Lila wants its clinical-program forecasts judged without hindsight
Lila already reports laboratory discovery and preclinical results. Its translational-biology role asks scientists to assess clinical programmes using mechanistic and human evidence, then score forecasts against real programme progression using only information available at the prediction date. They must also report when a contribution adds nothing. That could test what the assessments contribute beyond a persuasive biological explanation. The vacancy establishes the intended evaluation design, not forecast accuracy or better clinical outcomes.
The constraint
A convincing biological explanation can still fail to predict what happens to a clinical programme.
Another explanation
The work could be retrospective benchmark development rather than forecasts used in live programme decisions. Actual contamination controls, baseline comparisons and incremental predictive value are not demonstrated.
Evidence
- Lila Sciences · Senior/Principal ML Scientist, Translational Biology
scored using only information available at the prediction date, with the evidence boundary enforced against contamination
Context
- Lila's RNA discovery engine
The current company article reports preclinical internal results and explicitly limits their validation. It already establishes a therapeutics direction; it does not describe the prediction-date evaluation protocol in this vacancy. The page displays August 28, 2026; no archived first availability is asserted.
- Lila Solutions
The current solutions page already offers scientific reasoning models and laboratory access for discovery. The remaining increment is a specific advertised evaluation design for clinical-program judgment, not entry into medicine or a new forecasting method.
What would change this reading?
A dated evaluation protocol identifying the information cutoff, programme-progression outcome, leakage checks and comparison baseline, followed by scored results including null contributions. Time-bounded retrospective tests would count as evaluation evidence, not prospective forecasting.
For candidates
The advertised work combines mechanistic reasoning, human biomedical evidence and evaluation design. Demonstrating what a model adds beyond a time-bounded baseline matters as much as explaining its predictions.