Write songs by choosing every note

01 · Role in the system

What Jev does here

Three stages of choices: one setup call picks mode, tempo and form; one call per bar chooses among seven described triads; one call per note chooses among about twelve pitches and a rest plus a length. Code enumerates only the legal options and attaches the facts a musician would weigh - interval from the last note, chord tone or tension, whether a leap resolves, how far the cadence is - so Jev cannot break the song, only prefer. The README publishes a verbatim first response where code ranked the opening F#dim last of seven and Jev, reading the mood restless, put 42 percent on it in 569 ms. Every request and response is journaled to JSON, and the site replays each song at the speed it was written with the full distribution beside the sheet. The author reports about 250 ms and 1,600 input tokens a call; a 32-bar song is 60 to 80 calls, twenty seconds, half a cent.

02 · Control boundary

Where Jev sits

One Choice per musical decision over code-enumerated legal options with attached facts; determinism by construction, and a JSON trace drives the replay.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Musical quality is bounded by the facts code attaches and the diatonic option sets; Jev only prefers among them.
• Latency and cost figures are author-reported from the author’s own runs.
• Built in an afternoon for fun, per the README; the five published tunes are the corpus.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• beingcognitive ↗Community · github · public · checked 2026-09-22
• beingcognitive ↗Community · demo · public · checked 2026-09-22