Kid mode 🧒

This is the made-up-live part of a song called Plasma. Only the keyboard player got pulled out of the recording cleanly, so this whole page is about Page. He plays three notes that stack into a bright chord. For the first half the highest of the three is a note called C. Halfway through he moves a note called F up to the top instead. Same three notes, new favourite one. He also swaps from an organ sound over to a real piano partway through, and you can see the swap happen in the numbers. The grown-up breakdown is below, tap Grown-up mode.

ixi-style analysis · believe-you-me

Plasma 2025-04-20 jam peak · ixi analysis

An ixi-style read of the improv window, 4:45 to 9:15 within the extracted jam file, which is shorter than the released track, so these timings are file-relative and not LivePhish track positions. This jam's stems never split the bass, the drums or the guitar, so Mike, Fishman and Trey are unread here. What survives is demucs's piano output and its other output, plus a vocal lane used to gate the window. Read the caveats before the finding. The other lane is demucs's residual output and demucs names no instrument for it. That lane was audited on 2026-07-19 against a jam whose run produced all six stems, and the audit clears it of the two worst possibilities: it holds 37.0% of its energy below 250 Hz against a clean residual's 38.2% (an isolated bass lane runs 82.0%), and 3.7% in the 2k to 8k cymbal band against a real drum lane's 47.7%, at a tonal flatness of 0.022 against drums at 0.104. The missing bass and drums are not inside it. Its note-attack rate does run 2.2 times the clean residual and that excess is tonal, so guitar may be folded in, and the lane is therefore called keys-like rather than Page alone. Chroma also folds octaves, so the pitch-class ordering below is statistical weight. Register is not measured. With those in place: the window correlates to F major at 0.916 (third in the run behind Halley's Comet 0.953 and Twist 0.941, and off two lanes rather than six, so not strictly comparable). The measured change is that the three heaviest pitch classes stay F, A and C throughout, and the heaviest single class flips from C to F around 7:00. That is an emphasis shift inside one chord, not a change of change: C was the fifth of F major all along. The staves are pitch-tracked and then snapped to F major, which makes them a quantized reading and not a transcription in their own right.

