Capability matrix
The target is broad field utility, not a claim that every possible optical experiment belongs in one package.
| Scientific task | Status | Current route or required work |
|---|---|---|
| Tabular, TDT, NWB and selected public-data ingestion | Supported | Explicit schemas and channel roles |
| Doric, Neurophotometrics and pyPhotometry native import | Experimental | Typed native readers with format-specific schemas, inspection evidence, digital events and synthetic writer-contract fixtures |
| Sampling, gaps, dropout, saturation-like and flat-step QC | Supported | Outcome-blind preflight and report panels |
| Reference-channel correction | Supported | OLS and IRLS; subtraction and division remain distinct |
| Signal-only bleaching correction | Experimental | AsLS and double exponential need broader controls |
| Processing order, normalization and negative-control contract | Partial | Provenance exists; add operation-order validation, local/null-relative normalization, and explicit control roles |
| Sensor/isobestic-aware validity | Experimental | Open versioned profiles; wavelength, role, unit, detector, fiber, saturation, range, reference-coupling, lag and event-response evidence; cannot prove inertness or concentration |
| Hemodynamic and spectral crosstalk correction | Experimental | Independently calibrated or explicitly user-declared linear mixing matrix; rank/conditioning refusal, missing-pattern evidence, residuals, and optional channel holdouts; biological validation remains open |
| Multi-color spectral unmixing | Experimental | Typed component/channel/wavelength metadata and the same identifiable linear contract; no blind, nonlinear, or time-varying unmixing |
| Event alignment and window summaries | Supported | Retains events, sessions and animals |
| Interval/bout alignment and normalized progress | Experimental | Typed intervals, explicit onset/offset, duration and progress kernels; ordered filter/merge/split/context/overlap policies retain a complete source-lineage ledger and evidence fingerprint |
| Animal-level scalar/vector inference | Supported/experimental | Auditable animal-estimate ledger, paired and independent two-level contrasts, t intervals, animal bootstrap, standardized effects, support, and leave-one-animal-out influence; typed materializers now cover peri-event curves, spontaneous transients, state-band power, and multi-signal association while preserving domain denominators |
| Group-by-condition animal interactions | Experimental | Explicit two-group × two-condition difference in differences; complete animal-condition ledger, missing-cell exclusions, simultaneous bands, and animal influence |
| Whole-waveform intervals | Supported | Paired or independent animal-level pointwise and simultaneous bands with retained session-to-animal evidence |
| Published waveform bootstrap/permutation parity | Partial | Related animal-level resampling exists; add an explicit literature reproduction and estimand comparison |
| Trial-level functional mixed models | Planned | fastFMM bridge and numerical-parity fixtures |
| Behavioral/event-kernel GLMs | Experimental | Full-FIR, raised-cosine, history, duration/progress kernels, grouped validation, pointwise jackknife sensitivity, model multiverses, and paired family drops; the first simultaneous-band calibration failed its normalized-progress gate, so that band is opt-in only |
| Longitudinal learning trajectories | Interoperable / experimental | Validated neural-summary handoff to Unspool; public cross-package benchmark still required |
| Across-session photometry comparability | Experimental | Typed outcome-blind identity, preprocessing, coverage, baseline, reference-coupling and sampling diagnostics with warning/refusal states bound into the Unspool handoff |
| Single-signal autocorrelation, PSD and spectrograms | Experimental | Gap-separated autocorrelation pairs and window-weighted Welch PSD; complete-window spectrograms retain edge and missing-data evidence; exact-axis PSD and autocorrelation curves now support animal-level pointwise and simultaneous inference |
| State/epoch-conditioned spectral summaries | Experimental | User-supplied, non-overlapping epochs remain separate partitions; paired band-power inference aggregates sessions within animals |
| Multiscale long-duration summaries | Experimental | Named physical-time scales; hard gap/epoch boundaries; sample- and time-weighted metrics; complete window ledger; session-to-animal contrasts without biological “tonic/phasic” claims |
| Spontaneous transient detection | Experimental | Separate GuPPY-, PASTa-, and prominence-compatible candidate families; gap-aware detector evidence retained |
| Spontaneous transient quantification | Experimental | Non-z-scored amplitude/half-width/AUC, compound groups, exposure-adjusted summaries, detector-bound frozen baseline/control score gates, native gap-bounded cutouts and optional waveform-QC refusal |
| Spontaneous transient animal-level inference | Experimental | Paired/independent animal bootstrap and randomization for exposure-adjusted rate, amplitude, width, and AUC contrasts |
| Multi-site/multi-color coupling | Experimental | Explicit pair/site/sensor/optical metadata; joint-validity lagged association; declared event/behavior residualization; crosstalk review flags; blocked within-session nulls and animal-level contrasts |
| Coherence and phase analysis | Experimental | Window-weighted joint cross-spectra, state conditioning, explicit phase convention and band summaries; complete coherence curves now support exact-axis animal-level pointwise and simultaneous inference; multitaper and population phase uncertainty remain gaps |
| Dense spatial multi-fiber analysis | Experimental | Shared coordinate space/unit; complete gap-aware edge ledger; Fisher-z distance bins; within-session node-label null; edge→session→mouse paired or independent inference; constrained spatial nulls and raw-system validation remain gaps |
| Optogenetic-stimulation artifact handling | Experimental | Prospective time-only pulse masks compose with existing validity; separate recovery, censoring, detector-rail and negative-control diagnostics never adapt the mask |
| Sensor-kinetic modeling and deconvolution | Experimental / caution | Versioned calibrated or assumed parametric/sampled linear kernels; gap-bounded forward prediction; sampling, duration, transfer and regularization gates; conditional regularized recovery with reconstruction evidence; nonlinear models and latent uncertainty remain gaps |
| Behavior/pose/state discovery | Experimental interoperability boundary | Typed file/in-memory adapters, matched-pulse clock synchronization, current DeepLabCut/Keypoint-MoSeq/SLEAP/BORIS fixtures, and native loss-aware ndx-pose 0.3 inspection plus 2D/3D NWB round trip; a real camera-to-photometry clock fixture remains a gap; Unspool owns longitudinal models |
| Robustness multiverses | Supported | Named alternatives, compatibility, failures and provenance |
| NWB evidence and archival publication | Supported | Export, verification, signing and draft DOI handoff |
“Gap” means the question is common enough to deserve either implementation or a clear interoperability route. It does not mean the safest response is necessarily to add another algorithm; some capabilities require new validation data first.