AcuVizionAIEvidence review for AI · Belgium
Guardian by AcuVizionAI

AI results you can check, not just trust.

Guardian recomputes a model's reported metrics from the evidence supplied with them. When the numbers and the evidence disagree, the claim is blocked and the discrepancy stays visible for human review.

Registered Belgian company Private evaluation demo Built for EU AI Act documentation
Claim review · predictions.csvSynthetic example
Recomputed from rows –
BLOCKED_CONTRADICTED

–

Evidence: 24 rows, true → predicted

Edit the declared value, or click a row to change its prediction. Guardian recomputes on every change.

sha256: –

Guardian

From a reported number to a verdict.

Model reports carry numbers: accuracy, F1, AUC, error rates. Guardian checks each one against the files it was computed from, under the evaluation rules the team declared, and records the result in a signed evidence pack.

01 · Intake

Declare the claim and its evidence

The team supplies the reported metrics, the prediction and label files behind them, and the evaluation rules: metric definition, threshold, rounding.

02 · Recompute

Recompute every number

Guardian recomputes each metric from the rows. A difference is checked against the declared precision, so rounding never hides a real discrepancy.

03 · Verdict

Record a verdict per claim

Each claim is allowed, blocked as contradicted, or blocked for missing evidence. The verdicts, file hashes and rules go into a signed pack.

ALLOWED

The declared value matches the value recomputed from the supplied rows at the declared precision.

BLOCKED_CONTRADICTED

The supplied rows give a different value. The claim cannot be used until the report or the evidence is corrected.

BLOCKED_MISSING

A file the claim depends on is missing or does not match the declared IDs. Guardian does not guess.

Who it is for

  • Medical and scientific AI teams preparing technical documentation, such as EU AI Act Annex IV for high-risk systems
  • QA, regulatory and validation leads who sign off on model reports
  • Reviewers who need to check a vendor's reported performance

What it does not do

  • It does not certify a model or make it compliant
  • It does not judge whether the evaluation design is adequate
  • It does not replace clinical or regulatory review
Case study · our own result

We report our results the way Guardian would.

Our skin-lesion model on the ISIC MILK10k benchmark, written up with its limits next to its score.

ClaimValueStatus
Benchmark score (macro Dice over 11 diagnoses)Public leaderboard, 479 test lesions0.625Verified · link
Overall rank, including entries with external dataLive leaderboard, November 2026Top 10Verified · link
Rank among entries without external data1stVerified · link
Margin to the next two entries without external data0.611 and 0.6090.014 – 0.016Within noise
Test-set noise, one modelOur resampling of out-of-fold predictions to the test size≈ 0.03 SDTop 3 not separated
Training data and computeMILK10k training set only, public ImageNet weights1 × RTX 3090Declared
Several submissions were made and the best was kept, which inflates the score slightly. Open the leaderboard ↗
Research

Models for high-stakes, imbalanced problems.

Guardian grew out of our own modelling work, where rare classes, small test sets and leakage make reported numbers easy to overstate.

Dermatology

Multi-modal skin-lesion classification

Clinical and dermoscopic images combined with patient metadata, across 11 diagnoses including rare malignant classes.

ISIC MILK10k · 0.625 · no external data

Materials · VALTRIX

Property prediction from composition and structure

A public, reproducible package for JARVIS exfoliation energy: pinned commit, checksum-verified assets and a CPU replay script.

JARVIS leaderboard PR #34 · open since 15 Oct 2026 · status ↗

Risk

Time-aware tabular models

Point-in-time validation and cost-sensitive thresholds for rare-event detection, where a random split would leak the future.

V7-CORE++ · internal evaluation

Drug discovery

c-Myc/MAX protein interaction

Computational screening and free-energy calculations on a protein–protein interaction that is hard to drug.

Research in progress

Company

A Belgian deep-tech company with one working rule: no evidence, no claim.

Founded
Independent R&D since 2022, incorporated in Belgium in December 2025
Location
Near Leuven, Flanders
Founder
Kamran Soleimani, 20+ years in public-sector IT infrastructure · LinkedIn ↗
Code
github.com/AcuVizionAI ↗
Contact

See Guardian on a model report of your own.

Tell us which metrics you report and what evidence sits behind them. We will set up the private demo with a synthetic or anonymised example.

Write an email