Updated: 2026-10-22. This is the repository-side status summary for the occupancy work. The GitHub Wiki roadmap still needs a synchronized update; see follow-up issue #205.
- The differential-abundance config offers four per-taxon methods
(
nb_glm,clr_welch,zinb,hurdle_nb) and five compositional models (ancom_bc,aldex2,songbird,multinomial,dirichlet_multinomial); see the differential-abundance section.songbirdandmultinomialare two implementations of the same baseline-category multinomial model. zinbandhurdle_nbare available in the differential-abundance config and UI. They report the count effect and zero/occurrence effect and test group effects jointly. The UI surfaces model cautions; BH adjustment remains on.MetaManifold.Differential.fit_occupancyfits repeated-visit Boolean detection histories (sites × visits × taxa), with group occupancy ψ and a common detection probability p. This is a low-level estimator only.
Occupancy cannot yet be selected in AnalysisConfig or run through the regular analysis endpoint. The estimator does not accept detection covariates/formulas, primer or batch effects, or random effects; it refuses single-visit histories. The ZINB/hurdle models do not yet have user-defined component formulas, model selection diagnostics, or full scale benchmarks. These constraints and the closure/destructive-sampling caveat are detailed in Occupancy model notes.
Issue #205 tracks first-class repeated-detection input and configuration, detection and zero formulas, identifiability validation, diagnostics/reference comparisons, simulation recovery, benchmarks, provenance/schema completion, and updates to the Wiki and user/developer roadmaps. Do not describe the lower-level fitter as a complete occupancy workflow until those acceptance criteria pass.
Issue #18 records the originating feature request and the first implementation slice.