FCS (Flow Cytometry Standard)

FCS is the open file format that flow cytometers and their software write. OpenReadout returns each data set as a table of events with one column per parameter, plus the keywords and instrument metadata; compensation, transforms and gates are applied on request.

Derived from the public ISAC specification “Data File Standard for Flow Cytometry, Version FCS 3.1 — Normative Reference” (2009, CC BY-SA 3.0) and checked against public corpus files (FCS 2.0, 3.0, 3.1 from many instrument families). FlowIO (BSD-3-Clause) and fcsparser (MIT) were read as prior art and are used as reference readers. See docs/provenance/fcs.md.

FCS is an open standard, so unlike the microscopy formats most of this page restates a specification; the corpus notes record where real files depart from it. Section numbers (§) refer to the FCS 3.1 normative reference.

A file is one or more data sets. Each data set has a HEADER, a primary TEXT segment of keyword-value pairs, a DATA segment of events, and optionally a supplemental TEXT, an ANALYSIS segment and implementor-defined OTHER segments. $NEXTDATA links data sets. OpenReadout exposes each data set as one table: one row per event, one column per parameter.

FCS 3.2

FCS3.2 data sets (Spidlen et al. 2021) are read like 3.1 with these differences: $PnDATATYPE gives a measurement its own data type (I, F or D) when it differs from $DATATYPE, and widths, decoding, dtype and bit masks follow it per column; $MODE may be absent (list mode); $BEGINSTEXT/$ENDSTEXT/$BEGINANALYSIS/$ENDANALYSIS may be absent; $BEGINDATETIME/$ENDDATETIME (ISO 8601, with a zone) become acquisition_start/acquisition_end in place of $DATE/$BTIM/$ETIM; $CARRIERID, $CARRIERTYPE, $LOCATIONID, $FLOWRATE, $UNSTAINEDCENTERS, $UNSTAINEDINFO are copied to carrier_id, carrier_type, location_id, flow_rate, unstained_centers, unstained_info. Corpus: zenodo19221995-facsdiscover-zam36 (BD FACSDiscover S8, 440 measurements, delimiter CR).

HEADER (§3.1)

All offsets are ASCII decimal, right-justified in 8 bytes, relative to the first byte of the data set, and inclusive (the end offset is the last byte of the segment).

bytes	our name	content
0–5	`version`	`FCS2.0`, `FCS3.0`, `FCS3.1` or `FCS3.2`
6–9	–	spaces
10–17 / 18–25	`text`	primary TEXT begin / end
26–33 / 34–41	`data`	DATA begin / end
42–49 / 50–57	`analysis`	ANALYSIS begin / end (zeros or blanks when absent)
58…	`other`	optional OTHER segment begin/end pairs, 16 bytes each, until the first segment starts

When a segment lies (partly) beyond byte 99,999,999 its HEADER offsets are 0 and the real offsets are only in TEXT ($BEGINDATA/$ENDDATA, $BEGINANALYSIS/$ENDANALYSIS, $BEGINSTEXT/$ENDSTEXT). Blank fields (all spaces) are read as 0 and reported as blank_header_offset (seen in fcsparser-fake-large, flowio-coulter-lmd). HEADER_LEN = 58; header_len is 58 plus 16 per OTHER pair actually present (74 in fcsparser-cyflow-cube-8).

TEXT (§3.2)

The first byte of the primary TEXT is the delimiter (ASCII 1–126). Seen in the corpus: /, \, |, form feed 0x0C (BD FACSDiva, Cytek SpectroFlo), record separator 0x1E (Beckman Coulter Navios).
Pairs are keyword DELIM value DELIM. A delimiter inside a keyword or value is doubled. Keywords and values may not be empty, so a doubled delimiter is always an escape — except as the last two bytes of a segment, where it can only be a terminator plus an empty final value (flowio-data1 ends …Analysis Doc.\\).
Keywords are case-insensitive (get ignores case); FCS-defined keywords start with $. Values are UTF-8 in 3.1; values that are not valid UTF-8 (older Mac writers, e.g. CellQuest Pro\xAA) are read byte-for-byte as Latin-1 and reported as non_utf8_value.
Numeric values must not be padded (§3.2.17) but often are ($TOT/83411 / in FACSDiva exports, $BEGINDATA/ 3446/ in Guava) or carry leading zeros ($BEGINDATA/0000000000004161/). We trim; check reports padding as padded_numeric_value (info).
Trailing spaces or NULs after the final delimiter are ignored (Miltenyi supplemental TEXT ends with one space).

