedly does not.

An agent that can query the statute but not the interpretations will apply the rule fluently and wrongly. But don't worry, it'll sound completely 100% sure while it does. That's the synthetic-trust failure that shows up everywhere that AI touches high-stakes work, and it's arriving now at regulatory scale where the cost of confident misapplication is measured in enforcement actions and real harm. The defense is not a smarter model. It is infrastructure: make the interpretation layer, the accumulated human record of how the rules actually apply and where authorities disagree, available to the machine before it acts.

We already built this once

The reassuring thing that's worth ending on is that we did not have to invent disagreement-preservation for the age of AI. Humanity has been running the largest and longest instance of it in recorded history. Law already figured out that you preserve the interpretation and not just the rule, that you name who decided and on what facts, that you keep the dissent where the future can find it, and that a decision you can rely on is one whose reasoning survived.

What is new is not the idea. It's the inhuman reader. The next participant in the system doesn't need to spend a lifetime in the archive, it reads a lifetime's worth of work in a matter of seconds. So the work now is to take the oldest disagreement-preservation machine we have and make it legible to the newest kind of mind that will have to operate inside it.

  1. Keep the interpretation on the record.
  2. Keep it queryable.
  3. Keep a human authority at the point of decision.

That's the same short list I'd write for any decision worth trusting, human or machine. The law just got there first, and at a scale nothing else has matched.

The rule was never the hard part. The hard part is the memory of how we decided it applies, and who disagreed, kept in a form the future can actually read. Protect that and you can build on it. Lose it and all we're left with are rules that nobody can follow and no record of why they were ever important.

Want to understand your own readiness profile? Take the PAICE assessment to discover your strengths and opportunities.

Get Involved:

📖 The PAICE Legal Graph:

📖 Trust and Infrastructure:

Curious but short on time?

Take the 3-minute PAICE Pulse — a quick confidence check that maps how you see your own AI collaboration posture. No login required.