arrayprobe 0.1.0

Release summary — 0.1.0

What it contains. Two engines behind one driver: the UIOOPS engine, a port of UIOOPS v3.5
and the default for new data, and the SODA-2 engine, the independent second opinion. The earlier
port of UIOOPS v3.1.4, the anchor that reproduces archived UIOOPS products, is the validation
reference and is not part of this package. See the "Choosing an engine" documentation page.

Validation. Each engine is checked against the original it reproduces. The v3.1.4 port, kept outside this
package as the validation anchor, reproduces archived MATLAB products cell by cell — on one flight 33,027 of 33,027 cells, with
15,987,236 particles matching to the particle; the v3.5 and SODA-2 engines match their original
sources executing under Octave and GDL; and SODA-2 beside an independently archived SODA-2 output
accepts 18,017 particles against 17,817 on one flight (figures in
docs/04-validation.md; the
SODA-2 comparison in
docs/developer/SODA2_BESIDE_UIOPS.md).

Citation. SODA-2 is cited as Aaron Bansemer. (2026). abansemer/soda2: v2.26.0 (Version v2.26.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18331418 — the form its author, Aaron Bansemer, provided (2026-10-06); CITATION.cff carries it as a reference.

Corrections on by default (engine 3.5 only; the v3.1.4 validation port never applies them). Detail in
docs/developer/TWO_ENGINES.md.

Correction	Probes	Opt-out
Stray-pixel removal before sizing	every engine-3.5 run	--stray-pixel off or [enhancements] stray_pixel_fix = false
2D-S sized on the 2D-S bin table, not HVPS's	2D-S whose project name falls to v3.5's otherwise branch	--protective off or [enhancements] bin_table_fix = false
Particle-clock offset derived from the data	probetype 1 (CIP/PIP) outside UIOOPS's clock-offset table	--protective off
Zero-sample-volume seconds written NaN; duplicated seconds moved into the gap	probetype 2 (2D-S/HVPS)	--protective off

Corrections on by default for probes with no archived product. On CIP/PIP, 2DC/2DP and
Fast2DC the driver also writes each of these as a separate enh product, recorded in the .call
stamp; --protective off skips them, and the 2D-S/HVPS lineage that reproduces archived products
never gets them. Detail in enhancements_uiops/docs/CORRECTIONS.md.

Correction	Probes
Glitched header times dropped before they cascade into a 24-hour time error	CIP/PIP
particle_time recovered from its 8-bit integer collapse	2DC/2DP, Fast2DC
Duplicated airspeed seconds collapsed onto a regular one-second grid	CIP/PIP, 2DC/2DP, Fast2DC

Limits of this release.

• The engine and package rename follows in a later release; import names are unchanged here.
• Re-validation of the archived reference flights on current code covers 13 of 14, not counting the
  product of record.
• The held-out synthetic-dataset test is pending.
• Known gaps needing outside input: PECAN runs need a user-supplied parameter file; 2DP needs a
  user-supplied tas_max; SODA-2's ERODE border rule is unconfirmed under GDL; HVPS counts
  match the archived product only under a per-flight inter-arrival call inferred by comparison
  (within 0.5 % on four of the six 2016 flights and on 2017-08-26; 2016-09-10, 2016-09-25 and
  2017-08-13 open), while one fixed call gives 1.24 and 1.09 times the archived count on two
  flights; the archive's per-flight configuration is not recorded, so HVPS counts are not
  quoted as validated; CIP, PIP, 2DC and 2DP have no raw-plus-product reference pair, so they are
  checked at code level only. See the "Known gaps" documentation page.

What changed. This is the first release of the package, so there is no earlier release to
compare against: everything above is new in 0.1.0.

Does it change my numbers? Not applicable: there are no earlier released numbers for this
release to change.

Added

• arrayprobe inventory lists every raw input with its size and flags empty or truncated files.
  The driver runs the same check before decoding: an empty file is refused unless
  --accept-flagged is given, a file ending in a partial record is decoded with a warning, and
  the run's .call stamp records what was found.
• arrayprobe reproduce-report writes a report of what a run used and produced, for checking a
  result against another run.
• arrayprobe check-counts and the count_closure module show, second by second, that the binned
  counts add up to the accepted particles, with every excluded particle attributed to its reason.
  The comparison against a probe's own reported total is not available yet.
• arrayprobe attributed --layout particles writes one Parquet row per particle (needs the
  [parquet] extra); the default per-dimension layout is unchanged.
• Triage: an optional habit label per particle, from the UIOOPS v3.5 habit vocabulary, and the
  agreement between two reviewers as Cohen's kappa.
• Documentation: an "Add a campaign" page with a worked example whose output is checked by a
  test; inventory, particle-layout, reproduce-report and count-closure sections; the opt-ins
  notebook shows --accept-flagged; the installation page explains how to verify a download.
• Release workflow: build-provenance attestations (Sigstore) for every distribution file,
  checkable with gh attestation verify.