A² — Assay
European-built·Air-gapped

Evidence, claim by claim.

Integrity checking for research, reports and institutional writing, grounded in thousands of peer-reviewed, academic and professional resources.

Status: Pilot Access
The Problem

Fluent falsehood, not grammatical truth.

Modern generative systems write with remarkable fluency, convincing terminology and rhetorical authority, yet have no intrinsic concept of truth. Polished prose can hide claims that do not hold.

How it reads

Fluent, confident, persuasive

Independent audits reduce undetected bias in automated decisions. Most organisations already review their models every quarter. The regulation was withdrawn last year.

FluencyHigh
ConfidenceHigh
AuthorityHigh

Style alone says nothing about truth.

How Assay reads it

Claim by claim

Independent audits reduce undetected bias in automated decisions. The regulation was withdrawn last year.

Supported1
Needs evidence1
Review required1

Illustrative example.

Four ways confident writing fails

Stylistic checks cannot see inside an assertion. Assay can.

Surface fluency

A paragraph that reads like scholarship but is hollow at its factual core.

Fabricated citations

Authors, journals and references that sound right but do not exist.

Statistical distortion

Findings stretched, compressed or invented behind confident prose.

Retracted re-citation

Withdrawn studies cited for years after they were retracted.

The Definition

A forensic approach to institutional writing.

The point is not whether the words were typed by a person or produced by a machine. The point is whether what the words assert can be traced to evidence.

Assay is

  • A verifier of what writing claims, broken down into individual, checkable statements.
  • Grounded in Corpus Ethica and recognised academic registries, with awareness of retracted work.
  • Respectful of the author's own argument, interpretation and voice.
  • A producer of sealed, tamper-evident findings for high-stakes work.

Assay is not

  • A similarity checker that only compares text against other text.
  • An "AI detector" that guesses who, or what, wrote the words.
  • A punitive tool. Assay is advisory and informs expert judgement.
  • A wrapper around a single third-party model.

How Assay works

Decompose, trace, verdict, seal.

Step 1

Decompose

The document is broken into individual, checkable claims. Argument and opinion are recognised and set apart.

Step 2

Trace

Each claim is traced to evidence in Corpus Ethica and to recognised academic registries.

Step 3

Verdict

Every claim receives a plain verdict: supported, needs evidence, or review required.

Step 4

Seal

Findings are sealed into a tamper-evident record, with sign-off proportionate to the stakes.

Four verdicts. One honest score.

Every empirical claim lands on a clear spectrum. The author's voice never lowers the score.

Supported

The evidence supports the claim, with no active retraction.

Review required

The claim conflicts with the evidence, or cites a source that cannot be confirmed.

Author's voice

Argument, interpretation and stance, recognised and kept outside the score.

Integrity
score

Integrity = Supported ÷ all empirical claims

SupportedNeeds evidenceReview required

The author's voice never lowers the score

Beyond detection

From similarity, to authorship guessing, to verifying what is actually claimed.

Generation 1

Similarity checking

Compares text against other text. Finds copying, but not whether anything is true.

Asks: has this been written before?
Generation 2

“AI detection”

Guesses at authorship from style. It cannot see inside an assertion and penalises honest writers.

Asks: who, or what, wrote this?
Generation 3 · Assay

Claim verification

Decomposes writing into claims and traces each one to evidence. Integrity becomes a question of substance.

Asks: is what this says true?

Findings you can stand behind

Sealed, tamper-evident and signed in proportion to the stakes.

Single sign-off

Routine reviews and standard assessments.

Dual executive sign-off

Policy certifications and curriculum releases.

Board-level sign-off

High-stakes findings in regulated sectors, signed by all three.

Who it's for

For anyone whose credibility depends on what their documents claim.

Universities & faculty

Support academic integrity without guessing at authorship.

  • Supervised faculty pilots
  • Feedback that teaches, not punishes
  • Fair to humanities and sciences alike

Research institutions

Strengthen the evidential basis of reports and publications.

  • Citation and retraction awareness
  • Pre-publication review
  • Sealed review records

Policy & public bodies

Verify the claims behind consultations, briefings and guidance.

  • Evidence-traced briefings
  • Air-gapped where required
  • Transparent findings

Boards & regulated sectors

Assurance for documents that carry legal or financial weight.

  • Board-level sign-off
  • Tamper-evident records
  • Deployable fully air-gapped
Architecture

Built the ETHOS way

Our models and deterministic expert engines are developed in-house, within the UK and EU, and deployable fully air-gapped.

Air-gapped by choice

Runs entirely on your own infrastructure when required.

Grounded in Corpus Ethica

Thousands of peer-reviewed, academic and professional resources.

Part of A²

Works alongside the Assurance Studio across the ETHOS platforms.

Join the Assay pilot

Assay is available through our Pilot Access programme for selected institutional partners.

Request Pilot Access