Glossary

In short: one fixed meaning for cell, bin, second, boundary, non-zero and counts, plus the four nested denominators an agreement figure can be computed over. A percentage means nothing until you know which one it used.

See also Validation, which uses every term defined here.

This page is the single definition. The paper's number registry carries the same definitions under validation.vocabulary, and the manuscript restates them; if the three disagree, this page and the registry are right.

Two terms are easy to confuse in a way that changes a headline: the share of cells where both files record zero and the share of particles in the excluded bins are different quantities over different populations, and both land near the same number.

The grid

A flight product is a two-dimensional grid of numbers, plus its axes.

Term	Definition
second	One row. One second of flight time — the product's native time step. A flight has ~30,000 of them.
bin	One column. One particle-size interval. The 2D-S grid has 128, smallest first.
cell	One row–column pair: a single number. One second, one size bin. A flight grid holds ~3.8 million cells.
counts	The `Counts` variable: whole particles recorded in that cell. An integer — which is why agreement can be exact equality rather than a tolerance.
size distribution	The `Size Distribution cm-3` variable: the same grid, as a concentration. Both are compared; both give the same headline.

Which cells get counted

Four populations, each a strict subset of the one before. Numbers are from one 2D-S flight of a field campaign (20 September 2016) in the validation set.

Population	Cells	What it is
grid	3,846,656	every second × every bin
both-present	3,202,944	both files hold a value rather than fill (`NaN`)
non-zero	55,017	both-present, and at least one side is non-zero
used-bin non-zero	13,039	the same, restricted to bins 6 and above

both-present is the widest honest denominator, but 95.8–98.8 % of it is zero equals zero — agreement about absence. A percentage over it looks excellent almost regardless of what the code does.
used-bin non-zero is the headline denominator: only cells where at least one file recorded a particle, in the bins the science uses. It is ~200× smaller and correspondingly harder to satisfy.

The two are never mixed, and both are always printed. A check prints both every time and names what the looser one is made of.

The boundary

Bins 1–5 are excluded from every headline figure. They sit at the diode-pitch noise floor and the science does not use them — the field's convention, not a threshold tuned here. Boundary, unqualified, always means the line between bin 5 and bin 6.

Bins 1–5 also hold 99.4–99.8 % of all particles by count, which is what makes the headline denominator small. That is not the ~98 % zero-equals-zero share above; the two are different populations, each measured on its own.

Quantity	Value	Population
share of particles in bins 1–5	99.4–99.8 %	all particles in the archived product
share of cells that are zero==zero	95.8–98.8 %	both-present cells

Agreement

Term	Definition
agreeing cell	Both files hold the same value. Exact equality, not within a tolerance.
differing cell	Both files hold a value and the values differ.
uncovered second	A second present in one file and absent from the other. Invisible to any cell-wise percentage, so coverage is always reported separately.
used-bin residual	The differing cells that survive into bins 6+. Across the six validated 2D-S flights of the validation set: 48 cells, of which none has both sides non-zero — the two implementations differ only on whether a second has particles, never on how many. This is a 2D-S result and does not transfer: on the HVPS most differing cells are the other kind, both sides non-zero and disagreeing, and bins 6+ is not a usable HVPS range, since it puts almost nothing there.

Files

Term	Definition
archived product	The archived MATLAB output the published science was written from. For the validated 2D-S flights these are `base.combined.H.2DS_2.cdf`. All 29 held products carry the `_2` suffix*, hardcoded in the validation tool so nothing else can be compared against by accident.
`proc` file	Per-particle output of stage 3 (framing): one row per particle, ~40 variables.
`sd` file	Per-second size distribution, stage 4: one file per raw file.
combined product	The combine stage: one flight's `sd` files stitched into the single product a paper cites, and what validation compares.
`.call` sidecar	A text file beside every output recording exactly how the stage was invoked, because no UIOOPS output records which call produced it and two reachable calls give a 58 % difference in accepted particles.

Pipeline stages and test terms

Term	Definition
golden	A stored file of expected output, written by running the original's own code on a fixed input — the UIOOPS `.m` sources under GNU Octave, or SODA-2's IDL `.pro` sources under GDL — that a test compares the port's output against. Most are committed and run on a fresh clone; goldens cut from real campaign data are not distributed, so the checks that need them report as skipped. Octave and GDL are needed only to regenerate goldens (Installation).
`iQC`	A v3.5 `imgProc_sm` switch, default `0`. `iQC=1` runs a denoise pass on each particle image (`imgProc_sm.m:801-846`, ported as `arrayprobe_uiops.qc.qc_image`) before any measurement is taken from it, and adds the `subparticle_num` and `subparticle_mindist` variables to the `proc` file. The value is recorded in the `.call` sidecar; the v3.5 engine refuses `--chunks` when it is `1`.
S2	The decode stage (`2` in the `stages` setting of `arrayprobe run`): turns a raw probe file into a decoded image file plus its `.tas.csv` true-air-speed sidecar. The driver's log lines for it begin `S2`.
S3	The framing stage (`3` in `stages`): frames particles out of the decoded images and writes the per-particle `proc` file.
S4	The size-distribution stage (`4` in `stages`): computes one-second size distributions and writes the `sd` file. The combine step that follows is `c` in `stages` and has no S number.

Evidence

Term	Definition
archived product	The strongest layer: output written by the original MATLAB. Limited to the configurations one campaign used.
differential oracle	The original's own `.m` files run unmodified under Octave. Reaches configurations no product covers, including inputs that crash the original.
documented semantics	The weakest layer: a reading of the source plus a pinned test, labelled as such wherever it carries a claim, because both sides come from the same reading.
measured / derived / read	How a finding was established. Measured: run on real data; derived: computed from measured quantities; read: established by reading the source and flagged for the reader's own check, never asserted.

Probe and microphysics terms

Definitions as used in How a size distribution is built.

Term	Definition
optical array probe	An airborne instrument that records particle shadows on a row of photodiodes across the airflow; its raw file holds images, not sizes or concentrations.
diode	One photodiode in the probe's array; the number of diodes times the resolution is the width the array images (Probes).
slice	One clock of the array: the record of which diodes were shadowed at that instant. A particle passing through becomes a stack of slices.
habit	A particle's shape class, assigned after Holroyd (1987) and written per particle to the `proc` file.
reject code	The per-particle code that decides whether a particle is accepted into the size distribution.
inter-arrival time	The time between one particle and the next; the shattering test and the live time are both built from it.
shattering	An ice crystal striking the probe tips breaks into fragments that arrive in a burst; particles whose inter-arrival time is shorter than a threshold are removed as fragments.
sample area	The effective area over which a particle of a given size is counted, set by `sa_method`: centre-in, entire-in, or the reconstruction correction that widens the array.
true air speed (TAS)	The aircraft's speed through the air, from the probe's own record in the decode's `.tas.csv` sidecar.
live time	The part of each second the probe could record: the sum of inter-arrival times less the time lost to overloads; written as `sum_IntArr`.
sample volume	Sample area × true air speed × live time, per bin; it turns a count into a concentration and is written as `sample_vol`.