Supplemental TEXT and ANALYSIS

$BEGINSTEXT/$ENDSTEXT locate the supplemental TEXT; it uses the primary delimiter and may or may not start with it (Miltenyi: starts with /; flowio-m0-wm278-s1-zero-gain: starts directly with SORTSTATS|, and sits after DATA). Seen oddities: the range equals the primary TEXT (flowio-b01-kc-a-w-91-us, not re-parsed), and a ZIP archive instead of keywords (fcsparser-cyflow-cube-8, signature PK\x03\x04; the same file’s TEXT says P$CFGTYPE/ZIP/). ANALYSIS (§3.4) has the TEXT syntax; its location comes from $BEGINANALYSIS/$ENDANALYSIS, else the HEADER. Keyword lookups for normalized fields search the primary TEXT, then the supplemental TEXT.

Required keywords (§3.2.18) and how we use them

keyword	our field	use
`$BYTEORD`	`byte_order`	`1,2,3,4` → `LittleEndian`, `4,3,2,1` → `BigEndian`; two-byte `1,2` / `2,1` appear in Beckman Coulter FCS 2.0 files; any other permutation → `Mixed` (reported, not decoded)
`$DATATYPE`	`data_type`	`I` `Integer`, `F` `Float`, `D` `Double`, `A` `Ascii`
`$MODE`	`mode`	`L` `List` (decoded); `C` `Correlated`, `U` `Uncorrelated` histograms (deprecated; described, reads exit 6)
`$PAR`	`parameters`	number of `$Pn*` sets read
`$TOT`	`event_count`	rows
`$PnB`	`width`	`Fixed(n)`: bits for I/F/D, characters for A; `FreeFormat` for `*`
`$PnN`	`short_name`	column `name`
`$PnR`	`range`, `range_text`	column `range`; the I bit mask
`$PnE`	`amplification`	`[decades, offset]` in column `extra.amplification`
`$BEGINDATA`, `$ENDDATA`	`data_range`	DATA location (see below)
`$BEGINSTEXT`, `$ENDSTEXT`	`supplemental`	supplemental TEXT location
`$BEGINANALYSIS`, `$ENDANALYSIS`	`analysis`	ANALYSIS location
`$NEXTDATA`	`next_data`	offset of the next data set relative to the start of this one; 0 = last

Optional parameter keywords copied into column extra: $PnS → label; $PnG → gain; $PnV → detector_voltage (volts); $PnL → excitation_wavelength_nm (list; 3.1 allows several); $PnO → excitation_power_mw; $PnF → filter; $PnT → detector_type; $PnP → percent_emitted; $PnD → display_scale (verbatim); $PnCALIBRATION → calibration (verbatim). Column extra.bits is $PnB, extra.range_keyword is $PnR as written, and extra.bit_mask appears for I data when the mask is narrower than the field.

Data-set keywords copied into table extra: $CYT/$CYTSN → instrument.model/serial_number; $SYS → system; $DATE → acquisition_date; $DATE+$BTIM/$ETIM → acquisition_start/acquisition_end; $FIL → file_name (also the table name); $SRC → source; $EXP → experimenter; $OP → operator; $INST → institution; $SMNO → specimen; $CELLS → cells; $PROJ → project; $COM → comment; $PLATEID/$PLATENAME/$WELLID → plate_id/plate_name/well_id; $ORIGINALITY; $LAST_MODIFIER; $LAST_MODIFIED → last_modified; $VOL → volume_nl; $TIMESTEP → timestep_s; $TR → trigger {parameter, threshold}; $LOST → events_lost; $ABRT → events_aborted. Also fcs_version, data_set_offset, datatype, byte_order, mode, event_width_bytes, supplemental_text, analysis_segment, spillover, software, vendor_keywords.

