# UniPert feasibility protocol

## Current executed scope

- Formal Cell paper and PubMed record verified.
- Official public repository audited at commit `2f5d46930dcbdeb92073a13e898abe6e363e679a`.
- Code licence recorded as GPL-3.0; article access and patent notice remain separate rights facts.
- 21/21 governed candidates map to reviewed human UniProt primary accessions.
- Encoder runs: 0. Embedding results: 0. Predictive benchmarks: 0. DMD functional outcomes: 0.

## Required next execution order

1. U1: freeze the official environment and compute candidate embeddings with checksums.
2. U2: run cause-space sanity checks and known-relation retrieval against ESM2-only and random embeddings.
3. U3: run the G2 held-out-family/pathway test against native ridge, ESM2-only, nearest-neighbour and random-embedding baselines.
4. U4: run public gene-to-compound transfer with scaffold holdout and a data-volume learning curve.
5. Publish exposure labels, leakage checks, intervals, failures and null results.

## Adoption rule

Adopt UniPert as a feature only if it improves a predeclared held-out task over simple baselines with uncertainty, survives exposure and leakage audit, and does not degrade worst-group performance.

## Claim boundary

This release verifies sources, implementation limits and 21-candidate identifier readiness. It publishes zero UniPert embeddings, zero UniPert predictive benchmarks and zero DMD functional outcomes.