Phraseless

What Phraseless does

Find what to protect. Test one change.

Paste a scene, essay, or chapter. Phraseless compares its measurable patterns with two genre-scoped shelves - 16 named authors across 80 public-domain works - shows the closest measured relationships and exact evidence, then names one effect to protect and one reversible change to test. It does not rewrite the piece.

Analyze my draft - freeInspect a sample result

How it becomes useful

From a shelf position to a decision you can make.

Each Read separates what the text measurably does from what it might mean, then proposes one reversible move and lets you compare it with the original.

  1. 01

    Observed

    The passage, measurement, comparison set, and scope that the claim actually rests on.

  2. 02

    Possible reading

    A bounded interpretation of that evidence - clearly separated from what was measured.

  3. 03

    Try

    One reversible operation, plus the job or quality the revision should preserve.

  4. 04

    Check

    Keep the original, keep the revision, reject both, or stay unsure. The metric never decides for you.

What you can inspect

Every consequential claim carries a receipt.

Receipts let you challenge the output without needing access to Phraseless’s internal tuning. They expose the basis and boundary of a claim - not the machinery used to find it.

Audit one yourself

A receipt with enough detail to reject.

This example is synthetic. Its controls change only this example in the current tab. They are not saved, sent, or counted as product feedback.

Synthetic specimen · no writer source

A four-word interruption acts as this paragraph’s hinge.

EXECUTED

Exact supporting passage

The rain worked at the glass while Mara folded the letter into smaller and smaller squares. The kettle went quiet. She left both things on the table and carried only the silence upstairs.
Measurement and scope
Sentence lengths are 16, 4, and 13 words. The claim applies to this three-sentence paragraph only.
Possible reading
The abrupt loss of sound may turn the paragraph from physical action toward withdrawal. That meaning was not measured.
Caveat
Sentence length cannot establish intent, emotion, or whether the paragraph succeeds.
Reversible experiment
Move “The kettle went quiet” after the final sentence. Preserve the sudden loss of sound. If the paragraph loses its pivot, undo the move.
Does this read earn the claim?

Demo only. Nothing is saved, sent, or counted. No answer is no signal.

Three honest states

Missing evidence stays visible.

EXECUTED

The declared check ran and its evidence passed integrity verification.

NOT ASSESSED

The source, baseline, context, or support needed for the claim was absent.

FAILED

The check ran, but its operation or evidence integrity did not hold.

The naming test

We measured whether we could promise you an author.

“You write like Austen” is only worth saying if a stranger’s prose would not earn the same sentence. In July 2026 we tested that promise directly: hundreds of paste-sized passages from our shelf authors’ own held-out novels, scored against passages from nineteen novelists of the same era who are not on the shelf.

Three independent methods - our production measurement, the standard stylometry family, and impostor-pool verification - hit the same wall. At paste length, prose by an off-shelf novelist lands as close to a shelf author as that author’s own writing. Picking which shelf author sits nearest is strong: right about 95 times in 100 on long passages. But no method could keep false names under our published ceiling of two in a hundred without going silent on most true ones.

So the product tells you the true thing instead: where your piece sits, by how much, with the passages behind it. The naming gate stays published and unpassed. If a stronger shelf and method ever clear it, names come back with a receipt. Until then, a ranking you can check beats a match we cannot defend.

Public coverage receipt

What the analyzer has actually covered.

16 authors · 80 works · receipt version 2

Public-domain Literary V1

version 2026-07-24 · verified 2026-07-26

Status
EXECUTED
Named authors
8
Declared works
40
Construction passagesiExcerpts from the declared works used to establish the analyzer’s reference patterns.
192
Holdout passagesiDifferent excerpts kept out of calibration, then used to check whether those comparisons held.
128
Acceptance-validated profilesiProfiles whose acceptance calibration executed: self/impostor distance separation held on real holdout data. Even these profiles are shown as a ranked position, not a claimed relative - naming waits on a stricter shelf-level gate we have published and not yet passed.
1
Bounded-ranking-only profilesiProfiles eligible only for bounded ranking: closest inside this declared shelf, not a stable or acceptance-validated relationship.
7

Public-domain Essay V1

version 2026-07-26 · verified 2026-07-26

Status
EXECUTED
Named authors
8
Declared works
40
Construction passagesiExcerpts from the declared works used to establish the analyzer’s reference patterns.
192
Holdout passagesiDifferent excerpts kept out of calibration, then used to check whether those comparisons held.
128
Acceptance-validated profilesiProfiles whose acceptance calibration executed: self/impostor distance separation held on real holdout data. Even these profiles are shown as a ranked position, not a claimed relative - naming waits on a stricter shelf-level gate we have published and not yet passed.
0
Bounded-ranking-only profilesiProfiles eligible only for bounded ranking: closest inside this declared shelf, not a stable or acceptance-validated relationship.
8
Declared limitations
  • Two declared shelves span 80 public-domain works by 16 named authors: Literary V1 (creative prose) and Essay V1 (essays and criticism). Neither shelf is a map of literature.
  • Acceptance validation is EXECUTED for 1 of the 16 profiles. The remaining 15 carry bounded-ranking authority only: closest inside the declared shelf, not a stable or acceptance-validated relationship.
  • A closest match on a shelf means closest inside that declared set. It is not an authorship or identity finding.
  • The analyzer measures textual patterns. It does not determine authorship, influence, plagiarism, AI use, intent, or objective quality.
  • Literary V1 covers creative prose from 758 to 2319 eligible words; Essay V1 covers essays from 972 to 3322 eligible words. Lengths outside these ranges can be measured, but the match carries a narrower claim.

Privacy is part of validity

The default Read stays in your browser.

Default

Local Read

Your pasted writing is measured in this browser. The default flow does not send the document to a Phraseless analysis API.

Separate opt-in

Model comparison

No model call occurs until you inspect the bounded evidence bundle and consent. Returned findings must cite evidence already known to the Read.

The output has brakes

What Phraseless refuses to conclude.

No mind reading

A click cannot prove intent, emotion, personality, or what you meant to do.

No authorship theater

A literary relative is not proof of identity, influence, plagiarism, or AI use.

No quality oracle

Metric movement can describe a change. It cannot declare one version better.

No green check on absence

A check that did not run is NOT ASSESSED. Silence is no signal.

Transparency without an instruction manual

We publish the receipt, not the tuning.

You can audit what supported a result and what limited it. Exact thresholds, ranking weights, prompt text, provider routing, anti-abuse controls, and regression assets stay private because they are product IP and because publishing them would make the checks easier to game.

Run a free local ReadSee the complete productInspect operational proofBack to Phraseless