Dates and times

$DATE: dd-mmm-yyyy (spec). Also seen: dd-Mmm-yy (22-Sep-13, FCS 2.0 CellQuest; two-digit years < 70 → 20yy, else 19yy — inferred), yyyy-Mmm-dd (Miltenyi MACSQuantify), mixed-case months, trailing spaces.
$BTIM/$ETIM: hh:mm:ss.cc (3.1, hundredths), hh:mm:ss:tt (2.0/3.0, sixtieths per 3.1 Appendix B, converted to hundredths: 17:29:39:51 → 17:29:39.85), hh:mm:ss. A fourth field with three digits (09:42:05:509, CyFlow) is not a sixtieth; it is dropped rather than guessed.
Output is ISO-8601 without a time zone (FCS records none); info adds a note saying so (book/src/guides/metadata.md). acquisition_end rolls to the next day when $ETIM < $BTIM.

Spillover (`spillover`)

$SPILLOVER (3.1, §3.2.20): n,name1,…,namen,s11,s12,…,snn, row-major, sij = spillover from parameter i into j; names are $PnN. $SPILL (FlowKit’s compensation example) and SPILL (BD FACSDiva and others; not FCS keywords) have the same layout; they are used only when every name in them is a $PnN of the same data set (true for the DiVa files; false for the Stratedigm files, whose SPILL names were redacted). $COMP (FCS 3.0) is exposed as a matrix with parameters empty when the value lists no names; no claim is made about whether it is a spillover or a compensation matrix, and it is never applied. info reports the matrix; table --compensate and analyze gate apply it (docs/formats/flowjo-wsp.md).

Scale values, compensation, transforms, gates

read_table and export return raw DATA values. openreadout table FILE --compensate [--transform …] [--workspace W.wsp|--gatingml G.xml --population PATH] and openreadout analyze gate first convert them to scale values (FCS 3.1 §3.2.19–20): $PnE f1,f2 with f1 > 0 gives 10^(f1·x/$PnR)·f2 (f2 = 0 read as 1); a $PnG other than 0 or 1 divides; the Time parameter is multiplied by $TIMESTEP and never divided by a gain (as FlowIO does). Then compensation, transforms and gates as described in docs/formats/flowjo-wsp.md; table records every step in processing.

Instrument families (`platform`) and parameter roles

tables[].extra.platform names the family that wrote the data set, recognised from keywords the file carries (provenance: Source::Inferred), and adds what that family’s keywords mean:

`family`	recognised by	corpus files	extra fields
`bd-spectral`	`CREATOR` starts `BD FACSChorus`, or `BDSPECTRAL UNMIXED` present	`zenodo19221995-facsdiscover-zam36`	`unmixing` {`stored`, `parameters` (from `BDSPECTRAL UNMIXED`), `method` (`BDSPECTRAL UNMIXING METHOD`), `applied` (`BDSPECTRAL APPLY`)}, `detector_count` (the `SPILL` detectors), `imaging_features` (`$PnFEATURE` values other than Area/Height/Width), `cytometer_configuration`, `lasers`; columns: `unmixed_fluorescence`, `imaging_feature`, `spectral_detector`
`bd-facsdiva`	`CREATOR` starts `BD FACSDiva`	`fcsparser-facs-diva`, `-fortessa-a01`, `-hts-lsr-ii-d06`, `flowio-100715`	`lasers` (`LASERnNAME`, `LASERnDELAY` → `delay`, `LASERnASF` → `area_scaling_factor`), `experiment` (`EXPERIMENT NAME`), `tube` (`TUBE NAME`), `cytometer_configuration`, `cst_setup_status`, `compensation_applied_in_acquisition` (`APPLY COMPENSATION`)
`cytek-spectral`	`CREATOR` starts `SpectroFlo`, or `$CYT` `Aurora` / `Northern Lights` / `NL-…`	`zenodo17457137-aurora-beads`, `fcsparser-cytek-nl-2000-header`	`lasers`, `detector_count`, `detectors_by_laser` (full-spectrum detectors `<laser><n>-A` with laser prefixes `UV`, `V`, `B`, `YG`, `R` named by the file’s `LASERnNAME`), `unmixing` (`stored: false` when `$PnTYPE` says `Raw_Fluorescence` or `$SPILLOVER` is the identity; `true` when no detector-named parameters are present), `user_setting`, `group`, `tube`
`beckman-cytoflex`	`CYTEXPERTFIL` present, or `$CYT` starts `CytoFLEX`	`zenodo18439538-cytoflex`	`software` `CytExpert`, `tube` (`TBNM`), `group` (`CGNM`), `plate_number` (`PLTNO`), `compensation_channels` (`COMPCHH`)
`sony-spectral`	`$CYT` `SA3800`, `ID7000`, `SP6800`	`zenodo7971252-sony-sa3800`	`log_display_parameters` (`$PnD/Logarithmic,…/`)
`sony-sorter`	`$CYT` `SH800`, `MA900`	–	–
`mass-cytometry`	`$CYT` containing the word `CYTOF`, `DVSSCIENCES`, `FLUIDIGM` or `HELIOS`, or three or more metal-tag channels	`zenodo10510047-cytof-mouse`	`mass_channel_count`, `isotopes` (`142Nd`), `markers_by_isotope` (from `$PnS` `142Nd_CD19` → `CD19`), `note`

