Signals,
before stories.

Twenty pinned recordings yielded 9,600 candidates and 174 fixed selections. Six complete new-session participants support four comparisons; adjusted p values are 1.00, 0.25, 0.50 and 1.00.

Only two new-person participants met the input rules, so that cohort remains unestimable. Independent recorded-ledger reproduction verified all 4,002 event partitions; neither this result nor model agreement establishes semantic meaning.

We measured extrema, inflections, cycles and bursts on the same 122 selected segments, retaining every candidate and rejection across 2,806 event partitions.

Event-to-encoder alignment differs between CodeBrain and CBraMod. Adjusted p values are 0.50, 0.25, 1.00 and 1.00. Recurrent numerical turns do not yet establish universal tokens or transitions between brain states.

CodeBrain and CBraMod receive the same 122 selected wave segments and controls. Within-block original-minus-phase contrasts are positive in all five paired people.

Geometry and change contrasts have adjusted p = 0.125. Five-person test resolution, learned priors, shared recording effects and unknown training overlap keep the universal-geometry question open.

A frozen quality census checks 5,760 blocks across already examined recordings. It selects 122 time-distributed inputs; five people supply sufficient paired-night support, and the pretrained encoder completes 610 passes.

Geometry correlations are small, even where they exceed the specified shift controls. Change-profile tests do not clear correction. Every rejected and qualified-but-unselected block remains traceable.

A fixed public EEG encoder runs on nine of 240 candidate segments, with four waveform controls each. The strict quality rules leave no complete eligible participant pairs, so the planned agreement tests cannot be estimated.

All exclusions and segment observations remain visible. The result establishes a reproducible model execution path, without establishing universal tokens or brain meaning.

Twelve new ear-EEG recordings add 86.38 qualified hours. Six participants each contribute two nights; the frozen comparison analyzes 48 hours and asks whether session summaries recur across nights.

All four documented ear channels are used. No stage or personal labels enter numerical discovery. Stable person or sensor effects can contribute to recurrence; this does not establish universal tokens.

The full open around-ear inventory contains 55 recordings. Fifty-four qualify, totaling 223.47 sample-hours. We compare the timing and geometry of changes in waveform shape, spectrum and sensor coordination, with phase and nuisance controls.

The methods receive numerical waves without identity or task labels. These measurements test repeatable signal structure; they do not establish a universal brain language. Experiment 005 is a prepared Neurable protocol, with no physical collection yet.

Eight around-ear recordings were checked numerically and tokenized with three unsupervised feature families. All 27 seed and vocabulary settings were retained and reproduced.