Cognitive Authorship Analysis

Was a human mind present

when this was written?

Impronta reads the traces that human writers leave without knowing it — the clustering of uncertainty, the rhythm of thought, the marks of a mind at work. Not what AI sounds like. What cognition leaves behind.

🔒Zero Text Retention — Architectural
📄Peer-Reviewed Methodology
🌍ESL FPR Published: 3.8%
Not for Sole Evidence
Analysis Input · Impronta v5 Process Archaeology · 23 Features
Stored in your browser only
I
II
III
IV
V
VI

Diagnostic indicators only · Not for use as sole evidence in any academic integrity proceeding · Your text is never stored

The Method

Giovanni Morelli read paintings the way Impronta reads text

In the 1880s, Morelli discovered that art forgers faithfully reproduced the grand compositional elements — the Madonna's posture, the colour palette, the architecture. What they could not reproduce were the small, unconscious details: the fold of an earlobe, the curve of a knuckle. Those details emerged from the hand itself, not from intention. Forgers cannot fake what was never consciously made.

AI language models are skilled forgers. They reproduce the grand rhetorical structure of human essays. What they cannot reproduce are the cognitive process traces: hedges clustering around genuine uncertainty, sentence rhythm driven by cognitive load, the marks of a mind revising itself mid-sentence. These traces emerge from the writing process itself — and autoregressive token prediction has no writing process.

Impronta reads the knuckles.
Theoretical framework: Process Archaeology.
Primary metric: HumAI@5%FPR — humanized AI recall at a 5%
false positive rate on genuine human academic texts.
Reference: Ginzburg, C. (1983). Clues: Roots of an evidential paradigm.
Fig. I — Comparative Study of Ear Forms
After G. Morelli, 1891

Raphael
Botticelli
Perugino
Signorelli

"The details the hand produces without knowing it
are the only details that cannot be forged."

Fig. II — Cognitive Anatomy of the Writing Hand
Six Layers · Process Archaeology · Impronta v5
LAYER I Peripheral Tissue sentCV · hedgeCluster LAYER II Cognitive Load Arch. entropyVar · paraCV LAYER III Knowledge Sedimentation ttr · lexCohesion LAYER IV Process Traces repairs · temporal LAYER V Semantic Trajectory terminalReturn · coherence LAYER VI Dependency Structure prop_density · spaCy

"The features that reveal cognitive origin
correspond to the traces the hand leaves without knowing."

How It Works

Four steps. Twenty-six features. One probability.

The core detection runs on Impronta's own server — no text is transmitted to external AI services until you request the rhetorical analysis layer.

01
Submit text
Paste or upload analytical prose of ≥250 words. Select genre and ESL mode if applicable.
02
Extract 26 features
Six cognitive layers are computed server-side via Python, spaCy, and XGBoost. No external API call.
03
Score and classify
Features pass through a calibrated XGBoost model trained on 6,400 texts across four authorship classes.
04
Receive report
An interpretable report returns feature-level scores, confidence interval, adversarial spectrum position, and rhetorical analysis.
What the features show

Every human value in Fig. III is the mean from a corpus of genuine academic essays. Every AI figure is from the same texts rewritten by large language models with explicit humanization instructions — the hardest detection case.

The gap between the bars persists across every model family tested — Mistral, LLaMA, Qwen, Gemma, Yi-Large — because these are cognitive process absences, not distributional artifacts of any particular training corpus.

You cannot attack all twenty-six traces simultaneously with a single prompt.

Fig. III — Cognitive Process Signatures
Human vs. Adversarially Humanized AI · Mean values
Human
Humanized AI

hedgeCluster
78
31 Epistemic clustering
sentCV
66
22 Rhythm variance
temporal
55
18 Time anchoring
repairs
42
8 Self-correction
entropyVar
61
24 Info. density var.
paraCV
58
21 Paragraph length var.
Cognitive Rhythm
sentCV · hedgeCluster
entropyVar · paraCV
Knowledge Arch.
ttr · availSpec
lexCohesion · arcVar
Process Fossils
repairs · temporal
revFossils · grounded_hedge
Privacy Architecture

The AI detector that cannot access your students' work

When you submit a document to Impronta, what happens to it?
Text enters our server over HTTPS. It is processed in RAM only — never written to disk, never logged, never passed to a database. Twenty-six features are extracted. A score is computed and returned to you. The text ceases to exist in our systems the moment the response is transmitted. Not eventually. Immediately. There is no deletion schedule because there is nothing to delete.

