Forensic-grade authorship verification from the terminal. Build a per-author profile by fingerprinting their writing, then ask of any text: was this written by that author? — and get a calibrated probability, not a vibe.

Pure Rust, single static binary, no model and no network required. Built on the agent-cli-framework: JSON envelopes, semantic exit codes, and a machine-readable agent-info manifest, so agents and humans use it the same way.

LLM-era writing tools need a trustworthy ruler. A hand-weighted similarity score with no standardization and no calibration can't tell you whether text is "in an author's voice" — optimizing against it just games the metric. This tool implements the method forensic and academic stylometry actually use, and reports a calibrated verdict you can defend.

It is also the independent judge for a sibling rewriting tool: kept separate on purpose, run only on held-out text, never optimized against.

cargo install --git https://github.com/paperfoot/stylometry-cli # from source

# Fingerprint each author from a folder of .md/.txt (a file also works)
stylometry profile build adams --corpus ./adams-nonfiction/
stylometry profile build wodehouse --corpus ./wodehouse/ # others double as background
stylometry profile build jerome --corpus ./jerome/
# Fit the verifier: delta -> P(same author), with AUC against the other profiles
stylometry calibrate adams
# Verdict on a new text
stylometry compare suspect.txt --profile adams
stylometry profile list

compare returns Cosine Delta, Classic Burrows Delta, the nearest profile, a background-rank score (a simple rank fraction, not full General Imposters), and (once calibrated) P(same author) with a same/different verdict. Every command takes --json; run stylometry agent-info for the full manifest.

1. Fingerprint. Tokenize a corpus into ~1,500-word chunks; count the most frequent words and character trigrams (the two strongest authorship signals in the PAN literature).
2. Standardize (Burrows Delta). z-score every feature against the combined reference of all profiles — so a frequency counts as "unusual relative to writers in general", which is the whole point of Delta over raw cosine.
3. Distance. Default Cosine Delta (Würzburg variant) to the author's z-space centroid; Classic Burrows Delta reported alongside.
4. Calibrate + verify. Fit a logistic delta → P(same author) using the author's own held-out chunks (leave-one-out) as positives and the other profiles as imposters; report AUC and the decision threshold. A background-rank score (a simple rank fraction, not full Koppel-Winter General Imposters) says how much closer the text is to the target than to any other profile. Calibration is bound to its reference set: change the profiles and compare flags the calibration stale rather than trusting a wrong threshold.

Adams vs three near-neighbour British comic authors (Jerome, Wodehouse, Chesterton), with two deliberately adversarial checks:

Two things keep these honest rather than flattering:

• Same-source control. Separating three authors whose texts share an identical plain-text source shows the signal is authorial style, not a formatting or provenance artifact.
• Leave-one-work-out. Holding out an entire book (different topic, never trained) and still verifying it is a real generalization test, unlike leave-one-chunk-out within a single book. 5 of 6 Adams works pass.

The one LOWO miss is the useful part: The Salmon of Doubt — Adams's non-fiction collection — is rejected by a fiction-built profile and attributed to Chesterton (another essayist). A profile only recognizes the register it was built from, so build it from the kind of writing you intend to verify.

Caveat: each author here is one long book, so author/book/topic are partly confounded and chunk-level AUC is optimistic. A topic-controlled evaluation (short texts, same-topic different-authors, open-set imposters) is v0.2 work; see ROADMAP.md. The method and math were independently reviewed by GPT-5.5 (Codex) and Gemini; their findings drove the calibration-binding, train/query feature-parity, and logistic-regularization fixes in this version.

Reproduce: eval/fetch_corpora.sh, then eval/validate.sh (smoke test) and eval/lowo.sh (the honest cross-work test). The build is ritalin-gated.

See ROADMAP.md. v0.2 adds a content-independent neural style embedding (StyleDistance) as a second, separately-calibrated axis, a frozen reference + chunk manifest so a fine-tuning tool can exclude the judge's text, the full PAN metric suite, and a pure-text "reads-as-LLM" axis.

MIT.