Built in 69933cf98 as StayAsleepVibe.kt: a one-pass sound walk that plays once and hands off — OSC bass sliding from the fourth down to the root under a single ringing kick, DX keys answering at twice the bass’s rate, off-eighth re-picks in the alley and the chase, street footsteps, a passing train, sirens, and one spoken line held back for the closing awake section. That line was tuned in 0b079532c and c4bf61abc so the setup and punchline land once, on the last downbeat, instead of repeating through the chase. A matching visualizer shipped alongside it: FaceMorphViz.kt draws a news broadcast on an old TV set, rabbit ears and all, whose story graphic is a face that dissolves eyes-first as the song plays — surging on loud passages, snapping to black-and-white on kicks, tearing on drum hits — introduced in 1c6bb331c.

Rust Belt is a swampy heartland-rock pocket at 87 BPM in D Dorian inspired by the Pretenders classic. The bass evolves the pocket, adding a round, fingered two-bar hook that never transposes until the chorus asks it to (IV → bVII → i; in dorian the major IV is the money chord).

What shipped. e6e1b9c53 added three pieces of arrangement machinery that were built together and share one wire: a storm voice, a per-edge transition-effect bank, and a per-track breathe cycle.

The storm is a 9th Pulsar voice, not a graph unit. pulsar_storm.h is 1,272 lines, seeded per instance and driven only by pulsar_rng, so a given seed always renders the same audio — no allocation, no statics, no logging after Init.

Rain is two fields at once, and voicing only one of them is what makes it read as a leak. Near field: Bernoulli dust, each drop a one-sample impulse into a band-pass it re-tunes on the way in, so every drop is its own size. Far field: a diffuse wash of the thousands of drops too distant to resolve, dark and gusting, rising super-linearly so it’s a hint behind a drizzle and the body of a downpour. Density is mapped geometrically, not linearly — a linear sweep spends its whole bottom half above the rate at which the ear stops resolving individual drops. The particle field and the smooth-random LFO are both adapted from Mutable Instruments Plaits (MIT, © Emilie Gillet).

Thunder is a five-octave fractal sum starting at 0.12 Hz (8.3 s) and doubling up to 1.9 Hz, undulating deep enough that troughs land 20 dB under the crests — the front swells and recedes as it travels rather than sitting at one level. Distance darkens the roll, but only so far: the pulsar’s output high-pass sits at 55 Hz, and a deeper tilt would push a far strike’s whole band underneath it and simply delete the thunder.

Lightning is a cascade, not a hit — and it falls. Four transients step down 3400 → 1900 → 950 → 420 Hz on widening gaps (32 / 42 / 52 ms), with tails that lengthen on the way down (35 → 130 ms) because a lower band rings longer. The tails overlap the next onset on purpose: the “clap-clap” is carried by the attacks, and a strike that went silent between steps was four ticks. A band-pass alone was a filter ping — the textbook synthetic transient, and exactly what the ear test called synthy — so a parallel grit path gives every hit a torn broadband body beside its pitch, a crackle modulator rips its amplitude, and a Lorenz stream wanders both the centre and the pitch/grit balance so no two cracks are the same shape. The descent runs crack → tear → boom → sub without a hole in it.

Then terrain echo: five irregularly spaced taps (37 / 71 / 118 / 179 / 247 ms) off one mono line with no feedback, so the tail is bounded by the longest tap by construction. The spacing is deliberately non-harmonic — evenly spaced taps comb, and a doubling chain reads as a tempo-synced delay rather than as ground. Distance is the whole dial and it runs the right way round: a bolt directly overhead barely echoes, a distant one answers off the hills for a quarter second.

Breathe is what makes it more than a weather layer. A bar-clocked swell sinks a track’s gain toward a floor and closes its tone with it, then rises back. In downpour the kit drops out and everyone else keeps playing under the roar — buried, not absent — breathing on co-prime periods of 2, 3 and 5 bars, so their swells only re-converge every 30 bars, far longer than the section lives. The ensemble audibly decoheres instead of just getting quieter. Then cloud break snaps every voice onto one shared 3-bar breath and walks the kit back in under full rain: the band reassembles in a single move. bars = 0 is exactly unity — the caller skips the audio path rather than multiplying by a computed 1.0 — so a vibe with no breathe renders bit-identically to a build without the feature.

