Public methodology
DAI Methodology v1.4 · Updated July 20, 2026
Digital Authenticity Intelligence
Digital Authenticity Intelligence, or DAI, is an evidence interpretation approach for evaluating what available evidence supports about digital media and the claim being made about it.
The DAI Methodology™ gives TRST a disciplined way to assess complex media without forcing it into a simple “real” or “fake” label.
Content Credentials help establish provenance. TRST helps interpret what the total evidence supports.
The core question
A better question than “real or fake”
Digital media can be captured, edited, exported, compressed, screenshotted, recomposed, generated, reposted, or redistributed. Those states are not interchangeable, and a responsible assessment has to account for them.
What does the available evidence most strongly support?
TRST applies that question to produce a short Investigation Brief: what was found, how confident the assessment is, and what could not be determined.
TRST and Content Credentials
Provenance is evidence, not the whole answer
Provenance standards such as C2PA and Content Credentials are foundational to the future of digital trust. When available, they can provide cryptographic information about a file’s origin, edits, and publication path.
DAI treats that information as a critical evidence family. TRST is not trying to replace provenance standards. It is designed to interpret provenance alongside other evidence.
But many real-world media claims require more than provenance alone. A genuine image can be used with a false caption. A credentialed file can be screenshotted and redistributed. An old image can be presented as new. A file may have no credentials because it was compressed, downloaded, screenshotted, or shared through platforms that remove metadata.
Content Credentials help explain where a file came from. TRST helps explain what the available evidence supports.
Absence of credentials does not prove that media is fake. A credential does not automatically prove that a caption, claim, context, or interpretation is true. DAI exists to interpret all available evidence together.
The core principle
Confidence is earned through the systematic reduction of relevant uncertainty.
Evidence
Available signals are examined together. No single detector, credential, or score decides the result.
Uncertainty
Missing evidence limits a conclusion. Missing credentials do not become proof of alteration, generation, or authenticity.
Confidence
Confidence reflects the strength and consistency of evidence supporting the specific finding being reported.
What v1.4 adds
Sharper rules for modern evidence failure modes
Version 1.3 consolidates case-tested rules for degraded files, rapid republication, direct user questions, and evidence gaps that should not be converted into confidence.
Forensics on best evidence
The submitted file is often the most degraded copy. DAI analyzes the strongest available specimen and names which copy was measured.
Single-release distribution pattern
Many exact copies in one short window can document one release being republished, not independent corroboration.
Question-focused response
When a direct question is asked, the assessment answers that question first, within the limits of the evidence.
Status is not confidence
What cannot be assessed carries an explicit status: Not established, Not determined, or Insufficient evidence.
How DAI evaluates media
Evidence becomes an assessment
TRST gathers distinct kinds of evidence, including provenance when available, interprets their relevance to the media and claim being examined, compares responsible explanations, and communicates the best-supported finding.
Observe
Collect available signals
Interpret
Account for media form
Compare
Evaluate explanations
Communicate
Produce a clear brief
Evidence areas
These signals are interpreted together, not treated as independent verdicts.
- Provenance and Content Credentials when available
- File and metadata signals
- Media-form observations
- Synthetic-generation indicators
- Processing and compression patterns
- Known online history
- Reference comparison when prior versions are available
- Media and claim context
Media form matters
Different media require different interpretation
A screenshot should not be judged like a camera photograph. A poster should not be judged like a news photo. An image of an identity document is not the same thing as the underlying physical document.
Camera capture
Evidence can support capture origin without guaranteeing that the scene is unaltered.
Screenshot
Screen-capture properties are interpreted as expected characteristics, not automatic manipulation.
Document image
TRST evaluates the submitted file, not the validity of the underlying physical document.
Designed graphic
Sharp edges, smooth fills, and layout geometry may be normal properties of authored media.
Composite or generated media
Declared, detected, and structural signals are distinguished rather than collapsed into one label.
Editorially processed media
Color, crop, format, or compression changes can be identified without calling the content manipulated.
Investigation Briefs
Designed for practical decisions
Briefs are structured for fast-moving editorial, investigative, and verification contexts. The main assessment appears first; technical detail remains available below it.
- 01Primary finding
- 02Confidence in that assessment
- 03File description
- 04Origin
- 05Material modification
- 06Synthetic generation
- 07Known history
- 08Why this assessment?
- 09Full investigation details
What TRST does not claim
TRST does not certify absolute truth or guarantee authenticity.
TRST does not replace C2PA, Content Credentials, or other provenance standards.
TRST does not treat missing credentials as proof of falsity.
TRST does not treat credentials as proof that a claim is true.
TRST does not replace editorial, legal, forensic, or investigative judgment.
It assesses the submitted media using the evidence available at the time. New evidence may change an assessment.
What remains proprietary
TRST publicly documents the principles and language of the methodology. Its implementation remains proprietary, including provider orchestration, prompts, scoring weights, thresholds, normalization logic, and commercial analysis workflows.
Put the methodology to work