_FUNCnnn.DAT

Binary spectrum data file. One file per function. Contains all spectra for that function, stored contiguously. Three record layouts (encodings) are known. The encoding is a property of the DAT records, not of the index: Encodings D and E appear behind both the 22-byte (Variant A) and the 30-byte (Variant B) index.

Encoding	Record size	Index variants	Layout
A	6 bytes	A	u16 ion count, floating-point m/z (4-bit exponent, 24-bit mantissa)
D	8 bytes	A, B	16.16 intensity, floating-point m/z (5-bit exponent, 27-bit mantissa)
E	12 bytes	A, B	compressed intensity, Encoding D m/z word, auxiliary word

Every encoding stores an uncalibrated m/z per record; the _HEADER.TXT T1 polynomial applies to sqrt(m/z). No lock-mass correction is applied by the reader.

Choosing the record width

• Variant A index: consecutive DAT offsets divided by the record count give 6, 8 or 12 bytes per record.
• Variant B index: the record has no record count. The width is judged from the data. Under the right width every record's bytes 4-7 are a position word with bit 26 set, and m/z never decreases within a scan. Up to 16 evenly spaced scans are sampled and the first one that fits exactly one of 8 or 12 bytes (with at least three records) decides. When none decides, the reader logs a warning and assumes 8 bytes; a wrong guess then fails at decode time with an error rather than producing data.

Encoding A: 6-byte records (Variant A index)

Status: Decoded; m/z checked against lock-mass references (2026-09-26)

Observed in: PXD058812, PXD003126, PXD010569, PXD021125, PXD029515, PXD041695 (older QTof and Q-Tof Premier-class instruments, MassLynx 4.0-4.1)

Key facts:

• File is a flat array of 6-byte records (no top-level file header)
• Scan boundaries and record counts come from IDX Variant A (offset u32@0x00, count in the low 24 bits of u32@0x04)
• Each record stores an ion count and a floating-point m/z word; there are no sentinel records and no TOF bins to rescale

6-byte Record Layout

Bytes	Type	Description
0-1	u16 LE	Ion count (TDC hits; small integers, 0 for range markers)
2	u8	High nibble: m/z exponent e. Low nibble: always 0 in the corpus
3-5	u24 LE	m/z mantissa M, normalized (bit 23 always set)

mz_uncal = M * 2^(e - 24)          # M in [2^23, 2^24), so mz_uncal in [2^(e-1), 2^e)
mz       = (T1(sqrt(mz_uncal)))^2  # _HEADER.TXT "Cal Function N" polynomial

The first and last records of a scan have count 0 and decode to the acquisition range in _FUNCTNS.INF: in PXD058812 70 CA FF C7 is 0xC7FFCA * 2^-17 = 100.000 and B0 4A FF F9 is 0xF9FF4A * 2^-13 = 1999.98 for a 100-2000 function. The decoder rejects records whose low nibble is non-zero or whose mantissa is not normalized.

Validation (clean-room)

• Every record in six bundles has a zero low nibble and a normalized mantissa, and decoded ranges match _FUNCTNS.INF to within 0.03 Da.
• Lock-mass functions, after T1 calibration: PXD003126 [Glu1]-fibrinopeptide B [M+2H]2+ +24 ppm (+87 ppm without T1); PXD021125 +25 and +38 ppm.
• PXD041695 background ions calibrate to 429.092 and 445.12-445.13, matching the common polysiloxane contaminants at 429.0887 and 445.1200; without T1 they are about 900 ppm high.
• PXD029515's lock function shows the Glu-fib envelope at a consistent -150 ppm on both isotopes, while PXD003126, acquired with identical _extern.inf geometry and software, sits at +24 ppm. The offset is attributed to that instrument's calibration at acquisition time; this has not been proven.

Encoding D: 8-byte records (either index variant)

Status: Decoded; m/z checked against physics and lock-mass references

Observed behind the 22-byte index in: PXD081045 (Vion IMS QTof, UNIFI 2.0 export), PXD001123 and PXD009047 (SYNAPT G2, MassLynx 4.1), PXD037102.

Observed behind the 30-byte index in: PXD001175, PXD001471, PXD002393, PXD005960, PXD035818 (SYNAPT G2-S); PXD066594, PXD068881, PXD079562, PXD080129 (SYNAPT G2-Si); PXD071342, PXD073126 (SYNAPT XS); PXD045625, PXD053170, PXD075602, PXD078353 (Xevo G2-XS, all non-lock functions and some lock functions); PXD069628 (Xevo G3).

