OmniPredict · Predictive biomedicine
A five-module biomedical prediction pipeline with a rigor contract: every constant must be classified as derived, provisional, or blocked, and no calibration advances without admissible evidence.
What it is
OmniPredict explores biomedical prediction through a five-stage computational pipeline tested with synthetic, non-clinical inputs. Its distinction is not just the model — it is the process: the system audits its own coefficients, declares which ones are provisional, and blocks promotion when traceable derivation or measurement is missing.
Formal coefficient derivation is part of the weekly scientific cycle, but the current readiness contract stops calibration because admissible biophysical data are absent.
Current stage: computational integration tested locally and structural readiness defined; none of the four components yet has the data package required for calibration, validation, or holdout reservation.
Public objective: build biomedical prediction where every parameter can be traced to evidence and biosafety constraints.
Next verifiable milestone: obtain independent, traceable, partitioned biophysical data for all four components before comparing or calibrating coefficients.
Lab updates
-
Readiness NO-GO
August 2026
The instrument is ready; admissible data do not exist yet
The preflight verified contract, provenance, and structure, but classified all four components as not ready because raw biophysical data with controls, uncertainty, and independent partitions are missing. A scientific NO-GO prevents a simulation from becoming a calibration.
-
Verifiable software
July 2026 · week 1
The pipeline removed an unbacked parameter — and stayed verified
The governed review detected a filtering constant with no scientific derivation backing it. It was removed from the system, the affected module was rebuilt, and the verification suite closed 5/5 green. This is the project's governance pattern in operation: science enters the code only through the authorized channel, or it doesn't enter.
This evidence verifies software integration and rules on synthetic inputs; it does not establish diagnosis, treatment, efficacy, clinical safety, or a biological mechanism. Human review adjudicates scope. The internal mechanism is never published.