Evidoa
AI-native research laboratory.
We build AI systems for scientific discovery: they generate hypotheses, test them adversarially, and we publish what survives — including what dies.
Current research
VCC-2026 — Arc Virtual Cell Challenge
Evidoa is participating in the Virtual Cell Challenge 2026: zero-shot prediction of CRISPRi perturbation effects in unseen cellular contexts. Work in progress covers data engineering, perturbation-grouped cross-validation, official scoring, and model development. No results are claimed here until they are verified and public.
How we work
- Every hypothesis is attacked before it is trusted — independent adversarial review is part of the pipeline, not an afterthought.
- Negative results and dead ends are recorded and reused, not quietly dropped.
- Methods are built for reproducibility: fixed protocols, append-only provenance, official evaluation tooling where it exists.
- Claims are separated by strength: computational evidence, independent verification, prior-art check — and only then a scientific claim.
Preprints
- Model shopping: how answer selection across AI models inflates scientific error rates
- Silent archival of World Development Indicators: data unavailability for retired WDI series through the World Bank API
Datasets & code
Contact
- Website evidoa.com
- Telegram @Evidoa
- Email Enable JavaScript to reveal the contact address.