Find the relevant material
Page detection, ink screening and frame selection establish where the examination begins.
THE ADMISSIBLE AI ENGINE
AI-assisted analysis.
Built around
the evidence.
Specialist comparison models, structured explanations and forensic signal analysis work together—connecting the question, the source and the output.
CONNECTED ANALYSIS ARCHITECTUREIllustrative
Learned comparisonMeasured signalsDefined checks
PURPOSE-SPECIFIC PROCESSING
Different questions.
Connected analytical tools.
Select an evidence walkthrough to see these roles in context.
Capture useful features
Visual descriptions, measured geometry, sound characteristics and file records expose different signals.
Compare or measure
Purpose-specific comparisons, calibrated evidence measures and defined checks address the selected question.
Keep the output traceable
Attach regions, references, timestamps and processing context to the resulting exhibit.
Purpose-specific intelligenceSeven evidence categoriesComplementary analysis methodsSource-connected outputs
01 / SPECIALIST CAPABILITIES
More than one model.
More than one kind of evidence.
The engine combines learned comparison, objective measurement, signal processing and file-integrity checks. Each contributes a different kind of evidence.
Specialist comparison models
Explore visual signature relationships, handwriting authorship evidence and voice characteristics with outputs appropriate to each comparison.
- Multi-reference signatures
- Handwriting comparisons
- Voice-similarity measurements
Structured visual explanations
Connect image-quality assessment, independent specimen descriptions and systematic feature tables to the measured evidence.
- Image assessability
- Observed similarities & differences
- Competing explanations
Speech & media screening
Create confidence-marked speech drafts and screen supported visual patterns for specific manipulation or generation indicators.
- Per-word uncertainty
- Sampled face indicators
- Image & video screening
Signal & geometric analysis
Measure captured motion, writing geometry, recording continuity, frame timing and calibrated scene relationships.
- Real capture dynamics
- Timing & continuity
- Calibrated measurements
File & provenance checks
Read file structure and embedded records, inspect content fingerprints and verify available signed provenance manifests.
- Metadata & structure
- Content integrity
- Provenance manifests
Non-generative clarification
Reduce noise or adjust contrast and level with captured information, producing distinct image, audio or video working copies.
- Image denoising
- Audio clarification
- Adjacent-frame processing
MODELS ≠ WORKFLOWS
A connected engine, not one model per tool.
A workflow can combine several learned components, reuse a shared component or rely on deterministic measurements alone. Analysis counts and unique model counts describe different things.
02 / FOLLOW THE QUESTION
See how the engine
fits your evidence.
Choose an evidence category to follow a representative processing path. These illustrations explain workflows; they do not process or assess uploaded files.
SIGNATURES / REPRESENTATIVE PATHWAY
Compare a signature across known references.
EVIDENCE IN
A questioned signature and suitable independently sourced references. PDF and image support depends on the chosen comparison.
- 01 / PREPARE
Separate questioned material from the reference set. Check format, visible detail and reference availability.
- 02 / COMPARE
Measure visual relationships with complementary comparators and, where supported, account for variation among known signatures.
- 03 / CONNECT
Keep descriptive similarity, independent distance and calibrated evidence outputs distinct, with the reference context retained.
OUTPUT CONTEXT
- Reference variation
- Visible stroke relationships
- Method-specific comparison outputs
Input requirements & output scope
Visual likeness does not reconstruct writing pressure or speed. Those measurements require genuine digitizer data.
HANDWRITING / REPRESENTATIVE PATHWAY
Connect measured writing features to a structured explanation.
EVIDENCE IN
Questioned and known writing, or samples from at least two candidate writers for candidate ranking.
- 01 / MEASURE
Extract writing structure and record supported geometry, spacing and stroke characteristics.
- 02 / COMPARE
Run the chosen authorship or candidate-ranking workflow, retaining the appropriate comparison output.
- 03 / EXPLAIN
Describe samples independently, organise observed differences and consider competing authorship explanations.
OUTPUT CONTEXT
- Authorship evidence
- Candidate-writer ranking
- Quantified writing features
Input requirements & output scope
Candidate ranking is closed-set and returns a nearest candidate. Authorship evidence depends on the comparison population and validation scope.
DOCUMENTS / REPRESENTATIVE PATHWAY
Combine document records with captured physical features.
EVIDENCE IN
Original supported files, page images, or purpose-made raking-light and transmitted-light photographs.
