Published reference
A result reported by an external study. It establishes what that study observed—not what oNeura independently predicted.
Shown below: corrected PLATO reference.
Evidence & reproducibility · Updated October 17, 2026
Evidence before
confidence.
A plausible simulation is not a validated prediction. Here is how we distinguish the two—and what needs to accompany a result before you rely on it for a research decision.
Current public status
Earlier comparison claims are under review.
We have withdrawn the previous trial-accuracy scorecards from this page. Their displayed comparisons did not provide the complete run artifacts, calibration history and held-out evaluation needed to substantiate validation claims. They should not be treated as independent validation.
No independently verified model prediction is currently published on this page. We are not claiming clinical validation, regulatory approval or a completed whole-body simulation. Research services are scoped case by case.
Read the label
Four different kinds of evidence.
Engineering demonstration
An illustration of implemented behavior. Working controls, deterministic output or passing software tests do not establish biological accuracy.
Our browser demos belong in this category.
Retrospective comparison
A model is compared with outcomes that may already be known during development. Inputs, fitted parameters, endpoint mapping and mismatches must be disclosed.
Prior public comparisons: under review.
Held-out validation
Predictions are frozen before evaluating data not used to select or tune the model. Report the predeclared metrics, all failures and uncertainty, with independent review identified separately.
No substantiated held-out study published here yet.
Published reference · not an oNeura prediction
A correction to the PLATO comparison.
In PLATO, the ticagrelor group’s primary composite endpoint—death from vascular causes, myocardial infarction or stroke—was 9.8% at 12 months, compared with 11.7% for clopidogrel. All-cause mortality was 4.5% with ticagrelor versus 5.9% with clopidogrel.
The previous site incorrectly labeled the composite as mortality and paired the published ticagrelor result with a simulated prasugrel entry. Those were not valid like-for-like comparisons. We have removed the claimed model score rather than inventing a replacement prediction.
Read Wallentin et al., NEJM (2009) ↗ · DOI and full publication ↗
Work in progress
Cardiac assay reproduction.
Our current scientific priority is reproducing the 33-drug cardiac assay study by Nguyen et al. (2017), starting with vehicle and verapamil. This is ongoing work, not a claim that the full paper has been reproduced.
A credible comparison must preserve the assay’s actual endpoints: an electrical duration is not a substitute for mechanical relaxation, and a reconstructed illustration is not a raw simulated time series. Data access, endpoint definitions and reproducible run artifacts remain part of the evidence requirements.
Before a result becomes a claim
What a study record needs.
- Question & endpoint. Intended research use, population or assay, units, timing and a predefined success criterion.
- Inputs & provenance. Data source, rights, transformations, boundary conditions and explicit assumptions.
- Calibration & holdout. What was fitted, what was known during development, and what was genuinely withheld.
- Reproduction instructions. Model and code versions, commands, environment, seeds and artifact hashes.
- All outputs. Raw results, comparisons, uncertainty, errors, exclusions and unsuccessful runs—not just favorable examples.
- Limits & review. Supported scope, known failure modes, reviewer role and permission to share the materials.
Start with an empty specification.
This JSON file is a blank template, not a completed study, validated result or customer case study. Empty fields must be completed and reviewed; downloading it establishes no evidence.
Download empty study templateHave a specific research question?
Start with the endpoint and the data you can share. We can discuss feasibility before defining a paid scope.
Discuss a research pilot →Or email bobby@blackweb.ai with a nonconfidential summary. This page does not send a message.