Cadence

What cadence is

Cadence is the rhythm of every sentence the firm writes. The single value the parameter takes is measured — sentences carry a median length of 14 to 18 words and a mean of 12 to 20, varied across the passage rather than monotone. Measured cadence permits short sentences for emphasis and a single long sentence per paragraph, and it caps subordination at main plus one level. The same cadence applies to a memo, a working paper, an alert, and a documentation page without softening.

One value, measured

The parameter takes exactly one value across the system, and the typed module rejects any other string at parse time. The fixed value carries an invariance: measured cadence applies to every surface, and the validator runs the same closed rule set against each source file. Length pressure does not relax the rhythm, and brevity pressure does not flatten it: a short label still varies its sentence shapes, and a long passage still holds the median in band.

Sub-properties

Cadence declares 14 sub-properties in its typed source. The length axis carries target sentence length, sentence-length distribution, short-sentence clustering, and long-sentence load. The structural axis carries clause depth, rhythm variation, parallel structure, and inversion. The punctuation axis carries comma density, em-dash use, semicolon use, and colon use. The boundary axis carries opening-sentence and closing-sentence cadence — the first sentence of every paragraph is claim-bearing, and every section closes on a sentence that lands.

Of the 14 sub-properties, 7 are backed by 7 typed rules in the validator, all 7 implemented and running at validation. One additional rule sits at runtime-only status pending a UI-source check. The remaining sub-properties are specified in Voice-Addendum.md and await typed rules.

Cadence visualised

The first exhibit below renders the box-plot atom for the first time on canon. The atom takes its claim from Design-System.md § Diagram › Box plot — distribution comparison across groups. Two boxes carry the sentence-length distribution across canon’s two channels — face and voice. Each distribution is computed at build time from every page in the channel; the box marks Q1, the median, and Q3, the whiskers reach the 1.5-times-IQR bound, and outlier sentences sit beyond as small filled circles. A horizontal accent band marks the 14-to-18-word target the median rule guards. The toggle below the diagram exposes the same five-number summary as a table for screen-reader and keyboard users.

Typed rules

The second exhibit below renders the typed rules read directly from @vaquum/voice/parameters/cadence. Each row carries the sub-property the rule guards, the rule’s prose description, the mechanism the validator uses, and the declared severity. The single-rule-per-row layout matches the reference-table form already used on register, temperature, precision, and compression.

Mechanisms

Cadence carries two mechanisms in its typed rule set, drawn from the closed set of four — regex, dictionary, parse, and editorial — declared at the validator level. One rule is a regex check on em-dash spacing. The remaining typed rules run as parse operations on the sentence and paragraph structure of the source file, drawn from the validator’s parse engine. No cadence rule loads a dictionary, and the editorial mechanism is reserved for sub-properties whose check the validator cannot decide alone.