Airport51Darwin
An Airport51 Inc instrument for written work
Darwin
“Origin of Pieces”
Evidence that your writing existed, and stood apart, before the flood of AI-generated lookalikes.
Darwin reads a work the way a naturalist reads a specimen: where it was first sighted, how long the record of it runs, and how much later material now crowds the same ground. It is not an archive. It is the provenance layer above the archives: others show a page, Darwin shows the evidence chain behind it.
Fig. 1 · baseline by quarter
The Human-Origin Prior Curve
Free tier · runs in your browser
The Human-Origin Snapshot
Point Darwin at a URL or paste the text itself. It queries the Internet Archive live, hunts public indexes for independent dated mentions of your work, scores the writing on your device across a full ledger of signals, and scans a live index of later content on the same topic. Your pasted text never leaves the browser; only topic keywords are sent to public indexes.
Darwin queries the Internet Archive (Wayback) live for the earliest capture and the full capture span, then fetches the readable article so the writing itself can be scored.
No work handy? Try the demo with a real 2014 essay.
A Snapshot reports strong, medium or weak human-origin evidence. It never claims content is human-made, and it is not a legal conclusion.
The pipeline
How Darwin reads a work
Four classes of evidence, weighed in order. No single signal decides the verdict, and AI-detection heuristics are treated as a weak witness only: they can lower confidence, never establish it.
Provenance & sightings
Live lookups against the Internet Archive: the earliest capture of your URL and the full span of captures since, corroborated by independent dated mentions found in public indexes (Hacker News submissions, Wikipedia citations). A long capture history plus a second witness is far harder to fake than a single timestamp.
The Prior Curve
Your work is placed on a transparent quarterly baseline (Fig. 1): the probability that a work of its kind and date is human-origin. High before 2022, declining through the post-ChatGPT flood.
The signal ledger, a weak witness
On-device analysis of the writing itself, in both directions: AI-leaning signals (sentence uniformity, generic phrases, spaced em dashes, formula constructions) and human-leaning ones (first-person voice, contractions, concrete specifics, fragments and asides), plus a distinctive-vocabulary fingerprint. These signals inform the verdict; they never decide it.
The similarity wave
Real similarity math (TF-IDF cosine plus word-shingle overlap) against dated later content pulled live from a public index by your work's own keywords. If a wave of lookalikes rose after you published, Darwin charts it by quarter and scores the dilution risk.
The truth boundary. Darwin certifies three things and is deliberately careful with a fourth. Others show a page; Darwin shows the evidence chain behind the page.
Existence
The piece appeared somewhere public, and Darwin can point to where it was observed.
Timestamp
When it appeared or was captured, corroborated across independent witnesses rather than a single source.
Integrity
Whether the observed record has stayed consistent across the capture span since it was first sighted.
Truth
Whether the work’s claims are factually correct. A page can exist, be old, and be unchanged, and still be wrong. Darwin does not rule on this.
The full report
The Beagle Log
Every voyage kept a log. The Beagle Log is Darwin's full, dated, reproducible record of a work's origin, and of everything that has drifted toward it since.
- Full similarity map across a live, continuously monitored later-content index, not a preview corpus.
- Dilution-risk breakdown: attribution confusion, synthetic flooding and market substitution, scored separately.
- Full Human-Origin Confidence and Distinctiveness scoring, versioned and reproducible.
- Evidence appendix as PDF and HTML, plus a creator-origin certificate you can attach to filings and disputes.
- Origin Monitor: ongoing alerts when new content starts crowding your work.
Be aboard for the first voyage.
The Beagle Log engine is fully specified and in build. Leave your email and you will be the first to run it, at launch pricing.
No spam. One email when the Log opens.
Methodology note
What a verdict means, and what it does not
Origin of Pieces verifies provenance, not absolute truth. It shows what was observed, when it was observed, where it was observed, and how strongly independent witnesses support the evidence chain. The Human-Origin Snapshot is a preliminary reading based on visible timestamp and provenance signals plus on-device text heuristics. It is not a statement that content is “human-made” and not a legal conclusion. It reports whether human-origin evidence is strong, medium or weak given the signals available.
Darwin never treats a single archive result as final. A finding is presented as corroborated by N independent witnesses, and one witness alone is flagged as not independently corroborated. AI-detection heuristics are weak witnesses only and are never the basis of a verdict. Scores are reproducible computations, presented as evidence strength rather than authority. Whether a work’s claims are factually correct sits outside Darwin’s boundary and is never scored. The Beagle Log adds the deep provenance scan, similarity-wave mapping against later AI-era content, distinctiveness scoring and the full dilution-risk analysis.
Darwin is an evidence-analysis instrument, not legal advice. Where data was not publicly retrievable it is reported as unspecified, never inferred.