The organs stop playing an octave below the note you asked for: pitch-audit has been reporting Drawbar, Percussive and Rock Organ plus Bagpipe as FAIL octave for long enough that the program rows had grown comments explaining why the tool was wrong -- "the 16' adds a HALF-strength sub for body, NOT the fundamental. (Audit latches the 16' -> reads octave-low; by-design, the pitch sounds right.)" It did not sound right. The Drawbar registration was 16' .45, 5-1/3' .30, 8' .90, 4' .55, 2' .30, which is ratios 0.5, 1.5, 1.0, 2.0, 4.0 -- relative to the 16', harmonics 1, 2, 3, 4 and 8 of f0/2. A complete harmonic series with a strong first partial is the textbook condition for hearing THAT as the fundamental, so this was not a detector latching onto a quiet bar, it was a listener hearing C3 for a played C4. Measured by subharmonic summation -- the virtual-pitch model, which asks what the ear assigns rather than where the lowest strong partial sits -- Drawbar read -1205 cents and Percussive -1209. master
The 5-1/3' turned out to be the real culprit rather than the 16' itself: at ratio 1.5 it is the sub's THIRD harmonic and no harmonic of f0 at all, which is exactly what makes f0/2 credible. Trimming it and adding a 2-2/3' (ratio 3.0, f0's own third) lets the 16' stay properly audible at 0.25 instead of having to fall to 0.05, which is where a naive trim-the-sub sweep pointed. Rock needed the same repair at 0.40/0.35, and Bagpipe's "drone" -- which tracks fc * 0.5, so at 0.5 mix it is an octave doubler rather than a drone -- came down to 0.25. Rock is the one worth remembering, because it hid. At C4 its percept sits on the unison, so a single-note check passes and I recorded it as a false positive on exactly that evidence. It only drops to f0/2 from F#4 upward: the sub has to win in the ear's dominance region, roughly 500-2000 Hz, and the higher the note the better placed f0/2 is to do it. The register I sampled was fine; the instrument was not. Bagpipe hid the same way. Sampling one note is not a test, which is the second time this session that lesson has cost a wrong conclusion. Verified across C2-C6 with production jitter: all four now within a few cents, worst -33 at the very bottom of Drawbar where the sub falls out of the dominance region anyway. The project's own tools/pitch-audit goes FAIL 4 -> FAIL 1, and a verdict-by-verdict diff of all 128 programs shows only 16, 17 and 109 moving, none regressing. The remaining FAIL is Rock, and that one really is a detector artifact: its two detectors split 523/131 -- second harmonic and sub -- and neither picks the fundamental, which at C4 carries 28.9% of the power against the sub's 1.9%. Overdrive puts it in the rich/noisy category the audit's own header warns time-domain detectors about. The comments now say which of these to believe and why, so the next person does not have to re-derive it.