ν Nomos
§ Research / Prior art

We're not the first to think about this.

The rules-as-code field is fifty years old. Every project below taught us something; several have active teams we hope to collaborate with. Nomos is best understood as a layer built on their shoulders — not a greenfield replacement.

← back to the thesis

§ 01

Languages, systems, and research programs.

Catala

Inria · 2020–present

active

Default logic as a first-class feature, mirroring how legislation structures exceptions. Compiles annotated legislation to OCaml/Python. Lawyer-readable PDFs.

take

The default-logic design (Lawsky 2017); the article-style syntax; the provenance discipline; the possible reuse of its IR.

leave

OCaml-first toolchain; no LLM integration; academic DX; heavy specification burden.

OpenFisca

France, Etalab · 2011–present

active

Tax and benefit law modeled as typed Python. Production-deployed by multiple governments (FR, UK, NZ, Tunisia, Senegal).

take

Proof that rules-as-code works at government scale. Tax-rule modeling patterns. Parameter vs. variable distinction.

leave

Python-only; imperative; no temporal typing; no LLM bridge.

Blawx

Lexpedite / Jason Morris · 2020–present

active

Visual rules-as-code for lawyers over s(CASP) — a goal-directed ASP engine. Generates natural-language explanations for every answer.

take

Explanation-first UX. Lawyer-as-primary-user framing. The CodeX Prize ecosystem.

leave

Visual/Blockly interface for us — not where serious adoption lives. Prolog runtime dependency.

Logical English

Kowalski / Imperial · 2020–present

active

Controlled natural language compiles to Prolog. Lawyers write near-English prose that an engine can reason over.

take

The controlled-natural-language idea as a possible future frontend. Kowalski's framing of law-as-logic.

leave

Small tooling; research-only; parsing ambiguities that prose brings.

s(CASP)

various · research

academic

Goal-directed Answer Set Programming with automatic justification traces. Blawx uses it as its engine.

take

Justification-tracking as a runtime property. Possible backend option for Nomos later.

leave

SWI-Prolog dependency; niche community; poor browser story.

LegalRuleML

OASIS · 2013–present

standard

XML interchange format for legal rules. Interop between rule engines.

take

A possible interop export target. The abstract model of rule + exception + defeasibility.

leave

Not a programming language. Verbose. Not what humans write directly.

Accord Project / Ergo

Linux Foundation · 2018–~2022

dormant

Smart legal contracts. Cicero templates + Ergo DSL + Concerto data models.

take

Concerto (still alive) as a type-system base. Contract-specific vocabulary.

leave

Ergo itself is largely dormant. Blockchain framing. Template brittleness.

L4

Singapore Management University · 2019–present

research

Haskell-based legal DSL with formal verification and visualization.

take

Multi-target compilation (LP, English, diagrams). Attention to formal semantics.

leave

Small community; Haskell barrier; research-only.

Symboleo

U. of Toronto · 2020–present

research

Formal contract specification language with runtime monitoring.

take

Event-based semantics for contract performance. Obligation/right/permission algebra.

leave

Narrow contract focus; research prototype.

Kowalski et al. (1981)

British Nationality Act · 1981

historical

Landmark encoding of an entire statute in Prolog. Awarded the CodeX Prize in 2021.

take

The foundational proof that statutes are computable. A reminder that the logic was never the hard part.

leave

Purely prose-out-of-band; all facts had to be hand-entered as Prolog terms.


§ 02

Reusable OSS — the layer below us.

Everything below is permissively-licensed, battle-tested, and either already integrated or scheduled. Reusing these is how Nomos ships as a side project instead of a PhD.

Eyecite

US legal citation extractor. ~3k GitHub stars. BSD.

→ Wrap as Nomos's US citation resolver.

CourtListener API

Authoritative US case-law archive. Free Law Project.

→ Resolve case numbers → full opinions in authority references.

Blackstone

spaCy NER pipeline for UK legal text.

→ UK citation resolver for the UK jurisdiction pack.

Concerto

Accord Project's type-modeling DSL. Still active, GSoC 2026.

→ Type-system base, extended with temporal + defeasible annotations.

Akoma Ntoso / LegalDocML

OASIS XML standard for legislation worldwide.

→ Import statutes → emit Nomos rule skeletons.

CUAD

Contract Understanding Atticus Dataset. 13k+ clause labels.

→ Evaluate extract<Clause> — publish numbers in the README.

MAUD

Merger Agreement Understanding Dataset. 47k labels.

→ Evaluate extract<DealTerm>.

ACORD

Atticus Clause Retrieval Dataset. 126k query-clause pairs.

→ Evaluate retrieval-based extract operations.

OpenRouter

LLM gateway. Used for extract<T> today.

→ Currently integrated. See packages/llm.

Pile of Law · MultiLegalPile · LeXFiles

Pretraining corpora (256GB → 689GB).

→ Only if we ever need to fine-tune a Nomos-specialized extractor.

CommonAccord

Legal-docs-as-linked-data. Prior art on structured markup.

→ Cite as inspiration; possible import/export target.

Docassemble

Guided legal interviews. ~1.5k stars.

→ Output target — compile a Nomos program to an interactive interview.

Restate

Durable execution runtime (same choice as Weft).

→ Back human-in-loop gates that must survive restarts.


§ 03

Credit discipline.

Every paper or project above is cited in our README, our research pages, and — eventually — in any academic write-up. "I didn't know about them" is not a defensible position after fifty years of work.

If we borrow an idea, we name the source in a code comment. If we reuse a library, it's a declared dependency with the upstream license. If you see something unattributed, open an issue.