The Oscillome · 23 — Corrections
Every one of these was a confident, published conclusion that later evidence overturned. The most recent, and the most costly, is at the top. They are kept here in full, because a self-experiment that only records its successes is not measuring anything.
Claimed: a four-recording non-psychedelic resting group showed that psychedelic-level alpha appears without psychedelics. That group was the basis for the state-matched comparisons in §16, §07 and §15.
Reality: three of the four — now DMT experiment 001, DMT experiment 007 and DMT experiment 010 — were DMT sessions. The group was three-quarters psychedelic. The reason it resembled the psychedelic group is that it largely was one.
Cost: the psychedelic-versus-rest comparison is withdrawn entirely; one in-cohort non-psychedelic resting recording remains. The §16 core finding survives because it rests on a within-session eye-state contrast rather than on that group. Collecting confirmed non-psychedelic resting sessions is now the archive's highest-priority gap.
Claimed: comparing the archive's first recording with its last showed a 57% loss of standing-alpha organization and a doubling of the gamma share, presented as a possible change across the months.
Reality: the first endpoint, DMT experiment 001, was archived as a baseline but was a DMT session (Correction 04). The comparison set a DMT session against office work; it said nothing about how the months changed anything.
Cost: the comparison is withdrawn (§09), and the long-term question is now asked with trends fitted across every recording in each group.
Claimed: the archive's own self-supervised transformer separated the session labels in 29 of 37 held-out sessions, a result since superseded, and it was presented as the embedding result.
Reality: retrained on all 42 recordings in release 4.5, it does not beat the time of day alone, and it fails eye-state matching: 31 of 40 held-out sessions, against 27 for the time of day alone and 35 for sixteen hand-built spectral features.
Cost: the embedding result moved to the published foundation models (LaBraM 39 of 40, REVE 37), with time-of-day, eye-state and hand-built baselines beside every score (§21).
Claimed: every psychedelic session showed elevated alpha, contradicting the literature.
Reality: the non-psychedelic comparison group was office work. Against the same person's eyes-closed baseline, 11 of the 12 sessions labeled acute at the time are lower. The archive replicates alpha suppression.
Cost: every acute-vs-non-psychedelic contrast in the project had to be recomputed.
Claimed: LSD's alpha acceleration, replicated across five sessions, was a publishable brief report.
Reality: it replicates Carhart-Harris et al. 2016 (PNAS; MEG in 14 of the 20 participants; peak alpha about 10 Hz under placebo and 12 Hz under LSD)1. Not novel. The LSD-versus-other-psychedelic contrast rested then on 5 against 7 uncontrolled sessions at p ≈ 0.02 (6 against 7 now, p = 0.014, §17), and its one internal control — a within-session ramp — vanished at proper spectral resolution.
Cost: the claim was downgraded from finding to hypothesis.
Claimed: harmala alkaloids should drive an olivocerebellar oscillation at 8–12 Hz, making ayahuasca uniquely tremorigenic.
Reality: the head accelerometer says otherwise. The share of head motion in the 8–12 Hz band (8–12 Hz power over 2–8 Hz power) is highest under LSD (median 0.68) and psilocybin (0.64); ayahuasca (0.50) sits barely above ordinary sessions (0.44). Ayahuasca's dominant head-motion frequency, 6.12 Hz, lies outside the 8–12 Hz band altogether, against 8.78 Hz for LSD and 9.00 Hz for psilocybin.
Cost: a planned prediction was demoted, and tremor monitoring was extended to every future session.
Individual alpha frequency is treated in the literature as one of the most trait-like measures in EEG, with heritability estimates near 80%. In this archive, its intraclass correlation across recordings is 0.15. Within-recording variability (SD 0.74 Hz) is more than twice between-recording variability (0.31 Hz).
Minute-to-minute wander exceeds every effect being chased. That single number is why the session — not the window — is the unit of analysis here, why n for LSD is its six sessions rather than the hundreds of windows they contain, and why the next phase of this work is an active perturbation experiment rather than more passive recording.
Correction 02, revisited. That correction said the within-session LSD ramp "vanished at proper spectral resolution". In the LSD-labeled session added in v3 the ramp is present at 0.125 Hz resolution — the 1/f-corrected posterior peak rose about 1 Hz across the hour, as a step after the third sweep more than a slope — and the driven resonance rose in step, which the earlier sessions had no way to show. One dosed session (a known, small quantity), with the dose clock time unlogged, and two short non-psychedelic sessions with positive slopes of their own: the ramp is back as an observation after a dose, not yet as a psychedelic effect.
Correction 03, checked. The 8–12 Hz accelerometer share was lower during the LSD-labeled sweeps (0.28–0.36) than during the non-psychedelic ones (0.48–0.61) while the resonance sat higher, so the resonance is not simply tracking head motion, although the sweep with the highest f0 also carried the most absolute 8–12 Hz motion.
The ICC of 0.15, split by timescale. Between non-psychedelic days the driven f0 repeated to 0.03 Hz while the same days' spontaneous peaks differed by 0.40 Hz: part of the between-day wander is peak-picking. Within the LSD-labeled session the driven f0 moved about 1 Hz and the spontaneous peak moved with it: the within-session wander is real, and the resonance itself moves. §18.
A single-subject EEG archive: 56 recordings, 100.1 hours, 19 acute sessions, four psychedelic classes, 34 photic resonance sweeps, thirteen light-noise runs, one brain, February to October 2026. All 56 recordings form the analysis cohort.