Text enters server (HTTPS only)
RAM only — never written to disk
26 features extracted
Score returned to you
Text ceases to exist

"Impronta's zero-retention design is not a policy choice. It is an architectural consequence. You cannot delete data that was never stored. Submitted text is never logged, never backed up, never retained in any form. No Impronta employee can access it — because it is not there."

What we retain (per account) vs. what we never retain
What we store
CNI score · 26 feature values (26 numbers)
Word count · Timestamp · Mode selected
Genre selection · ESL mode flag
None of this is your text or student identity
What we never store
Submitted text · Sentence-level text
Student name · Course information
Document title · Highlighted text
We do not have it — it was never kept
How Impronta compares to major competitors
Platform Text stored? Default retention Trains on submissions? Student IP? Methodology published?
Turnitin Yes Indefinitely (unless inst. requests deletion) Not publicly denied Non-revocable license granted No
GPTZero Unverified claim Session (unverified) Not stated Not addressed No
Originality.ai Yes — by default Stored in account history Not addressed Not addressed No
Copyleaks Enterprise-configurable Not published Not addressed Not addressed No
Winston AI Not published Not published Not addressed Not addressed No
Impronta Never — architectural Zero — nothing persisted Never — architectural No license acquired Yes — peer reviewed

Sources: Turnitin Privacy Policy (Feb 2026) · GPTZero privacy policy · Originality.ai terms · Competitor documentation reviewed May 2026. "Unverified" indicates a privacy claim was made but is not independently auditable.

Request DPA for Institutions Read Full Privacy Architecture
Adversarial Detection

We detect what every other detector misses

Process archaeology targets the hardest case: AI text explicitly humanized to evade detection. On this task, no published system comes close.

Adversarial Detection Spectrum

Where text falls on the authorship continuum — from verified human to clean AI generation

Human Lightly
humanized AI
Aggressively
humanized AI
Clean AI
HumAI@5%FPR — Humanized AI Recall at 5% False Positive Rate

The proportion of explicitly humanized AI texts correctly flagged when the detection threshold is set such that no more than 5% of genuine human texts are falsely flagged. This is the operationally correct metric for academic integrity. No other published detector reports this metric.

System HumAI@5%FPR AUC (academic) Cross-model (5 families) Method published? Notes
Impronta v5 79.5% 0.9445 AUC 0.925 mean Yes — peer reviewed +52.8pp over RADAR
RADAR (NeurIPS 2023) 26.7% 0.789 AUC 0.774 mean Yes (adversarial training) Designed for adversarial — still fails
roberta-mixed-detector 38.5% 0.937 In-distribution only No Excellent on clean AI; fails humanized
chatgpt-detector 28.7% 0.779 Poor cross-model No Near-chance on non-ChatGPT models
Binoculars-Light 24.1% 0.605 Requires 28GB GPU Partial CPU approximation only

Source: corpus_v3 test set (n=784) · Primary benchmark · All systems evaluated on identical test set with identical threshold protocol. Cross-model test: gsingh benchmark, format-normalized, DeLong p<.001 on 4/5 comparisons. Limitations disclosed in full methodology.

Evidence

What we have demonstrated — and what we have not

Confirmed
79.5%

Humanized AI recall at 5% false positive rate on adversarially humanized academic essays — surpassing RADAR by 52.8pp (z=8.56, p<.001)

+15.1pp

Mean AUC advantage over RADAR across 5 completely independent model families (Mistral, LLaMA, Qwen, Gemma, Yi-Large) — zero training overlap, p<.001 on 4/5

0.925

Mean AUC on 5 independent model families tested on journalistic prose (domain shift confirmed) — process archaeology generalizes across models and domains

100%

Detection maintained against targeted adversarial hedge-clustering prompt attack. The attack paradoxically increased detectability by over-regularizing hedge distribution.

3.8%

False positive rate on Chinese L1 English exam essays — below the native English rate of 15.4% at the same threshold. ESL risk is lower than predicted by SLA literature for this population.

