The Oscillome

What the evidence currently supports

In short: nine months of one person's EEG, from a fourteen-electrode headset, on ordinary days and during psychedelic sessions. The biggest change in these recordings is not any compound but whether the eyes are open or closed: closing them makes alpha (the brain's dominant resting oscillation, about ten cycles per second) several times stronger, so every other comparison has to take eye state into account. With that in hand:

• Alpha is faster in the LSD sessions: every LSD session sits above every ordinary one.
• Alpha is not a pure sine wave. It carries a smaller companion at twice its frequency that rises and falls in step with it, most strongly in the LSD sessions, though eyes-closed ordinary sessions reach the same range.
• The direction in which alpha sweeps across the scalp, from the back of the head towards the front more than the reverse, persists with the eyes closed under LSD and not otherwise.
• Five sets of equations for the brain's cell populations, fitted to every two minutes of signal, say that over a whole recording alpha behaves like a bell struck at random: it rings at its own frequency and dies away. Minute by minute the picture is finer: with the eyes open the bell is heavily damped; closing them brings it within seconds to the edge of ringing by itself (a self-sustaining oscillation, a limit cycle), an edge that one whole-head pattern of alpha crosses in some eyes-closed minutes and not in others, on ordinary days too, and for more of the time in the psychedelic sessions.
• Machine-learning models trained on other people's EEG tell the psychedelic sessions from the ordinary ones in 39 of 40, but those sessions also differ in time of day, room, posture and eye behavior, so the models may be reading the setting as much as the brain. They recognize the limit-cycle state and do not anticipate it. The decisive test, to fix the models and predict sessions recorded later, has now been run once: frozen in September and read on thirteen later sessions, the models called all thirteen and ranked the one psychedelic session first, where the time of day alone called eight; one psychedelic session supports a rank, not a rate, and the eye-state estimate alone also separates the thirteen, so the setting is not yet ruled out (§21).

1. Eye closure raises posterior alpha power several-fold within a session (6.5× in the reference session; 2.2× to 7.9× in every fixed block since). It is the largest effect in the archive and conditions every comparison below (§16). Replicated within archive
2. Alpha runs faster in every LSD-labeled session — 10.3–10.9 Hz against an ordinary median of 9.8 — and the four classes order LSD > DMT > ayahuasca > psilocybin within one person (§06, §17). Supported association
3. Flicker drives an alpha resonance at 9.7 Hz that reproduces to 0.03 Hz across days; in one session it sat 0.3–0.7 Hz higher forty minutes after an LSD dose of a known, small quantity than in the same session's pre-dose sweep (§18). Replicated shift: single session
4. Alpha is strongly phase-coupled to a peak at twice its frequency, most of all under LSD — evidence that strongly supports reading the “beta” peak at 21–22 Hz as a harmonic of alpha's non-sinusoidal waveform rather than an unrelated oscillation; a separate oscillator locked 2:1 is not excluded (§11). Supported association
5. Alpha's back-to-front lag asymmetry survives eye closure under LSD and not otherwise: +0.25 to +0.35 in six of six sessions against a median of +0.01, and +0.02 before the dose against +0.41 after it within one session — the archive's first difference that holds with eye state matched, though outside the timed blocks the matching uses a classifier built from alpha itself, so it also matches partly on alpha strength (§13). Supported association
6. Broadband light noise evokes a perceptual echo near 10.8 Hz, replicating a published effect with a consumer headset and an LED panel (§19). Replicated within archive
7. A simple resonator is enough over a whole recording; minute by minute, with the eyes closed, it is not always. Across five neural-mass and mean-field models fitted to each of the fourteen electrodes of every two-minute window, a lightly damped, noise-driven linear resonator explains the alpha spectrum and amplitude statistics at this resolution — a quality factor of 5.5 at the median electrode-window — and it predicts: fitted to half of a recording at one electrode, the two linear models give the spectrum of the other half with R² 0.94 and 0.96 (532 recording-electrodes), and the flicker resonance of §18 from unstimulated EEG to within 0.29–0.34 Hz. The amplitude test has power at the 152 recording-electrodes with a strong alpha peak, and all 152 sit at the linear limit; a Hopf limit cycle predicts held-out envelope power better than the linear resonator in only 37 of them, by -0.0004 nats per sample, so over a whole recording nothing in the data requires a limit cycle; the next entry looks window by window. Neural-mass and mean-field models are usually fitted once per subject; we did not identify a prior study fitting them at this density to one person's archive (§20). Observed
8. One collective mode crosses a Hopf bifurcation, and the psychedelic sessions spend more time beyond it. Read as a whole-head dynamical system, the recordings have one wave mode that approaches an instability, the front–back alpha pattern, and it crosses a supercritical Hopf point: with the eyes closed it holds a limit cycle in 26 % of minutes in LSD sessions, 37 % in the other psychedelic sessions and 13 % in ordinary ones, and almost never with the eyes open. The crossing is a drift through the point at the scale of minutes with no change at the eyes; beyond it the band condenses into that one pattern and its harmonic locks to it. A ring of Jansen–Rit columns with a front-to-back gradient crosses the same bifurcation; a homogeneous field cannot. The learned representations recognize the regime and do not anticipate it (§22). Supported association
9. A pretrained foundation model reads the labels, but so might the room. The published EEG foundation model LaBraM, frozen, read by a calibrated read-out that never saw the session, separates the session labels in 39 of 40 held-out sessions (REVE 37); LoRA fine-tuning either on these recordings without labels adds nothing, and kept beside the frozen features it costs nothing. Residual eye behavior, setting, posture, time-of-day and other session-level nuisance variables remain entangled with the label, so the next test was prospective: the representation and read-out were frozen and read on thirteen sessions recorded afterwards (§21, Result 22): all thirteen called, the one psychedelic session ranked first, every EEG read-out better than the time of day by 0.4 to 0.6 nats per session; one psychedelic session supports a rank, not a rate. We did not identify a prior evaluation of EEG foundation models on one person's longitudinal, altered-state archive (§21). Supported association
10. The resonator's fitted constants are complementary to the learned representations. The 41 constants of the neural-mass and mean-field models, fitted at each of the fourteen electrodes, place 38 of 40 sessions on their own, are the best single representation for two quantities no network was trained on (the long-range memory of the alpha envelope and alpha phase coupling with alpha strength removed), and given as inputs to a network trained from scratch on this archive — a network borrowing REVE's design, not a foundation model — lift it from 36 to 38 sessions (log loss 0.31 → 0.24, three training seeds). For the strong pretrained networks on the label they are redundant: those already contain what a resonator measures (§21). Supported association

