Defined configuration
A result belongs to a specified asset set, version, configuration, transformation path, and verification procedure. Those conditions are part of the evidence, not footnotes to ignore.
Evidence & methodology
Evidence before claims.
This page summarises the public benchmark context, a documented redistribution result, and the current external-evaluation status. It is designed to make the scope and limits of each statement visible, not to claim universal performance.
Documented benchmark scope
What the public figures cover
12,331
unique images
160,303
controlled transformations
61,655
negative controls
99.55%
baseline perfect UID recovery: 12,276 of 12,331 images
Results reflect the documented benchmark version, dataset, configuration, and transformation set. They do not imply equivalent performance across all edits or deliberate removal attacks. Counts refer to documented internal benchmark runs and do not constitute independent external validation.
Method and limits
How to read the results
Negative controls and failure modes
Negative controls and non-recoveries matter alongside successful recovery. Aggressive geometric edits, AI-based removal attacks, and heavy compositing remain active work.
Measured visual-quality results are configuration-specific and vary by image, embedding configuration, and output format. Publication of additional detail is limited to material that is safe to share publicly; private benchmark bundles and investor-only material are not linked here.
Baseline result
99.55% baseline perfect UID recovery: 12,276 of 12,331 images
This is a baseline UID-recovery result from the documented benchmark configuration. It reflects defined test conditions and is not a generic accuracy, authenticity, deepfake-detection, protection-rate, or all-transformations claim.
Redistribution result
98.5% UID recovery after manual WhatsApp transfer · 985/1,000
This is a UID-recovery result from the documented manual WhatsApp-transfer test. It is not a generic accuracy, authenticity, deepfake-detection, protection-rate, or all-transformations claim.
Measured output
Recovery of the embedded UID after the specified transfer path, reported with its visible numerator and denominator.
Test boundary
A documented manual transfer condition, not a substitute for results across every messaging service, edit, or removal attempt.
What it supports
A checkable origin-continuity signal for this defined path, not a judgment that content is true.
External evaluation
EIC Accelerator Step 1 passed
puit.is received a score of 3.00/4.00 and was invited to Step 2 in July 2026. Evaluators judged all aspects of TRL 5 convincingly completed, with evidence supporting the claimed TRL 6.
The EIC Accelerator evaluation is ongoing; puit.is is not EIC-funded.
Need evidence for a specific workflow?
A scoped partner test can define the assets, transformation path, acceptance criteria, negative controls, and the evidence outputs appropriate to a real workflow.
Test your workflow