Processor Fault

s3g Processor Fault 8ch is a zero-audio-input, eight-output CLAP instrument that turns a generated byte field, an arbitrary file, or the waveform ancestry of a PCM WAV into continuously changing audio. It combines raw-data scanning, stochastic breakpoint geometry, damaged codec models, receiver-driven bass reconstruction, protocol modulation, and a bounded pressure, wavefold, and resonant-feedback stage. It can run continuously or respond as a velocity-sensitive MIDI instrument.

Raw-data scanning reads bytes as a moving waveform, while stochastic breakpoint geometry connects selected values with evolving curves. The codec section then introduces prediction errors, quantization, missing blocks, or damaged reconstruction as sound-generating processes rather than as file-format emulation.

The editor organizes its upper controls across three pages. Use SOUND for the field, codec, FM-style algorithm chart, M1/M2/M3 route summaries, and immediate BASS CORE amount. Use BASS LAB for receiver, pitch, body, high-gain character, articulation, and low-width controls, and open MOD LAB to edit every operator setting. MOD ON/MOD OFF remains visible on all three pages and bypasses the operator graph without disabling the bass core. Patch actions, performance mode, and the envelope editor stay in the lower strip. In a reduced host window, scroll the editor to reach controls outside the visible area.

Workflow

1. Set a REAPER track to eight channels and insert s3g Processor Fault 8ch.
2. Route the eight ordinary output lanes to the intended main speaker outputs, a downstream direct panner, or Output Autogain Stereo/Quad for fold-down monitoring.
3. Choose a preset or begin with INIT. Use OPEN ANY to replace the generated field with any file. A recognized WAV opened from INIT selects WAVE TRACE.
4. Use FREE for continuous sound, or select MIDI and play notes from the DAW.
5. Set SCAN, GEOMETRY, CHAOS, and FOLD for the underlying motion.
6. Choose a codec, then use RATE, BITS, CARRIER, and DAMAGE to break its continuity.
7. Raise BASS CORE, then open BASS LAB to choose how the codec becomes low-frequency excitation. Start with DIVIDE for stable subharmonics, DEMOD for contour, or ERROR for percussive faults; use TRACE to retain the raw codec identity. Add FUZZ, shape it with METAL, and introduce FEEDBACK last.
8. On SOUND, choose a CURATED CIRCUIT and read its operator flow in the chart. Start with a factory patch or CURATED RANDOM for a bounded variation, then open MOD LAB to change the algorithm or an operator's source, destination, rate, ratio, index, feedback, clock, or envelope response.
9. Raise DRIVE, SHRED, and RESONANCE to expose the byte-field contours and reinforce the modal body.
10. Choose a ROUTE and set SPREAD for the eight-channel distribution.

A useful first experiment is a low displayed codec update rate with high FOLD, followed by enough DRIVE and SHRED to turn the held vectors into pronounced curves. Then raise BASS CORE, select DIVIDE, and lower TRACE until the codec's carrier crossings resolve into a low body without disappearing completely. Raise FUZZ for a compressed upper growl, move METAL to find the desired edge and low-mid cut, then add restrained FEEDBACK to lengthen that texture.

Performance

FREE runs the generator continuously and bypasses the amplitude envelope. MIDI makes the field engine a monophonic, last-note-priority voice. Note-on retriggers the attack from the current envelope level, velocity controls voice level, and releasing the active note returns to the most recently held note. When no notes remain, the release runs to silence and the field freezes once the envelope becomes idle.

MIDI note 60 (C4) is the patch's reference note. KEY TRACK determines which domains receive its equal-tempered ratio: SCAN changes traversal and breakpoint timing while leaving the bass target fixed; BODY keys the divider and modal-filter target while the field keeps its free-running scan rate; and BODY + SCAN keys both. OCTAVE places that target at C2, C3, or C4 for -2 OCT, -1 OCT, or UNISON (zero-octave offset), and GLIDE smooths changes between held notes. Codec crossings can pull the measured result away from an exact oscillator pitch, especially with loaded material, so tune by ear when the source has strong irregularities.

A MIDI note-on also excites EXCITE, using velocity to accent the codec-derived attack without bending the body pitch. In FREE mode the receiver remains continuously driven by codec crossings, envelopes, and damage rather than retriggering a separate drum voice.

The envelope graph is directly editable. Drag the attack and release endpoints horizontally, the decay breakpoint in both axes, and the sustain breakpoint vertically. The dashed live-level trace and highlighted breakpoint follow the active MIDI envelope stage.

