RE
LI
A
BLE
180G{DATA}
014
2026
SPOKEN WORD
GENERATIVE
∞ : ∞
THE WORK
A generative cut-up poetry engine. Source material — subjects, actions, conditions, endings, phrases, and narrative inserts — is recombined in real time by a probabilistic grammar that shifts with an internal intensity arc. No session is the same. The narrator is always present and never to be trusted.
The B-side externalises the grammar as concrete poetry: every fragment rendered as text, building a palimpsest of the session in real time. Voice is muted. Only the beat continues.
FORMAT
Single-file browser work. No server. No memory. Each session a unique pressing. The piece runs until the window closes.
LABEL
180G{DATA} — 014 — DORKHOLM — 2026
SOURCE MATERIAL
Six fragment pools: 50 subjects [s], 50 actions [a], 50 conditions/conclusions [c], 50 exit/glitch commands [e], 62 phrases [p], 25 narrative pieces [n]. Pools are scanned at load time via HEAD requests — only confirmed files enter the manifest. An anti-repetition window of 6 prevents consecutive reuse per category.
GRAMMAR ENGINE
Each transmission is a probabilistic sentence construction. Roll weights: 8% glitch (isolated [e], possibly stuttered), 10% phrase ([p] with turntable treatment), 5% narrative ([n], never consecutive), remainder constructs a sentence from the pools. Sentence structure varies by an intensity state (slow random walk, updated every 20–40s): full s/a/c/e, headless a/c/e, elliptical s/e, or fragment s/a. Endings may stutter (2–3× near-glued repetition). Silence markers are injected between clauses at low probability, weighted by intensity.
AUDIO CHAIN
Built with Tone.js. All buffers are cached on first load; subsequent reuse is zero-cost. Playback timing uses Tone's Web Audio clock — buffers are loaded in parallel before the clock is touched, so scheduling is never stale. Voice chain: Volume → BandpassFilter(1200Hz, Q 0.8) → Freeverb(room 0.35, wet 0.28) → Limiter(-6dB). Narrative chain: same path but bypasses bandpass, lighter reverb (room 0.18, variable wet 0.08–0.26 per clip). Phrase processing: per-fire randomised turntable treatment — drag (rate 0.82–0.92), pitch-up (1.06–1.18), bit crush (5–6 bit), highpass (300–800Hz), or wide rate variance.
BEAT ENGINE
Two stems (b.wav, b1.wav) loop continuously from first play — position is never reset. An exclusive crossfade engine (7s fade-out, 5s overlapping fade-in) alternates between stems every 45–90s. A coin toss on init determines which stem opens. Playback rate drifts ±5% every 4–12s. No sidechain compression — the two stems never overlap in volume.
B-SIDE
Same grammar engine and timing as A-side. Voice and narrative audio chains are muted. The beat continues. Each transmission renders as concrete poetry: grammatical role determines font size and style. Lines accumulate bottom-to-top, fading through five opacity grades as new material arrives.