Scan
Run the file locally and capture its structural fingerprint without creating a cloud copy.
The Ocolta evidence model
A better answer than “trust the score.”
Start with a reproducible local scan. When a case needs a model-assisted second pass, opt in to a separate Deep Review that reports evidence, confidence, benign explanations, and what remains unknown.
0files uploaded by the public scanner
3result states: clear, review, unable to determine
2.0published heuristic ruleset — not a fraud benchmark
Version 2.0 is a transparent, non-additive heuristic triage ruleset, not a validated fraud-detection benchmark. No accuracy, false-positive, or false-negative claim is made. Inspect the published judgment steps on the methodology page.
Integrity Scan · on-device
Signals you can trace back to the file.
File identity
Binary signature, declared MIME type, file size, PDF version, page count, and a SHA-256 fingerprint for case continuity.
Creation context
Producer and creator metadata, creation and modification timestamps, and whether those facts fit the claimed source workflow.
Revision structure
Incremental-save markers and prior revision pointers, explained as evidence that can be legitimate in signed or form-filled PDFs.
Active content
Embedded scripts, file attachments, and digital-signature structures that deserve different handling.
Document math
Bank statement and paystub relationships, recalculated from values a reviewer confirms from the document.
Deep Review · $5, or Deep Review+ · $10
More context, with a different privacy boundary.
After explicit consent, Ocolta sends the selected file to its server for the opt-in model review. The model separates fraud-risk indicators from AI-generation indicators and reports where each observation appears, its confidence, plausible benign explanations, limitations, and proportionate next checks.
Model output is not a finding of fact.
It can be incomplete, inaccurate, or unable to determine. A reported indicator never proves fraud, forgery, authenticity, AI generation, or intent, and it must not replace source verification or human judgment.
What it does not claim
Honest limits make the evidence useful.
Ocolta does not confirm account ownership, contact an employer or bank, authenticate an issuer, or determine applicant eligibility. A clean local scan or a Deep Review with no supported indicators means only that the relevant workflow did not surface a supported reason for follow-up.
Editing is not proof of fraud.
A bank may optimize a PDF. An applicant may combine pages. A signature tool may add a revision. The result explains plausible benign causes and asks for stronger source evidence.
Risk is not a protected-class-neutral decision.
Use a written, consistently applied policy. Ocolta supports document review; it is not an automatic approval, denial, tenant-screening, or adverse-action system.
The review loop
From file to defensible next step.
Reconcile
Confirm the values you can see, then test the arithmetic and consistency rules that should hold.
Verify
Use the report to request the missing original, source confirmation, or adjacent evidence—not to make an unsupported accusation.
Start with evidence
Know what deserves a closer look.
Run a private, browser-based integrity scan. Your file never leaves your device.
Scan a document