Changes in version 0.6.1                        

Documentation

  - A vignette carries one task end to end on synthetic or bundled data,
    offline, and is built and run by R CMD check.
  - Runnable @examples added or corrected across 26 help pages. Each
    runs offline in seconds, writes nothing outside tempdir(), and is
    executed by R CMD check; anything needing a device, a download or an
    optional backend is fenced with the reason stated.
  - The README's quick start runs as written: it attaches the package,
    builds its own inputs, and uses only hard dependencies.

                        Changes in version 0.6.0                        

First-level cortical-activation GLM.

  - nirsActivationGLM(): fits every haemoglobin channel against
    HRF-convolved task regressors (per-condition boxcars over the
    stimulus blocks convolved with a canonical double-gamma HRF), a
    discrete-cosine drift basis, and optional short-separation nuisance
    regressors, returning per-channel/per-condition activation beta, SE
    and t. Supports a temporal-derivative basis and per-channel AR(1)
    prewhitening (fNIRS residuals are strongly autocorrelated, so naive
    OLS t-statistics are inflated).
  - nirsActivationContrast(): linear contrasts over the condition betas
    (estimate, SE, t, p per channel).

                        Changes in version 0.5.1                        

  - Fixed pruneChannels() / quality-contract validation on R >= 4.4.0:
    R 4.4 changed is.atomic(NULL) from TRUE to FALSE, so
    .nirs_quality_plain() wrongly rejected governed quality results
    carrying NULL optional-parameter sentinels (e.g. step_seconds =
    NULL) as 'stale or malformed'. NULL is now handled explicitly as a
    plain leaf (its pre-4.4 behaviour).
  - Made the MBLL reference-fixture self-consistency check portable: the
    stored intensity is recomputed from optical density via exp() (a
    transcendental whose last bit is toolchain-dependent), so the exact
    (tolerance = 0) compare is now tolerance = 1e-12 (~2300x above the
    measured cross-platform noise). File-hash governance is unaffected.

                        Changes in version 0.5.0                        

  - Add source-bound scalp-coupling, PHOEBE-style peak-power, and
    spectral SNR quality metrics with auditable mark/drop channel
    pruning.
  - Add a synchronous governed HbO neurofeedback controller that
    publishes atomic multi-target updates through PhysioStream
    biofeedback scopes.

                        Changes in version 0.4.0                        

  - Add source-bound short-channel identification, centred nearest/all
    short-separation regression, explicit physiology-band extraction,
    and GLM-ready nuisance-design matrices.

                        Changes in version 0.3.0                        

  - Add governed motion-artifact detection, TDDR, shift-invariant db2
    wavelet correction, and cubic-spline correction.

                        Changes in version 0.2.0                        

  - Add governed natural-log optical density, differential pathlength,
    extinction-coefficient, and modified Beer-Lambert conversions.

                        Changes in version 0.1.0                        

  - Add strict SNIRF v1.1 continuous-wave read/write support.
  - Add governed measurement-list and source-detector distance helpers.