- 01 / INSPECT
Read supported embedded metadata and check document-history inconsistencies.
- 02 / EXTRACT
Locate signature candidates and link each crop to its document page.
- 03 / REVEAL
Apply suitable enhancement to captured indentation or watermark signals, or decode eligible printer marks.
OUTPUT CONTEXT
- Embedded file records
- Source-linked signature crops
- Capture-specific enhancement
Input requirements & output scope
Metadata is editable. Raking light, transmitted light and eligible printer marks are separate capture requirements; ordinary scans cannot supply every signal.
IMAGES / REPRESENTATIVE PATHWAY
Investigate a specific image-authenticity question.
EVIDENCE IN
An original image, a defined question and suitable native-resolution references for camera-pattern comparison.
- 01 / SELECT
Match the question to duplication, compression, source characteristics, geometry or provenance.
- 02 / ANALYSE
Produce suitable region maps, correspondence points, source-comparison measurements or manifest checks.
- 03 / CONNECT
Retain the original and show how the selected output relates to the suspected region or source claim.
OUTPUT CONTEXT
- Duplication correspondences
- Manipulation indicators
- Signed provenance checks
Input requirements & output scope
Compression workflows need a suitable JPEG history. Missing provenance is inconclusive; statistical screening and cryptographic verification answer different questions.
AUDIO / REPRESENTATIVE PATHWAY
Separate speech, speaker and recording-content questions.
EVIDENCE IN
Original audio or a supported video audio track; known material is needed for a comparison.
- 01 / PRESERVE
Keep original audio distinct from any listening copy and inspect the available recording information.
- 02 / PROCESS
Choose confidence-marked transcription, voice similarity, shared-content comparison or continuity analysis.
- 03 / EXPLAIN
Keep uncertain words visible and connect comparison scores or continuity traces to their recording context.
OUTPUT CONTEXT
- Word-level uncertainty
- Voice versus content comparison
- Continuity and listening copies
Input requirements & output scope
A raw voice-similarity score is not a calibrated identity conclusion. Hum continuity needs a detectable trace; transcription uses the original recording.
VIDEO / REPRESENTATIVE PATHWAY
Connect the frame sequence to its timing and source.
EVIDENCE IN
Original video; camera references, timing anchors or scene calibration when the selected workflow requires them.
- 01 / READ
Inspect container properties, encoded structure and the decoded frame/timestamp record.
- 02 / EXAMINE
Choose full-sequence timing checks, sampled screening, camera alignment or calibrated measurement.
- 03 / CONNECT
Keep source timing, sampling coverage, uncertainty and derived viewing copies visible in the output record.
OUTPUT CONTEXT
- Frame and timing records
- Targeted synthetic-media screening
- Aligned, calibrated measurements
Input requirements & output scope
Sampled screening does not examine every frame. Speed and geometric measurements need suitable source timing and calibration.
SPECIALIST TOOLS / REPRESENTATIVE PATHWAY
Read the technical traces that sit between disciplines.
EVIDENCE IN
Suitable original image or video files, with source history, timing anchors or physical calibration as required.
- 01 / DEFINE
Identify an encoding, recapture, multi-camera timing or motion-analysis question.
- 02 / MEASURE
Inspect supported encoding families, local statistics, timing relationships or codec-level motion information.
- 03 / RECORD
Preserve reference coverage, timing uncertainty and the technical context needed to understand the result.
OUTPUT CONTEXT
- Encoding and recapture
- Camera synchronisation
- Codec motion and timing
Input requirements & output scope
These specialist tools connect to the corresponding image or video workflows. Compatible formats, reference coverage and codec support vary by method.
HOW THE ANALYSIS WORKS
A page becomes a linked specimen.
EVIDENCE / EXPLAINEDGuided demonstration
SPECIMEN AGREEMENT DEMO / 01
Authorisation record
Fictional document for a guided workflow.
Signature
Candidate region
SOURCE PAGE 01 → LINKED SPECIMEN
Page context retained 01 → specimen A
Detection & extraction / IN PRACTICE
From a page to a specimen.
A signature on an agreement
01 / Prepare the page
Keep the source page and its position in the document. Supported page rendering can produce a 600 dpi export; the original capture determines the detail available.
RECORDSource page · rendering context
02 / Locate the candidate
Locate signature-shaped regions. The wider mark-extraction workflow also screens ink and ranks candidates such as initials, names and company chops.
