ShootFlow reads Phase One's IIQ raw files directly, and since this morning it also applies the factory calibration that every Phase One back writes into each frame: defect maps, per-tile flat fields and gain curves. That check covered ten backs. It left out the newer formats (IIQ S, 16-bit IIQ L) and the aerial cameras, and it left one question open. On an iXU180 frame, an aerial mapping camera, ShootFlow's blue came out 5.2% lower relative to green than LibRaw's. The working guess was that the calibration would close that gap.
It didn't. Here is what this run found (full note).
The calibration matches on every format
All 13 remaining public Phase One files on raw.pixls.us went through the sensel-for-sensel comparison against LibRaw 0.22: the iXU180, the IQ180 in both compression settings, the IQ3 100MP three ways, the Trichromatic, the IQ4 150MP three ways, and two iXM aerial bodies. That is 1.42 billion sensels. Nine files equal LibRaw on every sensel. The IQ180 frames are within 3 codes, from float rounding in a spline. The iXU180 is within 2 codes everywhere except the four columns its file marks dead, which ShootFlow fills its own way. Each frame takes 0.44 to 0.76 s to calibrate, for frames up to 151 megapixels.
The first pass said the opposite
The first run of the check said the iXU180 was almost entirely wrong: 0.001% of sensels equal, LibRaw 4-21% higher. The check itself had two problems, and the calibration code had none.
LibRaw's renderer lowers its white point to the frame's brightest value when that value is close to the nominal one. On this frame that scaled every value by 3.3%. Turning that adjustment off removed the uniform part.
The rest was more interesting. An IIQ file can carry two tables of black-level offsets, one for each row and one for each column. ShootFlow adds them, and so does LibRaw for the IIQ L and uncompressed formats. LibRaw's IIQ S loader never reads them. So on IIQ S files, LibRaw's sensels are missing the tables before any calibration starts. With the tables zeroed on our side as well, the iXU180 matched.
Where the 5.2% went
Measured as each colour's mean, the calibration moves the iXU180's blue/green ratio by 0.2%, while the offset tables move it by 5.6%. With the tables zeroed and the calibration on, our levels equal LibRaw's to the fourth decimal. The IQ180 has tables too, but they span only -4 to +8 codes, so there the two readers agree whatever you do with them.
The iXU180's row table is not noise. It is a clean ramp: about -30 codes at the top and bottom of the sensor, falling to -333 at the middle row, the same on both halves. On a black level of 1024, that lowers the middle of the frame by up to a third of the pedestal.
Which reader is right?
We don't know yet, and the note says so. ShootFlow keeps applying the tables. The file stores them, LibRaw applies the same tables on every other IIQ format, and nothing marks them as IIQ L-only, so LibRaw's omission looks like a gap in one loader rather than a choice. But this frame can't prove it. It is a bright aerial scene with no shadows, the margins outside the crop hold filler rather than masked black sensels, and the seams at the sensor's split lines look the same either way.
A lens-cap dark frame from an iXU180, or a Capture One export of the same file, would settle it. If you fly one of these cameras and can share either, that is the next step. Until then, an iXU180 IIQ S frame developed in ShootFlow will show about 5% less blue relative to green than the same frame through LibRaw-based tools such as darktable or RawTherapee, most of all in the middle rows.
The checks are tools/iiq_calibration_check.py and tools/iiq_offsets_check.py in the photo-editing venture. Both run against the public CC0 files on raw.pixls.us.