Every platform also has vendor, technology (conventional, spectral, mass), model ($CYT) and software (CREATOR). Per column, extra.channel_kind is time, scatter (FSC…, SSC…), fluorescence, spectral_detector (Cytek detectors, with laser), mass (<Element><mass>Di/Dd, element checked against the periodic table, mass 75–209; with metal_tag {element, mass, isotope} and marker), background (BCKG…), event_length (Event_length), gaussian_parameter (Center, Offset, Width, Residual) or other; extra.measure is area/height/width from the -A/-H/-W suffix; extra.parameter_type is $PnTYPE (FCS 3.2; Raw_Fluorescence, Time, … in the Aurora file), and the other FCS 3.2 measurement keywords are copied as detector ($PnDET), feature ($PnFEATURE), dye ($PnTAG), analyte ($PnANALYTE) and datatype ($PnDATATYPE, when given). info --view explain describes spectral and mass-cytometry files in these terms and the experiment model carries cytometry_technology, spectral_detectors and mass_channels method parameters and the vendor as instrument manufacturer.

DATA (§3.3)

Locating DATA

Candidates: the TEXT pair ($BEGINDATA, $ENDDATA, unless both 0) and the HEADER pair (unless both 0). FCS 3.x tries TEXT first; FCS 2.0 tries the HEADER first.
The first candidate whose length (end − begin + 1) equals $TOT × event width, or is one byte longer, wins; otherwise the first candidate.
Differing pairs → offset_discrepancy (warning). In flowio-data-start-offset-discrepancy the HEADER says 5555–6188 (634 bytes) and TEXT 6081–6188 (108 bytes = 2 events × 54 bytes): TEXT wins. In -stop- the HEADER end lies past the end of the file.
Length one byte longer than needed → data_end_off_by_one (warning; Miltenyi, Coulter). Shorter → data_length_mismatch (error). Past end of file → truncated (error, exit 4).

Decoding (list mode)

Events are stored one after another; within an event, parameters in order 1…$PAR, each $PnB wide.

I — unsigned integers of $PnB / 8 bytes (any whole number of bytes up to 8; 8, 16, 24, 32 seen), in $BYTEORD order, then masked: mask = next_power_of_two(ceil($PnR)) − 1 when that is narrower than the field, else all ones (§3.3; $PnR/1024/ → 1023; $PnR/11209599/ on 32 bits → 2²⁴−1; $PnR/4294967296/ on 32 bits → no mask). $PnB not a multiple of 8 (bit-packed) → unsupported (exit 6).
F — IEEE-754 binary32 ($PnB must be 32); D — binary64 ($PnB 64). No mask; values may be negative or exceed $PnR.
A — fixed width: $PnB characters per value, parsed as a decimal number (blank field → 0). Free format ($PnB/*/): values separated by space, tab, comma, CR or LF, runs of separators counting as one. Deprecated in 3.1; no corpus file uses it (unit tests only).

read_table returns values as f64 (lossless for every width ≤ 32 bits and for F/D), raw: not compensated and not scaled by $PnE, $PnG or $TIMESTEP. Storage dtype per column: I → uint8/uint16/uint32/uint64 (smallest that holds $PnB bits; 24-bit → uint32), F → float32, D and A → float64.

