E1 · Source-domain baseline
Real BBBC039 baseline evaluation is retained, including the corrected RGB instance-mask decoder and held-out test evidence.
Research programmeEvidence first
Domain-robust nuclear instance segmentation.
BioNuclei is a reproducible research programme for fluorescence bioimaging. Experimental status below reflects the retained research record; infrastructure status is kept separate from scientific interpretation.
Validation sequence
Evidence accumulated in staged gates.
Each stage has a defined role. Completed stages are described as completed; remaining publication gates are not silently promoted to finished results.
E2/E3 · Cross-domain diagnosis
S-BIAD634/S-BSST265 target data were acquired and profiled; zero-shot BBBC039-to-S-BIAD634 transfer and failure diagnosis are retained.
E4 · Domain robustness
Intensity-domain randomization was executed across the 79-image target pool with retained provenance and a matched image-level comparison against the diagnostic baseline.
E5 · Ablations and baselines
Controlled ablations, conventional baseline comparison, prespecified matched statistics and the aggregate integrity gate are retained.
E6 · Few-shot adaptation
The standardized 1%, 5%, 10% and 25% adaptation protocol was executed and archived. Annotation-efficiency analysis is retained separately and does not imply a monotonic benefit.
E7 · Independent validation
Independent BBBC038 validation was executed and retained. This supports external validation of the frozen workflow, not a claim of universal biological generalization.
Reproducibility
Evidence remains traceable to immutable execution records.
The retained record includes exact workflow runs and artifact digests, a clean-environment reproduction, the permanent BBBC039 checkpoint release, and documented protocol boundaries.
Environment gate
Clean-environment reproduction passed under the retained workflow record. Scientific evidence is not regenerated merely for presentation.
BBBC039 model release
The verified BBBC039 checkpoint is distributed through the repository's permanent GitHub Release and is hash-checked by the public service at startup.
Publication gates
What remains open.
Remaining work concerns release/reporting integrity rather than inventing new scientific evidence.
Final experiment freeze
Freeze the final experiment command and archive all final manifests before paper-level numerical reporting is treated as immutable.
Figure and table audit
Generate final figures and tables from versioned machine-readable outputs and trace every reported number to an evidence artifact.
Literature and claim audit
Complete the final literature/novelty review, narrow unsupported wording, and document limitations and appropriate uncertainty.
Scope boundary
BioNuclei is the scientific substrate.
BioMCP, BioWF, BioSkills and BioFM are separate interoperability or research directions. They are not evidence for the BioNuclei benchmark and are not used to inflate its scientific claims.