What did not survive is reported at the same length: more alpha under psychedelics (withdrawn), a doubled alpha “wavenumber” (mostly eye state, §12), brighter light speeding alpha up (§19), any lasting trace across the months (§09), thalamocortical dysrhythmia (§10), and interaction between frequency bands (§13).

What limits it

• One person, no placebo, no blinding. Repeated measurement removes between-person variability; it does not remove time, order, expectancy or setting. Nothing here generalizes without testing.
• The session label is confounded with eye state and environment across the passive archive. Acute sessions are a median 34% eyes-closed, ordinary recordings 1.2%.
• Fourteen scalp electrodes, 0.2–43 Hz, no source localization. Phase gradients cannot distinguish a propagating wave from a lagged standing pattern; one generator cannot be told from several locked together.
• Unequal and mostly legacy acute sessions: 6 LSD, 7 DMT, 2 ayahuasca, 1 psilocybin. Most have no reliable dose or timing record; the September protocol records the dose and is adding minute-level timing.

What makes it unusual

• Four psychedelic classes in one nervous system. LSD, DMT, ayahuasca and psilocybin share 5-HT_2A agonism but differ sharply in receptor promiscuity and kinetics. Comparing them between subjects confounds the psychedelic with the person. Here the person is held constant — though dose, order and setting are not — and §17 makes that comparison directly.
• More non-psychedelic recordings than acute ones. Most naturalistic psychedelic work has no non-psychedelic comparison at all. This archive has more non-psychedelic hours than acute — which turned out to matter enormously (§16).
• Eight months of continuity. One headset, one montage, one person, one pipeline, from February to October 2026. Longitudinal drift can be measured rather than assumed absent.
• Neural-mass models and foundation models on the same data. Five neural-mass and mean-field models fitted to every electrode of every window, five pretrained EEG foundation models and transformers trained in-house on this archive alone are fitted or applied to every window of the archive and judged on the same held-out sessions — the two literatures rarely meet on one dataset.
• Ordinary life included. Office work, meetings, evenings at home. The mundane recordings are not filler — one of them produced the most personally useful finding here.

What it cannot be

• Not a clinical study. Fourteen saline electrodes, no source localization, no normative database, no blinding, no placebo.
• Not generalizable. Every number describes one person. That is the design, not a flaw — but it means nothing here transfers to you without testing.
• Not dose-controlled. Most legacy acute sessions have no reliable dose or timing record. The protocol introduced in September records the dose (a known, small quantity in LSD experiment 016) and is adding minute-level timing; legacy and prospectively recorded sessions are kept distinct throughout.
• Not free of confounds. The most important result on this site is a demonstration of how badly one confound distorted an initially exciting conclusion.