Oracle hash (corpus tests)

xxh3 in corpus/oracle/<id>.json → tables[] = xxh3-128 over every value of the data set as little-endian f64, column-major (all events of parameter 1, then parameter 2, …). oracle/gen.py computes it from FlowIO (as_array(preprocess=False)) or, for data sets FlowIO cannot decode (24-bit integers, blank HEADER offsets), from fcsparser; the corpus test computes it from read_table. FlowIO 1.4 ignores FCS 3.2’s $PnDATATYPE, so for FCS 3.2 files with mixed types the oracle re-reads the DATA segment with a NumPy record type built from each measurement’s type (fcsparser, which also ignores it, is kept as a disagreeing second opinion).

CRC (§3.5)

Eight ASCII bytes right after the last segment of a data set. CRC-16 with the CCITT polynomial, each input byte bit-reversed and initial value 0 (processed LSB-first: reflected polynomial 0x8408; check value of 123456789 is 0x2189). 00000000 means “not computed” (DiVa, Miltenyi 3.0, Navios). No corpus file stores a computed CRC, so a mismatch is a warning, not corruption; byte-swapped equality is accepted.

Multiple data sets

$NEXTDATA is relative to the current data set’s first byte (FCS 2.0 wording, followed by FlowIO; fcsparser treats it as absolute, which only works for the second data set). Guava Muse writes four data sets; Beckman Coulter .lmd files hold an FCS 2.0 data set followed by an FCS 3.0 copy with 32-bit values whose TEXT comes after its DATA. The walk stops at 0, at an offset past the end of the file (truncated), or at an offset already visited (bad_next_data).

`check` finding codes

truncated, data_length_mismatch, missing_data_offsets, missing_keyword, bad_keyword, unsupported_width, bad_offset, bad_next_data, bad_data_set (errors); offset_discrepancy, data_end_off_by_one, duplicate_keyword, duplicate_parameter_name, text_unterminated, dangling_keyword, empty_value, nonprintable_keyword, bad_delimiter, mixed_byte_order, histogram_mode, crc_mismatch, non_keyword_segment, no_events (warnings); padded_numeric_value, blank_header_offset, non_utf8_value, log_zero_offset, crc_ok, crc_not_computed (info).

Instrument-specific keywords (`vendor_keywords`)

Every keyword not starting with $ (primary and supplemental TEXT) is copied verbatim into table extra.vendor_keywords, grouped by a mechanical prefix (keyword_prefix): a namespace before an inner $ (FJ$ACQSTATE → FJ, GTI$SAMPLEID → GTI, P$CFGTYPE → P), a leading symbol (@SAMPLEID1 → @, &10PATIENT ID → &, #… → #), per-parameter keywords Pn… → Pn, letters followed by a number (LASER1NAME → LASERn), else the first word (CST SETUP STATUS → CST, ANALOG_COMP → ANALOG). Values over 1024 bytes are shortened in info; info --view full has them in full. No meaning is assigned except where stated above (CREATOR → software, SPILL → spillover).