PLAY and STOP fade and freeze both performance modes. Presets, CURATED RANDOM, and MUTATE shape the timbre without changing the selected mode or ADSR settings. In MIDI mode, any modulation operator set to ADSR follows this envelope at its selected index, so attacks can assemble a modulation stack and releases can withdraw it without changing the stored depth. FIXED runs that operator at full index. In FREE mode, both settings behave as fully open.

Byte Field and Geometry

The default source is a deterministic 65,536-byte virtual field. GEN FIELD captures the currently selected CODEC as a hidden generation profile, rolls a new seed, and builds data chosen to expose that codec's characteristic behavior. The source readout shows the captured profile in brackets. Changing CODEC afterward does not rewrite the field, so the same material can be compared across codecs; press GEN FIELD again to tailor a new field to the new selection.

The profiles use different data grammars rather than post-processing the same noise: differential and predictive modes receive continuous slopes with controlled innovations; companders receive quiet detail and sparse peaks; speech-like modes receive voiced, residual, and excitation structure; DISC receives repeated blocks and markers; SUBBAND receives separated slow and alternating detail; TRANSFORM receives sparse short-block spectra; MODEM receives symbol plateaus; FAX receives repeated scanlines; SIGMA 1-BIT receives slowly changing oversampled curves; APT and HF FAX receive sync, phasing, telemetry, and image regions; HELL receives doubled raster glyphs; MORSE and SPARK CW receive dot-dash keying fields; BAUDOT RTTY receives start/data/stop frames; and SSTV receives sync, porch, separator, and color scanlines. For an arbitrary file, OPEN ANY reads from byte zero and maps every byte to an unsigned 8-bit bipolar sample.

When OPEN ANY recognizes a supported WAV file, Fault parses its RIFF chunks, jumps to the actual data chunk, and decodes integer PCM at 8, 16, 24, or 32 bits or floating-point PCM at 32 or 64 bits. It applies bounded gain recovery, remaps every source frame into eight related lanes, and requantizes those lanes into the byte field. Mono becomes eight short-offset traces; stereo becomes four related pairs; sources with eight or more channels retain eight native lanes. This preserves contour, rhythm, and some pitch identity without retaining a recoverable clean signal. Files that are not recognized as supported WAV data are read literally from byte zero.

Raw files and the prepared eight-lane field are limited to 64 MiB. Longer sources show PARTIAL in the source readout. A shorter or irregularly sized source repeats as needed to form a power-of-two scan ring. The codec BITS control requantizes the field again later in the chain.

EVOLVE is free-running rather than transport-synchronized. It changes regional content without replacing the whole field, so a patch can retain its identity while its local byte relationships drift.

Codec Damage

The codec section treats the generated or file-derived signal as material for destructive encoding. These modes are synthesis models, not interoperable media codecs. Their design sources are collected in the Codec references.