68b00e82d wired the storm into the arrangement as one deliberate lap instead of a weighted graph: every section is worth hearing, and the downpour is worth hearing exactly once, which is also what pins the song near three minutes. 1439739 pushed the cloud break’s energy to 0.90 — above the verse it returns to, so the peak of the storm reads as a peak instead of a lull — and stretched its exit ramp to three bars so the macros have room to walk home before the band arrives. Recorded a few hours later.

Dog House is built on the Pulsar engine’s transformation from a beat machine into a generative song arrangement engine. At its core, Markov chord progressions drive the harmonic movement — weighted probability matrices determine which chord follows the next, so no two plays through the same section sound identical. The arrangement is a 6-section graph (intro → verse → chorus → solo → breakdown → outro) where transitions between sections are also probabilistic, creating songs that follow a familiar blues arc but never repeat exactly.

Four virtual band members — Drummer, Bassist, Keys, and FX — interact through personality traits (loudness, creativity, swing, drag) and handoff/pull-in matrices that govern who plays lead and who locks in behind them. Each member runs 32-step patterns with per-track bar strategies (Mutate, Fill, Call & Response) that evolve the patterns over time. In the solo section, Jam mode kicks in — band members lock into combos and riff on a shared lick, mutating it with each pass to create the spontaneous call-and-response feel of a live session. Dedicated Pulsar delay and reverb effect units (independent of the main voice chain) add spatial depth, while per-track mod LFOs and hold-step logic create evolving textures and drones beneath the arrangement.

This track is also the debut of the standalone DJ App, which packages Orphic modules (Pulsar, DJ turntable, Leslie, and Distortion) into a mobile-friendly UX with full Media Session integration across Android, iOS, and macOS — complete with a foreground service to keep the jam alive when backgrounded.

Itchy Scratchy puts the new DJ turntable front and center. The C++ DSP layer captures audio into a circular buffer and plays it back with cubic Hermite interpolation, giving scratches that warm, vinyl-like pitch response. Each deck can tap any source — Synth, Drums, Bass, or Master — and a constant-power crossfader blends between them without volume dips at center. On the Kotlin side, an MVI ViewModel drives a 60Hz physics simulation for platter momentum, while the Compose UI renders radial waveform platters and a vertical crossfader in the Cleveland Guardians palette.

Lazy Susan showcases the new Horn (Leslie rotary speaker) DSP effect. The C++ engine splits the input through a Linkwitz-Riley crossover into separate horn and drum rotor channels, each spinning with independent inertia to model the real-world lag of heavy rotating elements. Stereo amplitude modulation recreates the Doppler-like swirl that defines the Leslie sound. On the Kotlin side, a physics-based animation drives the rotor visuals in the HornPanel, with tactile controls for Speed, Ratio, Depth, Amount, Mix, and Brake. Rotor phase and audio peaks feed back into the engine’s visualization ring buffers, so you can watch the rotors spin in sync with what you hear.

Written at the Balch Hotel in Dufur, Oregon with Mt. Hood visible in the background. The day before (February 27th), the Mt. Hood lava visualization was added to the synth and rendering performance was optimized. On recording day, control event handling was unified via setPluginControl. The cozy and charming 1907 hotel room, watching the sunset behind Mt. Hood in the window with my fabulous wife Benedicte, was captured perfectly in this visualization and song.

The first track performed using Maestro Mode – conducting the synth with hand gestures via MediaPipe. Camera-based hand tracking was added on February 21st, along with CPU-saving inactive DSP plugin disabling. On recording day (Feb 22nd), Maestro Mode was reworked to use individual voice gating with Thumbs-Up hold control, ASL gesture control routing was made context-dependent, and state snapshotting was centralized for REPL and gestures. The core:foundation module had also been decomposed into specialized modules with a JVM 21 upgrade just two days prior. The aspect ratio (1006/1080) captures the portrait-mode camera view of the hand tracking in action.