writer (from the file)	corpus files	keywords seen
BD FACSDiva 6.2 (`CREATOR`)	`fcsparser-facs-diva`, `-fortessa-a01`, `-fake-large`, `-hts-lsr-ii-d06`, `flowio-100715`	`CREATOR`, `TUBE NAME`, `EXPERIMENT NAME`, `GUID`, `SPILL`, `APPLY COMPENSATION`, `THRESHOLD`, `WINDOW EXTENSION`, `FSC ASF`, `AUTOBS`, `EXPORT USER NAME`, `EXPORT TIME`, `CST SETUP STATUS`, `CST BEADS LOT ID`, `CST SETUP DATE`, `CST BASELINE DATE`, `CYTOMETER CONFIG NAME`, `CYTOMETER CONFIG CREATE DATE`, `CYTNUM`, `LASERnNAME`, `LASERnDELAY`, `LASERnASF`, `PLATE NAME`, `PLATE ID`, `WELL ID`, `SAMPLE ID`, `PnDISPLAY`, `PnBS`, `PnMS`
Cytek SpectroFlo 2.2.0 (`CREATOR`, `$CYT/Aurora/`)	`fcsparser-cytek-nl-2000-header`	`CREATOR`, `LASERnNAME`, `LASERnDELAY`, `LASERnASF`, `FSC ASF`, `WINDOW EXTENSION`, `THRESHOLD`, `APPLY COMPENSATION`, `USERSETTINGNAME`, `TUBENAME`, `GROUPNAME`, `CHARSET`, `PnDISPLAY`
FlowJo Collectors’ Edition 7.5 on a Cytek xP5	`fcsparser-cytek-xp5`	`CREATOR`, `ANALOG_COMP`, `FJ$ACQSTATE`, `FJ_$P1R`, `LASERnNAME`, `SAMPLEID`

Observed corpus values

file	version	delimiter	`$DATATYPE` / `$PnB`	`$BYTEORD`	data sets	notes
`fcsparser-facs-diva`	3.0	0x0C	F / 32	4,3,2,1	1	83411 events, SPILL 8×8
`fcsparser-cytek-xp5`	3.0	`\`	I / 24	4,3,2,1	1	8 × 3 bytes × 23126 = DATA length
`fcsparser-cyflow-cube-8`	3.0	`/`	I / 8,16,32	1,2,3,4	1	OTHER segment = STEXT = ZIP
`fcsparser-guava-muse`	3.0	`/`	F / 32	1,2,3,4	4	relative `$NEXTDATA`
`flowio-coulter-lmd`	2.0 + 3.0	`\`	I / 16, then I / 32	1,2 then 1,2,3,4	2	second TEXT after its DATA
`flowio-m0-wm278-s1-zero-gain`	3.1	`\|`	I / 32	1,2,3,4	1	`$PnR/4294967296/`, STEXT after DATA
`fcsparser-miltenyi-fcs31`	3.1	`/`	F / 32	1,2,3,4	1	STEXT 2722–127220, DATA end off by one
`zenodo17457137-aurora-beads`	3.1	0x0C	F / 32	1,2,3,4	1	71 parameters: 64 full-spectrum detectors, `$PnTYPE`, 64×64 identity `$SPILLOVER`
`zenodo18439538-cytoflex`	3.0	0x0C	F / 32	4,3,2,1	1	zero-padded `$BEGINDATA`, 8×8 `$SPILLOVER` over H and A detectors
`zenodo7971252-sony-sa3800`	3.1	`\`	F / 32	1,2,3,4	1	`$BEGINSTEXT` at 64 (inside the primary TEXT), `$PnD/Logarithmic,6,1/`
`zenodo10510047-cytof-mouse`	3.0	`\|`	F / 32	1,2,3,4	1	51 metal channels; keyword `$CYTSN_$DATE` (an empty `$CYTSN` value merged with `$DATE`)
`flowkit-gml-events`	2.0	`\`	I / 16	4,3,2,1	1	`$P1G/3.67/`, `$P2G/8/` and `$PnE/4,0/` on four channels: scale values differ from raw

Vocabulary (every public identifier in `openreadout-fcs` must appear here)