• PCM applies direct bit quantization.
• DELTA quantizes the difference from a running predictor.
• ADPCM quantizes prediction error with a multiplicatively adapted step: large codes expand it quickly and small codes contract it slowly.
• MU-LAW and A-LAW encode and decode eight-bit G.711-style sign, segment, and mantissa codewords. BITS removes low codeword planes below eight bits, while DAMAGE flips code bits.
• CELP uses damaged pitch, gain, and codebook behavior. Missing packets replay pitch-synchronous history with progressive attenuation, followed by a short recovery crossfade. It is not a G.729 encoder, decoder, or bitstream implementation.
• DISC introduces block errors whose duration selects the response: short gaps are interpolated with a 32-sample lookahead, medium gaps repeat or extrapolate fading history, and long gaps decay toward mute. It does not implement CD or DVD framing, modulation, or error correction.
• CVSD tracks the signal with a one-bit adaptive slope. Three identical decisions expand the step; alternating decisions contract it; bit errors and a clock-relative reconstruction filter expose failure.
• SUBBAND uses a downsampled two-band Haar quadrature-mirror bank with independently damaged adaptive quantizers for its low and high bands.
• LPC replaces increasing amounts of prediction residual with pulse and noise excitation through an adaptive all-pole model.
• TRANSFORM performs 32-sample DCT blocks with a damaged scale factor, frequency-weighted coefficient bit allocation, missing blocks, and coefficient-bit upsets.
• PREDICT uses a second-order predictor with quantized residuals, missing corrections, and occasional residual upsets.
• MODEM maps amplitude into damaged FSK-like symbols and carrier tones.
• FAX maps level and slope into fixed QPSK, eight-QAM, or sixteen-QAM constellations, then exposes raised-cosine symbol transitions, training sequences, equalizer lag, and phase slips.
• SIGMA 1-BIT uses a first-order integrated error-feedback loop followed by a bit-clock-relative reconstruction filter.
• HYBRID combines a predictive base layer with a correction layer that can thin, vanish, or invert.
• APT treats the signal as weather-satellite image brightness and amplitude-modulates a 2400 Hz subcarrier. Its 2080-word composite line carries paired image regions, telemetry wedges, minute markers, a seven-cycle 1040 Hz sync train, and a seven-pulse 832 Hz sync train. DAMAGE adds word errors, stale video, line slips, fading, carrier phase jumps, and receiver noise. It models the Automatic Picture Transmission sound and line grammar, not an interoperable RF transmission.
• HF FAX scans brightness as frequency between the historical 1500 Hz black and 2300 Hz white limits around 1900 Hz. Each line begins with phasing and start bars; DAMAGE introduces wrong pixels, black dropouts, line slant, flutter, and phase slips.
• HELL is a Feld-Hellschreiber-style 5-by-7 raster teleprinter with doubled vertical pixels and keyed 900 Hz ink. Clock mismatch and DAMAGE smear, repeat, drop, or invert raster cells.
• MORSE converts the field into International Morse letters, preserving one-unit dots and intra-element gaps, three-unit dashes and character gaps, and a keyed 700 Hz tone. DAMAGE drops elements and adds chirp and key-edge roughness.
• SPARK CW uses the same Morse timing but replaces the clean continuous-wave tone with damped, chirping, harmonic, noise-rich spark bursts.
• BAUDOT RTTY sends ITA2-style five-bit letters with a start bit and stop interval as 170 Hz-shift FSK around 1700 Hz. It includes letters/figures shifts; DAMAGE flips data and shift decisions.
• SSTV uses a slow-scan-television-style line: 1200 Hz sync, 1500 Hz porch and separators, then three 320-pixel color scans mapped from 1500 to 2300 Hz. DAMAGE causes false sync, frequency errors, and horizontal line slips.

Bass Core

BASS LAB reconstructs a coherent low voice from the selected codec rather than placing an unrelated sub oscillator underneath it. Codec and modulation crossings clock a rounded binary divider; the selected receiver weights that divided signal with the waveform, transmission envelope, or codec errors; and a moderately damped modal filter focuses the result around the selected pitch. Because the low cycles must come from the running codec, their timing and grain retain the patch's transmission character. TRACE determines how much of the raw Fault path remains audible. The compact BASS CORE control on SOUND is the same BODY amount shown in the lab.

The high-gain path is part of the reconstructed body, not a second global distortion stage. A pitch-following split keeps a clean low anchor outside the nonlinear loop while the upper body enters a fuzz and contour circuit. Its short regeneration time follows the body pitch, with DC filtering, bandwidth limiting, saturation, and an envelope governor keeping the loop bounded. This lets a note acquire sustained abrasion and upper resonance without allowing feedback to replace the fundamental with low-frequency rumble. With both FUZZ and FEEDBACK at zero, this path is inactive, so METAL has no audible effect.

The bass core and the modulation graph can reinforce one another, but they have separate enable paths. M1, M2, and M3 may still address BODY, RING, or STRIKE to animate the same resonant region, while their codec, clock, and data changes alter the crossings that generate the subharmonic. MOD OFF removes those operator contributions without switching off the receiver, dedicated pitch, excite amount, trace blend, or low-width processing.

Protocol Modulation

Protocol modulation adds three operators, M1, M2, and M3, and six selectable stacking algorithms to the codec. Like FM synthesis, one process acts on another, but an operator is not limited to a sinusoid and its destination is not limited to oscillator frequency. Raster sync, telegraph keying, framed bits, image scanlines, field data, and the codec's own output can become control signals, turning transmission structures into a compact set of timbre controls.

All three operators share the same source vocabulary. The tonal sources are SINE, TRIANGLE, and stepped NOISE. FIELD follows the current byte-field signal; SYNC and GATE supply narrow and square pulses. APT, HF FAX, HELL, MORSE, SPARK CW, BAUDOT RTTY, and SSTV reuse simplified line, raster, keying, framing, and scan grammars as modulation shapes. FEEDBACK derives a bounded operator from the codec's recent output, turning the carrier/modulator distinction into a recursive loop.