RECORDCandidate region · page coordinates
03 / Trim to the ink
Reduce surrounding whitespace around the selected ink region. Keep the original page available alongside the extracted specimen.
RECORDIndividual crop · original retained
04 / Connect the specimen
Compare the extracted sample with references where supported. Retain the crop-to-page relationship with the comparison and report exhibit.
RECORDSource-linked specimen · comparison input
Demonstration material · annotations explain the workflow; no case result is calculated.
03 / FROM MEASUREMENT TO EXPLANATION
Make the reasoning
as visible as the result.
In the handwriting-comparison workflow, structured visual reasoning connects independent observations and alternative explanations to the computed comparison evidence.
- 01
Assess the image
Identify whether the available writing is suitable for a meaningful visual description.
- 02
Describe independently
Document each specimen’s visible characteristics before bringing the samples into direct comparison.
- 03
Build the feature table
Organise letterforms, spacing and other observed agreements or differences systematically.
- 04
Consider alternatives
Explore common authorship, disguise, coincidental similarity, reuse and image artifacts as relevant explanations.
- 05
Connect to the measured output
State whether the explanation agrees or disagrees with the computed evidence; do not invent an authenticity percentage.
04 / UNDERSTAND THE OUTPUT
Different signals.
Clear distinctions.
The output format follows the question. A similarity value, a diagnostic map and a signed provenance record are not interchangeable answers.
Comparison evidence
Descriptive similarity and raw distances describe a measured relationship. Calibrated likelihood ratios, where available, express relative support under specified propositions.
Screening indicators
Region maps, sampled image scores and continuity flags highlight a specific pattern or inconsistency within the method’s stated coverage.
Integrity & provenance
Content fingerprints and available signed manifests address technical integrity or recorded provenance, rather than deciding whether everything depicted is true.
Measurements & working copies
Geometry, timing and pressure retain their capture and calibration context. Clarified copies remain distinct from the originals.
05 / REFERENCES & FURTHER READING
Explore the concepts
behind the capabilities.
Independent technical reading on evaluation, evidential weight, media processing and provenance. Platform-specific scope and validation are documented separately.
NISTAI RMF 1.0 · 2023
AI Risk Management Framework
Background on evaluating AI trustworthiness, validity and reliability. Useful for understanding why performance must be measured for the intended use.
Read the reference (opens a new tab)www.nist.govENFSIApproved version 3.0 · 2015 · PDF
Evaluative reporting in forensic science
Explains how findings can be evaluated under competing propositions and why evidential weight differs from a probability that a proposition is true.
Read the reference (opens a new tab)enfsi.euSWGDE20-A-001-2.0
Enhancement of digital audio
Background on recording quality, repeatable processing and the relationship between enhancement goals and the signal actually present in a recording.
Read the reference (opens a new tab)www.swgde.orgSWGDE18-V-001-2.0
Digital forensic video analysis
A technical overview of video examination, file handling, timing, clarification and the documentation of derived material.
Read the reference (opens a new tab)www.swgde.orgContent CredentialsOfficial FAQ
Provenance & signed records explained
Explains how signed provenance records describe content history. This is a different question from whether the depicted event or a statistical screening result is correct.
Read the reference (opens a new tab)c2pa.orgReferences explain the underlying concepts; their inclusion does not indicate product certification, endorsement or validation by the publishers.
06 / ENGINE QUESTIONS
A clearer view
of the technology.
Does each analysis use a separate AI model?
No. Some workflows combine multiple learned components, others reuse shared components, and deterministic checks do not require a learned model. A count of analyses is not a count of unique deployed models.
What does the AI-assisted explanation add?
The handwriting workflow can assess image suitability, describe specimens independently, organise a feature table, consider alternative explanations and relate its explanation to the computed comparison output.
Is every output an authenticity verdict?
No. Outputs include descriptive similarity, evidential-weight measures, raw comparison values, diagnostic maps, timing records, verified provenance information and separate clarified copies. The selected workflow defines what an output means.
Does the platform generate details during enhancement?
The described enhancement workflows use non-generative processing with information already captured in the original. They produce separate image, listening or viewing copies rather than reconstructing missing detail.
Where can I check a method’s requirements and validation?
Each module page describes the supported inputs, output type and technical scope. Research and validation resources document the available evidence separately from the external background references linked above.
YOUR EVIDENCE. A FOCUSED WORKFLOW.
See the engine
in context.
Explore the capabilities around the questions your team investigates.