ESL Fairness — Confirmed

Impronta is the only AI detection system that has published false positive rates by population group. The documented 3.8% ESL FPR (Chinese L1) contradicts the 15–30% ESL false positive rates reported across competing tools. Important caveat: this result is based on high-school exam essays; university-level L1 validation is ongoing.

Known Limitations
Short texts

Requires ≥250 words. Paragraph-level features collapse on shorter submissions, inverting the decision boundary. Use CNI-Short mode (AUC 0.655 on RAID benchmark) for shorter texts.

Not a verdict

Every score is a probability estimate with a documented ±0.08 confidence interval. Impronta is one signal among several in any human review process. Never sole evidence.

Clean AI detection

Impronta is designed for adversarially humanized academic prose. On journalism-domain clean AI text, distributional detectors (roberta-mixed AUC 0.991) outperform Impronta (0.925). Different tools for different tasks.

Domain validation

Calibrated on academic essays, tested on journalism. Legal text, technical documentation, and creative non-fiction have not been formally validated. Domain-appropriate confidence adjustments shown when genre is declared.

Construction dependency

The primary benchmark was constructed using humanization strategies designed by CNI's authors. Mitigated by three independent external evaluations but disclosed honestly. See full methodology for details.

Architecture

26 features across six cognitive layers

Each layer targets a different cognitive operation of the writing process — not what AI text looks like, but what human cognition produces.

Layer I
Peripheral Tissue
sentCV · hedgeDensity · hedgeCluster
Surface rhythm and epistemic distribution. The fastest-loading cognitive signatures — visible without reading for comprehension.
Layer II
Cognitive Load Architecture
entropyVar · paraCV · overAssert · presuppose
The macro-structural organization of informational density. Whether the document breathes like a mind at work or flows with mechanical uniformity.
Layer III
Knowledge Sedimentation
ttr · availSpec · lexCohesion · arcVar
The depth and specificity of disciplinary knowledge embedded in vocabulary, examples, and argumentative moves. The trace of having genuinely learned something.
Layer IV · Primary
Process Traces
repairs · temporal · revFossils · avgWordLen
Direct fossil evidence of the cognitive production process — self-repairs, temporal anchors, and revision artifacts. The strongest adversarial humanization signal.
Layer V
Semantic Trajectory
terminalReturn · localCoherence · embeddingDiv.
The global semantic arc — whether the text departs from, develops, and returns to its own ideas in the way a mind working through a problem does.
Layer VI · spaCy
Dependency Structure
prop_density · passive_density · syntactic_depth · floating_hedge · grounded_hedge · init_conj · np_timing
Syntactic microstructure via dependency parsing. The most adversarially resistant layer — grammatical architecture cannot be targeted by surface-level prompting.
The Arms Race Argument

We don't chase models.
We measure what every model lacks.

"Impronta detects AI text from five model families it has never seen,
including models released after training was completed."

Every distributional AI detector — systems trained to recognize what GPT-4 or Claude "sounds like" — must be retrained each time a new model is released. Their accuracy degrades as models improve and as humanization techniques evolve.

Process archaeology works differently. It measures the absence of cognitive production traces — traces that are present in human writing by definition and absent from AI generation regardless of the model, the scale, or the explicit humanization instruction. No language model generates text through the cognitive operations that produce these traces. That is not a training data limitation. It is a mechanistic difference.

The empirical confirmation: Impronta v5 achieves mean AUC 0.925 on five model families with zero training overlap, tested in a different domain (journalism) than its training domain (academic essays). The process archaeological features generalize because they measure something architecturally absent from all current language models.

For Educators

Free access for instructors,
for a full academic year

University instructors, professors, and teaching assistants with an institutional email address receive full Impronta access at no cost for 12 months — a complete academic year cycle. Academic integrity requires tools that are transparent, evidence-based, and held to scientific standards. Annual reconfirmation keeps the relationship current.

Verified by institutional email (.edu, .ac.uk, .edu.mx, and equivalents)
Full access for 12 months · Annual reconfirmation · Never shared with third parties

No institutional email? Instructors at community colleges, international universities, or using personal email may join the educator waitlist. We review and approve manually within 48 hours. Join the waitlist → educators@impronta.ai
Analyze free