identifier	meaning
`FcsReader`, `FcsDataset`, `FcsFile`, `FORMAT_ID`	reader entry points: format reader, opened file (core `Dataset`), parsed data-set chain, the id `fcs`
`open`, `path`, `file_len`, `data_sets`, `chain_findings`, `read_at`	opening a file; its path and size; the data sets found; findings about the `$NEXTDATA` chain; bounded read helper
`DataSet`, `index`, `offset`, `header`, `delimiter`, `text_range`, `text`, `supplemental`, `analysis`, `data_type`, `byte_order`, `mode`, `event_count`, `parameters`, `data_range`, `data_source`, `next_data`, `findings`	one data set and its parts (absolute offsets)
`keyword`, `event_width`, `expected_data_len`, `is_free_format`, `last_segment_end`	data-set helpers: keyword lookup (primary then supplemental), bytes per event, `$TOT × width`, `$PnB/*/` present, where the CRC field starts
`AuxSegment`, `range`, `keywords`, `note`	supplemental TEXT / ANALYSIS: location, parsed keywords, why not parsed
`OffsetSource` { `Text`, `Header` }, `name`	which pair located DATA
`Header`, `HeaderError` { `TooShort`, `NoSignature`, `BadOffset` }, `HEADER_LEN`, `VERSIONS`, `version`, `data`, `other`, `header_len`, `blank_fields`, `parse_header`, `looks_like_fcs`	HEADER parsing and signature test
`SegmentRange`, `begin`, `end`, `new`, `is_unset`, `byte_len`, `absolute`	inclusive byte range and helpers
`Keyword`, `KeywordSet`, `name`, `value`, `entries`, `duplicates`, `push`, `get`, `contains`, `len`, `is_empty`	keyword-value pairs, ordered, case-insensitive lookup
`TextParse`, `empty_values`, `dangling_keyword`, `nonprintable_keywords`, `latin1_values`, `unterminated`, `parse_keywords`	result of parsing a keyword segment and its irregularities
`parse_uint`, `parse_float`, `is_padded`	numeric keyword values (trimmed) and the padding test
`DataType` { `Integer`, `Float`, `Double`, `Ascii` }, `from_code`, `code`	`$DATATYPE`
`ByteOrder` { `LittleEndian`, `BigEndian`, `Mixed` }, `from_value`	`$BYTEORD`
`Mode` { `List`, `Correlated`, `Uncorrelated` }	`$MODE`
`FieldWidth` { `Fixed`, `FreeFormat` }, `width`	`$PnB`
`Parameter`, `number`, `short_name`, `label`, `range_text`, `amplification`, `gain`, `voltage`, `wavelengths_nm`, `filter`, `detector_type`, `power_mw`, `percent_emitted`, `display`, `calibration`, `data_type`, `detector`, `measurement_type`, `feature`, `tag`, `analyte`, `byte_width`, `bit_mask`, `read_parameters`	one parameter’s `$Pn*` keywords and derived widths/masks
`storage_dtype`, `decode_fixed`, `decode_free_ascii`	column dtype rule and DATA decoding
`SpilloverMatrix`, `values`, `parse_spillover`, `spillover`	spillover/compensation matrix and which keyword supplied it
`keyword_prefix`	mechanical grouping of non-`$` keywords for `vendor_keywords`
`scale_columns`, `file_matrix`, `parameter_names`, `is_time_parameter`, `is_fluorescence_parameter`	raw → scale values; the file’s own compensation matrix; `$PnN`/`$PnS` lists; parameter roles used by `--transform` defaults
`Family` { `BdFacsDiva`, `BdSpectral`, `CytekSpectral`, `BeckmanCytoflex`, `SonySpectral`, `SonySorter`, `MassCytometry` }, `family`, `vendor`, `technology`, `id`, `platform`	instrument family recognition and `extra.platform`
`channel_kind`, `column_extras`, `metal_tag`, `metal_marker`	per-column roles, metal tags and markers
`FcsDate`, `year`, `month`, `day`, `iso`, `next_day`, `parse_date`, `parse_time`, `parse_timestamp`, `acquisition_span`	`$DATE`/`$BTIM`/`$ETIM`/`$LAST_MODIFIED` to ISO-8601
`Crc16`, `update`, `finish`, `CrcField` { `Absent`, `NotComputed`, `Value`, `Other` }, `read_crc_field`	data-set CRC
`RowCondition`, `column`, `op`, `threshold`, `CompareOp` { `Greater`, `GreaterEqual`, `Less`, `LessEqual`, `Equal`, `NotEqual` }, `symbol`, `test`, `parse_conditions`, `filtered_slice`	`table --filter 'FITC-A > 1000' --count`: row conditions (`COLUMN OP NUMBER`, all must hold; NaN meets none) tested over the whole table on the values `table` returns, raw or compensated/transformed (module `filter`)

How this reader was derived, file by file: provenance log.