BioinvestGPT ApS · Research use only
BVCT evidence and prospective predictions
BioinvestGPT publishes every adjudicated BVCT prediction on its public prediction dashboard, the source of record. As of 24 September 2026, the company-reported track record is 95.6% accuracy (674/705 scored readouts), 89.7% positive predictive value and 99.1% negative predictive value. Each prediction is timestamp-audited to predate its trial readout.
Last updated · Maintained by BioinvestGPT ApS
What track record does the dashboard report?
A frozen, company-reported snapshot of the final adjudicated records on the public prediction dashboard, the source of record. Cohort cutoff: . Counts are per scored readout, as published on the dashboard.
| Measure | Scored readouts | Percentage |
|---|---|---|
| Accuracy | 674/705 | 95.6% |
| Positive predictive value (PPV) | 234/261 | 89.7% |
| Negative predictive value (NPV) | 440/444 | 99.1% |
Of 705 scored readouts, 234 were true positives, 440 true negatives, 27 false positives and 4 false negatives. A further 5 readouts could not be adjudicated and sit outside these denominators. The 705 scored readouts belong to 692 predictions: 13 predictions have more than one scored readout. Scoring follows the rule published on the dashboard.
These figures restate the dashboard; this page does not independently re-adjudicate the records. Check the dashboard for later changes, and the methodology for how to read each measure.
Where is the BVCT track record?
The public prediction dashboard at data.bioinvestgpt.com holds the final adjudicated records, with trial identifiers, prediction dates, prediction statements and outcome sources, including misses and readouts that could not be adjudicated. The figures on this site are a frozen snapshot of that record.
How are predictions timestamped?
Each prediction is recorded in a document signed through PandaDoc before the trial's readout, and the e-signature audit trail timestamps it. The process has been reviewed by a pharmaceutical partner; details on request. The signed prediction documents themselves are kept private; the dashboard lists each record.
A timestamp shows that a prediction existed at a particular time. It does not, by itself, establish predictive accuracy, clinical validity or independent scientific review; those need separate evidence.
How should I inspect a prediction?
- Identify the exact question. Record the trial ID (NCT number), intervention, population, comparator, endpoint and prediction index.
- Read the prediction. Preserve its wording, its date and any quantitative estimate.
- Check the timing. Confirm that the prediction date precedes the first relevant public result.
- Read the outcome source. Use the trial registry, the sponsor's original report or the peer-reviewed trial publication. Note the data cutoff and whether an endpoint is still pending.
- Apply the scoring rule published on the dashboard. Keep statistical trial success, clinical benefit relative to standard of care and commercial interpretation separate.
- Inspect limitations. Check exclusions, missing outcomes and repeated predictions on the same asset.
Why do accuracy and PPV differ?
Accuracy is the proportion of all adjudicated readouts scored correctly. Positive predictive value (PPV) is the proportion of positive predictions that met the positive-outcome definition. Negative predictive value (NPV) is the corresponding measure for negative predictions. Each has its own numerator and denominator.
A pooled percentage does not establish equal performance in every indication, modality or development phase. See the evaluation methodology for what is needed to interpret these measures.
How should a result be cited?
Include BioinvestGPT ApS, the prediction index, the drug and trial identifier (NCT number), the prediction date, the endpoint, the dashboard URL, the outcome-source URL and the date consulted. Keep the prediction and the subsequently observed result distinct.