The SOUND page draws the selected topology as an FM-style node chart: arrows show inter-operator influence, BUS or MUX nodes show parallel and time-slotted combinations, and the final node identifies the independent destinations. The three text summaries beneath it retain each operator's source, destination, and envelope mode. Its CURATED CIRCUIT menu installs one of two musically bounded operator recipes for the chosen topology while preserving the rest of the patch. MOD LAB repeats the live chart beside the full operator cards; its ALGORITHM menu changes only the graph, leaving the deep settings under direct control.

Protocol sources use sync regions and alphabets as compositional shapes; they are not interoperable encoders. Start with BODY, RING, STRIKE, FOLD, or SCAN to turn telegraph and image grammars into pitch, decay, impact, shape, or motion. Use CARRIER, DEVIATION, DATA, or DAMAGE when a more broadband, fractured result is desired.

Integrated Nonlinear Stage

The shaping stage follows the codec so held steps and prediction errors become wave geometry rather than remaining quiet digital artifacts.

• DRIVE raises level into pressure and saturation.
• SHRED moves the nonlinear path toward stronger folding and edge generation.
• RESONANCE raises a bounded internal feedback loop. Low settings follow a smoothed spectral-center estimate; above roughly one third, the loop progressively settles into a darker fixed body. BODY modulation can pull that body down by another two octaves, while the dedicated modal voice guarantees a coherent bass result even when the selected codec has a strong radio-frequency carrier.
• OUTPUT sets the final trim after internal calibration.

Feed-forward calibration and a slow, bounded level correction reduce large loudness jumps among codec and shaping settings. They do not normalize away transient or spectral differences. The final path removes DC and limits output to finite sample bounds.

Eight-Channel Routing

ROUTE changes how each output lane addresses the byte field:

• PARALLEL keeps lanes near a shared read path.
• DEINTERLEAVE distributes adjacent field data across eight lane-strided reads. For a prepared WAV, it follows the eight derived waveform lanes directly.
• PLANES offsets lanes into separate field regions.
• SHUFFLED uses repeatable hashed read addresses.
• DIVERGENT gives lanes different strides and long-term trajectories.

SPREAD operates on the mean plus each channel's residual. At zero, all lanes collapse toward the instantaneous eight-channel mean. At one, their full differences remain. Bounded energy compensation prevents spread changes from acting mainly as gain changes.

The display shows eight recent output waveforms. Each trace scales itself to reveal shape and is therefore not an absolute level meter; use the peak readout or a following Analyzer Meter for level comparison.

Presets and Patch Actions

INIT provides a neutral starting point with M1, M2, and M3 off, BASS CORE at zero, and the high-gain path closed. Each of the twelve active presets stores the complete modulation graph and bass setup; across the bank, every algorithm appears twice. M2 and M3 follow the ADSR, while M1 also follows it in crossed or transcoding presets whose complete stack is articulated together. Editing a sound control changes the menu state to CUSTOM.

• PLAY resumes the field from its frozen position with a 20 ms fade-in.
• STOP fades to silence over 20 ms, then freezes the synthesis engine without resetting it.
• OPEN ANY loads any file as a byte field. Supported WAV files are decoded automatically into eight lossy waveform-derived lanes; other files are read literally from byte zero.
• GEN FIELD leaves file mode, captures the selected codec as the field profile, and creates new codec-oriented source data while retaining the remaining control settings.
• CURATED RANDOM chooses one of the twelve proven modulation and bass circuits, retains its sub or percussion identity, then varies field motion, codec settings, receiver, body, excite, trace, fuzz, metal voicing, bass feedback, glide, low width, operator timing, ratios, indices, operator feedback, and nonlinear controls inside conservative limits. It does not invent an arbitrary source/destination graph. The guardrails favor BODY or BODY + SCAN key tracking, the lower octave settings, a centered low band, moderate texture and chaos, low codec damage and shred, bounded target-specific indices, and useful resonance; they exclude deviation, data, and damage from generated primary routes.
• MUTATE makes smaller changes around the current patch and usually preserves its seed, codec family, receiver, octave, and circuit. Occasional circuit, receiver, or octave changes stay inside the same bass and destination limits, so repeated mutations remain within the curated operating range.
• UNDO restores the state before the most recent preset, source, field, randomize, or mutate action.

Preset, source, field, codec, bit-depth, and routing changes crossfade instead of resetting abruptly. The automatable Run parameter mirrors the PLAY and STOP buttons. REAPER project state stores the transport state, performance mode, ADSR settings, complete operator graph, complete bass reconstruction and high-gain setup, field seed, captured generation profile, and external source path; held notes and source bytes are not embedded. If an external file is unavailable, Fault uses the generated field and identifies the missing source in its header; use OPEN ANY to locate or replace the file.