Key facts:

• No sentinel records; the first and last records are ordinary profile points
• Zero-intensity records are skipped by the decoder

8-byte Record Layout

Bytes	Type	Description
0-3	u32 LE	Intensity, unsigned 16.16 fixed point (bytes 0-1 are the fraction)
4-7	u32 LE	m/z word: 5-bit exponent E (bits 27-31), 27-bit mantissa F with bit 26 always set

mz_uncal = F * 2^(E - 27)          # F in [2^26, 2^27)
mz       = (T1(sqrt(mz_uncal)))^2

A position word without bit 26 set is a decode error.

Validation (clean-room)

22-byte index:

• Profile points are spaced by exactly one ADC sample. With flight time t = A_us * sqrt(m/z) from Lteff/Veff and the ADC Sample Frequency in _extern.inf, one sample is 2 * sqrt(m/z) * A_us / f. The median observed spacing divided by that prediction is 0.9986-1.0028 in every 100-Da band from 200 to 1900 m/z (PXD081045, 7.2 GHz) and 0.9998 on the SYNAPT G2 files (3.0 GHz). A linear TOF-bin reading cannot satisfy this.
• Isotope spacing is flat at 4 words per Da over 256-512 m/z and 2 per Da over 512-1024, the signature of an m/z (not time) mantissa.
• Lock mass after T1: SYNAPT G2 Glu-fib and Leu-Enk at -10 and -11 ppm. Vion Leu-Enk (ReferenceMass1 556.27658 in _extern.inf) at +20 to +69 ppm across 45 runs, with one offset per acquisition batch; the intense monoisotopic peak saturates and centroids 10-20 ppm higher than its isotopes. Without T1 the same peaks are 130-146 ppm low.

30-byte index:

• In 20 evenly spaced scans of every 8-byte function in the 18 bundles above, every record's bytes 4-7 have bit 26 set and m/z never decreases within a scan.
• Bytes 0-1 are non-zero in a minority of records (about 1-55% per function), with values such as 0xFF00 and 0xE000, consistent with the fraction of a 16.16 intensity.
• Lock mass, median over 20 lock-mass scans, after T1, counting only scans where the reference peak's 13C isotope sits one isotope spacing higher (examples/audit_corpus.rs):

Bundle	Instrument	Reference	ppm
PXD001471 57	SYNAPT G2-S	Glu-fib 2+	-17.3
PXD002393 S130426_21	SYNAPT G2-S	Glu-fib 2+	-75.6
PXD001175 S121126_06	SYNAPT G2-S	Glu-fib 2+	+161.7
PXD068881 CtpA_1076_2h_1	SYNAPT G2-Si	Leu-Enk	-29.6
PXD080129 186/203/205_nr15	SYNAPT G2-Si	Leu-Enk	+20.3 each
PXD071342 MDE_WT2_DIA	SYNAPT XS	Leu-Enk	+55.0
PXD073126 KMI_sFtsk_2	SYNAPT XS	Glu-fib 2+	+37.4
PXD075602 DHPR_11257-1	Xevo G2-XS	Leu-Enk	+66.5
PXD078353 (two runs)	Xevo G2-XS	Leu-Enk	+48.7, +42.2
PXD069628 HC18_CE, HC20_CE	Xevo G3	Leu-Enk	-97.6, -107.9

The PXD005960 lock function holds one scan whose dominant ion (m/z 825.1) is not a known lock compound. The larger offsets (PXD001175, PXD069628) are uniform across the reference peak and its isotope, which points to instrument calibration at acquisition time rather than the record model; this has not been proven.

Encoding E: 12-byte records (either index variant)

Original public LCT Premier acquisitions in MTBLS701 and MTBLS13770 pair the 22-byte index with 12-byte mass records. Consecutive index offsets divided by the preceding 24-bit record count establish the width. A separate optical function in MTBLS701 uses the existing 6-byte encoding and is unchanged.

The lock-mass functions of public Xevo G2-XS bundles PXD045625 (Abu_190520_Sha11, function 3) and PXD053170 (both runs, function 2) pair the 30-byte index with the same 12-byte layout; their other functions use Encoding D.

Offset	Size	Interpretation
0	4	Compressed intensity word, little-endian
4	4	Floating-point m/z word, same representation as Encoding D
8	4	Auxiliary word, meaning unresolved

For intensity word u, the measured normalized representation is:

mantissa = u & 0x001fffff
exponent = (u >> 22) & 0x1f
intensity = mantissa * 2^(exponent - 21)

Every nonzero intensity in the six LCT mass functions inspected has bit 20 set and bit 21 clear. Higher bits are flags rather than exponent bits; their meaning is unresolved. The decoder keeps flagged peaks and ignores the auxiliary word. It rejects unsupported nonzero mantissa patterns. The T1 header calibration applies to sqrt(m/z), as for Encoding D.

