Intermission.

The current description

A symbolic trajectory of written form.

PCDVR deterministically recodes written characters as categorical states and retains their order as a trajectory. The current tests show measurable written-form information in state composition, transitions, boundaries, and complete paths. They have not established separate word-family information beyond the tested ordinary-spelling controls.

This is not a semantic embedding, an etymology detector, a validated phonetic or articulatory geometry, or a demonstrated improvement over spelling-based classification.

What the earlier tests measured

Order, then boundaries.

In the independent 48-word, four-family sequence challenge, ordered trajectories scored 68.8%; 2,000 randomized-order controls averaged 43.9%. One control matched or exceeded the ordered score (plus-one empirical p ≈ 0.001). This tested family classification, not prediction of the word itself.

In the separate 47-word boundary challenge, the raw final character scored 66.0%, the PCDVR final state 66.1%, and the full path 70.2%. Similar final-character and final-state scores motivated a direct test against ordinary spelling controls.

Experiment 22C · 96 reference / 48 challenge words · Four families

Testing what remains after spelling controls.

The declared model used edit distance, character unigram/bigram/trigram measurements, written-vowel sequence, length difference, a two-character prefix, and one-, two-, and three-character suffix features to predict PCDVR pairwise trajectory distance.

Mean R² across five word-blocked cross-validation folds was 0.606—about 60.6% of distance variance explained on average across folds. This is a distance-prediction score, not classification accuracy.

After subtracting the model-predicted component, residual-based family classification produced:

These are label-permutation controls, not the within-word sequence shuffles used in Experiment 21P. Of 2,000 controls, 1,307 matched or exceeded the observed residual accuracy. Plus-one empirical p = (1,307 + 1) / (2,000 + 1) = 0.6537.

Experiment 22 did not detect residual word-family signal under the declared spelling controls in this fixed dataset.

This does not prove that every possible spelling model explains the representation, or that residual signal is absent in every dataset. The central 95% permutation interval was 12.5%–39.6%; it describes the controls, not a confidence interval for generalization.

Companion checks · 22A / 22B

Other views of the same challenge.

Intermission, not erasure

Same nodes. Different paths.

LISTEN, SILENT, ENLIST, TINSEL, and INLETS still share a state inventory while following different ordered trajectories. That representational distinction remains; it does not by itself establish extra linguistic information.

The current classification phase is paused. Existing results, including negative findings, remain available. Future questions about resolution, diacritics, compression, or deliberately phonetic mappings would be new investigations.

Next · Star Branch · No results yet

A different geometry to explore.

Can a PCDVR constellation or trajectory be matched quantitatively to real astronomical geometry—and how do those matches compare with randomized controls?

This begins a new question, not an astronomical claim about names or an extension of the word-family results.