Plasma (jam peak, 4:45 to 9:15) Phish · 2025-04-20 Moda Center, Portland

F major pitch classes · the residual lane leads (4:45 to 5:52) · C carries the most weight
Across the window the three heaviest pitch classes are C, F, A, the notes of an F major triad. Ordering here is by chroma weight, not by register: chroma folds every octave into one class, so this says which notes carry the most spectral energy. It says nothing about which note is physically highest. The section correlates to F major at 0.890. Activity sits mostly in demucs's other output, 59.8% of the two keys lanes, tracking 98.4% voiced across the loudest five seconds against the piano output's 42.2%. Mike, Fishman and Trey are unread. A 2026-07-19 audit of the other lane against an all-six-stem control shows the missing bass and drums are not inside it: 37.0% of its energy sits below 250 Hz against a clean residual's 38.2%, where an isolated bass lane runs 82.0%, and it carries 3.7% in the cymbal band against a real drum lane's 47.7%. Its attack rate does run 2.2 times the control, so guitar may be folded in with the keys.
the hand-off · the piano output takes over (5:52 to 7:00) · the weighting holds
Activity moves from the other output to the piano output, 40.2% up to 66.4%, with the piano's voiced rate rising 42.2% to 66.3%. Read plainly, that is a shift between two separated outputs. Reading it as Page moving from organ to piano is the obvious musical explanation and it is an inference, not a measurement, because neither lane is verified to contain only Page. The weighting does not move, still C, F, A, correlating F major at 0.923, the strongest section figure on the page.
the emphasis shift · F becomes the heaviest class (7:00 to 8:07) · same three classes, new order
The change. The ordering becomes F, A, C, with F now the heaviest class in place of C. This is a shift of emphasis inside one chord. It is not a change of root: C was already the fifth of F major in the earlier sections, so moving weight onto F strengthens a tonic that was there the whole time. The same three classes stay dominant. This measurement cannot rule out passing tones or separation artifacts elsewhere in the texture. The piano output tightens to 76.3% at 84.1% voiced. Vocal energy rises from 4.5 to 118.2, and here that rise does survive testing. A 2026-07-19 validation of the vocal lane shows level and voiced content rising together, from -72.1 dB at 1.0% voiced in the opening section to -37.3 dB at 10.1% voiced here, a 5.3x lift in voiced share against the rest of the window. The band is bringing the song back over the jam.
favorite momentThe emphasis moving onto F with the chord itself unchanged, a tonic simply getting heavier.
F emphasis held · the ride-out (8:07 to 9:15) · the ordering stays put to the end
The F, A, C ordering holds to the end. The piano output reaches 76.5%, and the other output thins to 9 tracked notes across five seconds against the piano's 16, most of it one sustained tone. The section correlates F major at 0.899. Vocal energy eases back to 50.8.
diagnostic · the same lanes unsnapped · no key assumption, chromatic spelling, K:C
The staves above were snapped to F major, and a snapped stave cannot be used as evidence for the key it was snapped to. Here is the measurement instead. Across both keys lanes the snap changed the pitch class of 35 of 110 notes (31.8%): the piano lane 14 of 49 (28.6%) at a median move of 13.9 cents, and the other lane 21 of 61 (34.4%) at 11.1 cents, against median source tuning errors of 12.2 and 11.0 cents. That is a third of the notes re-spelled, which is high. This stave redraws each lane's worst section with no key assumption at all. Read the snapped staves as illustration only. The F major finding on this page is built from the chroma correlation, which was never passed through a scale. Full per-section figures in _windows/plasma_piano_diagnostic.json and _windows/plasma_other_diagnostic.json.
Built from an autocorrelation and chroma pass over the three demucs lanes this jam produced: the piano output, the other output, and a vocal lane used to gate the window and to indicate probable vocal activity. Bass, drums and guitar never separated, so Mike, Fishman and Trey are unread. Three limits follow and none of them are cosmetic. First, demucs documents other as a residual output and warns that its six-source model leaks between piano and the residual. That lane was audited on 2026-07-19 against 2025-04-27 Twist, whose run produced all six stems so its own other is a true residual baseline. Over 180 seconds per lane: sub-250 Hz share 37.0% here against 38.2% for the clean residual and 82.0% for an isolated bass lane; 2k to 8k share 3.7% here against 47.7% for a real drum lane; spectral flatness 0.022 here (tonal) against 0.104 for drums. The missing bass and drums are therefore not meaningfully inside this lane. What the audit does not clear is guitar: onset rate is 2.46 per second here against the clean residual's 1.12, and that excess is tonal, so guitar content may be folded in with the keys. The lane is treated as keys-like rather than as Page alone. The vocal lane was validated the same day by testing structure rather than level, since energy alone proves neither singing nor a return: voiced-frame share runs 1.0, 1.9, 10.1 and 6.0% across the four sections against levels of -72.1, -57.8, -37.3 and -43.7 dB, so level and voiced content rise together and the vocal entry at 7:00 is supported at 5.3x the voiced share of the rest. The same test withdrew the equivalent claim on the Back on the Train sheet, so it is not a rubber stamp. Second, with no bass lane the ensemble root cannot be established from these stems at all. F is the best-supported reading of the keys material and it is not a verified root. Third, chroma discards octave and register, so every ordering here means spectral energy share. Which note is highest is not measured. The staves are snapped to F major, and snapping can manufacture correct-looking spelling by moving a detector estimate onto the nearest permitted degree, so treat them as illustrative. Window from the composite picker at 4:45 to 9:15, score 0.47, dominant feature interplay, file-relative. Krumhansl over piano plus other: F major 0.916, A minor 0.719. Correlation is a similarity score. It is not calibrated certainty. The earlier full-jam anatomy pass classified this Plasma as hybrid at 0.87 with its excursions in the first half, and this window is the back half, which fits a jam that wanders early and settles late. Scored 3.5 out of 10 by an adversarial ChatGPT deep-research pass on 2026-07-18; this revision is the response to it.
← Jam Library Back on the Train ixi Sigma Oasis ixi A Wave of Hope ixi Stem player