Across all 24,484,578 mass records in the two selected LCT acquisitions, raw positions are ordered within every scan and have the normalized Encoding D position pattern. Intensity sums excluding bit-28-marked points agree with the same-file index TIC within 113 ppm. This is a byte-derived consistency check, not a definition of that flag or external proof of absolute peak accuracy. No vendor software or vendor-derived expected output was used. The exact 213-byte _CHROMS.INF variant in issue #36 was not present and remains unresolved.

On the Xevo G2-XS lock-mass functions, every sampled scan decodes without an intensity-word error and leucine enkephalin lands at +11.0 (PXD045625), +9.5 and +10.4 ppm (PXD053170) with its 13C isotope one spacing higher. The intensity scale on these functions has not been checked against a TIC.

This support covers the twelve-byte mass functions. The separate MTBLS701 optical function still uses Encoding A and fails its existing m/z decoder on some scans. Canonical MTBLS13770 function 2 records also retain the existing MS2-without-precursor metadata behavior, which does not pass the shared core conformance check. Neither limitation is corrected by Encoding E.

Profile or centroid

Spectra are labelled per encoding, from the spacing of the stored points:

Encoding	Label	Evidence
A	profile	Peaks are runs of consecutive points one sample apart, for example PXD058812 molecular_mass_P15_01.raw scan 98: about 90 points at 0.009 Da steps from 600.68 to 601.47 tracing two isotope peaks
D	profile	Consecutive points are one ADC sample apart (see the Encoding D validation); in 30-byte-index functions 51-99% of consecutive steps are a single sample, and a lock-mass peak spans several points (PXD068881: 556.244, 556.254, 556.264, 556.275, 556.285)
E behind the 30-byte index	centroid	One point per peak at irregular spacing: in PXD053170 20231113_NSE_Sample_High.raw function 2, leucine enkephalin and its isotopes are single points at 556.159, 557.162 and 558.163 (uncalibrated), with neighbours 0.03-0.1 Da away; only 21-26% of steps are a single sample
E behind the 22-byte index (LCT Premier)	centroid	Same record layout as above. The point spacing of these functions has not been checked; the LCT bundles are not in the public corpus used here

_extern.inf Data Format reads Continuum for every function that declares it, including the Encoding E lock-mass functions above, so it is not used. The _FUNCTNS.INF +0x000 word is consistent with the encoding (bits 10-14 are 9 for Encoding A, 12 for D behind the 22-byte index, 28 for D behind the 30-byte index and 29 for E behind the 30-byte index; bit 15 marks lock mass), but no single bit separates profile from centroid, so it is not used either.

Ion mobility (SYNAPT HDMS): not decoded

The reader does not decode ion mobility. Scans from SYNAPT functions are returned as m/z and intensity only, mzML output carries no mobility array, and no run declares a mobility array kind.

Known facts:

• SYNAPT functions behind the 30-byte index use Encoding D records. Within every sampled scan m/z never decreases, so a stored scan is not split into drift-ordered blocks.
• Where drift time is recorded is unresolved. Candidates include the index records, the _PROCnnn files and per-bundle side files such as mob_cal.csv (shipped in the PXD080129 bundles).

Fields Under Investigation

• Encoding A: meaning of the byte 2 low nibble (always zero in the corpus)
• Encoding E: intensity flag bits and the auxiliary word
• Ion mobility: where drift time is stored for SYNAPT acquisitions

Reference Sources

• Empirical hex analysis: re/src/analysis/inspect.py
• Calibration: _extern.inf (Lteff, Veff, pusher cycle) + _HEADER.TXT (Cal Function N)
• Corpus samples:
  • PXD058812/molecular_mass_P15_01.raw (Encoding A, 197 scans, ~1050 rec/scan)
  • PXD058812/MS_fragmentation_P29_01.raw (Encoding A, 426 scans)
  • PXD066594/WANG.raw (30-byte index, Encoding D, 590 scans)
  • PXD068881/20220517_CtpA_1076_2h_1.raw (30-byte index, Encoding D, 1138 scans)
  • PXD075602/DHPR_11257-1.raw (30-byte index, Encoding D, 1150 scans)
  • PXD053170/20231113_NSE_Sample_High.raw (30-byte index; Encoding D survey, Encoding E lock mass)