A fast, rule-driven text linter, formatter, and transformer.

marque is a general-purpose rule engine for fast text processing — rules produce warnings, errors, fixes, and transformations, each with a confidence score the engine uses to decide what to auto-apply vs. surface as a suggestion. Built in the style of ruff: designed for perceptual instantaneity at any scale, operating on raw byte buffers with SIMD-accelerated scanning and an Aho-Corasick parser.

That's a fancy way of saying it's extremely fast. It can scan and fix a page faster than you can perceive the change.

The MVP ships a CAPCO/ISM classification-marking rule set targeting ODNI ISM-v2022-DEC (marque-ism and marque-capco crates). That's one application of the engine.

The roadmap expands into other U.S. Government control markings (CUI), foreign and multinational classification systems (e.g. NATO), and general-purpose text lint/transformation domains.

The next big feature upgrade will bring metadata curation, correction, extraction, and removal.

Design Philosophy

marque must be:

• exceptionally fast
• hardened and secure by design
• correct
• completely auditable
• helpful

Demo

Coming soon.

Install

cargo install marque

Or build from source:

git clone https://github.com/marquetools/marque
cd marque
cargo build --release -p marque

Quick start

# Lint a file (or stdin)
marque check report.txt
echo "(SERCET) draft content" | marque check -

# Preview fixes without writing
marque fix --dry-run report.txt

# Apply fixes in place; audit records stream to stderr as NDJSON
marque fix report.txt

Configuration lives in .marque.toml (project, committed) and .marque.local.toml (per-user identity, gitignored). See the CLI README for flags and exit codes, and marque-config for the full schema.

Workspace

marque-ism    ←── marque-core ────────────────────┐
marque-ism    ←── marque-rules ←── marque-capco ──┤
marque-scheme ←──────────────────  marque-capco ──┤
                                                  ↓
                                            marque-engine ←── marque-config
                                            ↑    ↑
                                   marque-wasm  marque-extract (non-WASM only)
                                            ↑
                                      marque-server
                                            ↑
                                       marque (CLI)

Read A ←── B as "B depends on A". marque-engine is the sole convergence point for the scanner chain (marque-core) and the rule chain (marque-capco). marque-scheme is the domain-neutral trait surface.

Crate	Role
marque	CLI binary — check, fix, metadata.
marque-core	Format-agnostic scanner + parser. SIMD via memchr, token matching via Aho-Corasick. WASM-safe.
marque-rules	Trait definitions only: Rule, Diagnostic, FixProposal, AppliedFix, Severity.
marque-engine	Pipeline orchestration. Engine (sync) + BatchEngine (async concurrent). Confidence gate + audit log live here.
marque-config	Layered config loading (CLI > env > .marque.local.toml > .marque.toml) with hard-fail validators.
marque-ism	ISM vocabulary types + generated CVE enums from ODNI schemas. Build-time codegen, no runtime I/O.
marque-capco	56 hand-written CAPCO rules (E001–E016, E020–E052, S001–S004, W002–W003, C001) consuming marque-ism predicates. Includes SAR (Special Access Required) validation per §H.5, structural SCI compartment + sub-compartment support per §A.6, per-SCI-system class ceilings + required-companion constraints per §H.4 (E042–E051, fix-and-warn), NODIS/EXDIS mutual-exclusion + NOFORN-required constraints per §H.9, REL TO list-grammar invariants per §H.8 (ordering E020 + no-duplicates E052), and the suggest-don't-fix channel (S004 rel-to-trigraph-suggest, issue #235).
marque-extract	Document text + metadata extraction. Stub — Kreuzberg integration pending.
marque-wasm	WASM target via wasm-pack. Byte-identical NDJSON output to the CLI.
marque-server	axum REST microservice wrapping marque-engine.

Domain scoping: marque-ism and marque-capco are the only crates that are ISM/CAPCO-specific. Everything else is general-purpose infrastructure designed to host additional rule sets (CUI, NATO, KFC recipe) without changes.

Status

MVP complete. Full lint → fix → audit pipeline for raw text with 39 CAPCO rules, including structural SAR validation per §H.5 and structural SCI compartment + sub-compartment validation per §A.6. CLI and WASM produce byte-identical NDJSON diagnostics. Configurable severity overrides, corrections map, and confidence thresholds. Batch processing via BatchEngine with concurrency control. Criterion benchmarks validate p95 ≤16ms on 10KB inputs.

Not yet built: marque-extract (extraction from a wide range of media), the metadata CLI subcommand, incremental LMDB cache (v0.2), server auth middleware.

License

This project is licensed under the Marque License 1.0 (LicenseRef-MarqueLicense-1.0). Copyright 2026, Knitli Inc.

Is it Open Source?

That depends on who you ask. The Open Source Initiative (OSI) and Free Software Foundation (FSF) both would say no, because their definitions don't consider any provision that limits uses as open source or libre ¹.

Marque is arguably open source under U.S. law².

However you classify it, the Marque License is what is known as "fair code" or "source available." It tries to balance a few competing interests:

• Source is right here in the open. An engineer can read, understand, and test it. I think that's a fundamental consumer right — you deserve to know what's under the hood and how meticulous I am about ensuring a high-quality product. You should also be able to freely conduct your own security testing, weighing the trade-off decisions I made against your own specific situation. That's nearly impossible with most enterprise software.

• You can freely use it for a lot of things. You can deploy Marque on an internal server, run the demo for your office, and much more. I built this because I knew the problem was solvable — the existing solutions just reflect what happens when nobody with real domain expertise is building the tools.

• You can't use Marque against me. I put a lot of time, care, and consideration into every detail of Marque. It's built on deep experience with marking systems. I let people use it for a lot of things for free — but I need to live, and I deserve some benefit from it too. I also don't want to see vendors who couldn't build this — and didn't — use it to undercut the people actually solving problems their customers have been waiting years for them to fix.

---

Footnotes

1. The definitions are very similar, but OSI uses the term 'open source' while the FSF uses 'libre'. ↩

2. Public Law 115-232 defines open source software as "software for which the human-readable source code is available for use, study, re-use, modification, enhancement, and re-distribution by the users of such software." Marque's source code is publicly available and meets this definition for internal use. Marque does not meet the Open Source Definition maintained by the Open Source Initiative or the Free Software Definition maintained by the Free Software Foundation, because it restricts commercial redistribution. ↩