Backtrace is a reverse-reverb / preverb workstation built by Sound Detective / Detective 47. A normal reverb answers a sound after it happens. Backtrace files its report first: a reversed swell that rises out of silence and lands exactly on the note that caused it.
Two engines share one panel. LIVE PREVERB is real-time and locked to your DAW's tempo. A worker thread renders the selected algorithmic reverb's tail for a unit impulse, reverses it, shapes it, and loads it as a convolution kernel. The dry signal is delayed by the length of that kernel, and the delay is reported to the host as latency — host delay compensation pulls the entire plugin earlier in time, and every note blooms a swell that leads into itself.
THE DUST VAULT is the capture-and-print engine. Record or import audio, render a forward FX tail from one of six delay flavors and five algorithmic reverbs, and reverse it into a tempo-fit printed swell. Then trim it, fade it, sweep a filter across it, print it, and drag the finished WAV straight back into your DAW.
Three house rules separate Backtrace from the usual suspects. The reversed reverb is generated from real algorithmic reverbs — never noise washes, never sampled IRs. It never time-stretches — a short tail is padded, a long one is decay-fitted, so what you hear is always the real bloom. And everything musical is host-automatable, so the swell can be played like an instrument.
Reverb is hindsight. This is premeditation.
Understanding the order of operations is the difference between using Backtrace as a preset box and using it as an instrument. Three chains matter: the real-time path you hear, the kernel build that feeds Live mode, and the offline print line inside the Vault.
THE CAPTURE TAP
The Vault's capture tap sits first, before anything else touches the signal. It snapshots the input exactly as it entered the plugin, so nothing downstream — not the mix, not the filters, not the output gain — ever colors a captured take. The evidence stays clean.
THE AUDITION LANE
While an audition runs — Play Swell, Play Source, Audition Tail, Preview Live — the playback buffer replaces the output entirely. You judge the printed evidence on its own, not smeared under the live input. A 15 ms ramp on stop means the tape never clicks off.
THE LIVE PATH
The wet feed is the input through a real-time pitch shifter (scaled by BLEND), through the WET-mode output filters, then convolved with the reversed-reverb kernel. The dry copy is delayed by exactly the convolution head plus the kernel length plus the shifter's latency — and that exact number is reported to the host. The delay is not overhead. It is the mechanism. Dry and wet meet at a soft ceiling of ±0.8.
THE OUTPUT STAGE
Everything — live output, auditions, passthrough — exits through the smoothed OUTPUT gain and the global HPF/LPF. The scope taps the signal after that filter, so the Evidence Scope draws exactly what leaves the plugin. The display never flatters.
THE KERNEL WORKSHOP
Kernels are never built on the audio thread. A worker fires on any tail-affecting change, debounced by 100 ms so a knob sweep coalesces into one build. Two convolution engines run in tandem: a new kernel loads into standby, settles for a full kernel length, then crossfades in over roughly 85 ms. Live tweaking never drops out.
THE PRINT LINE
The offline print never time-stretches. It decay-fits: render the forward tail, measure the −40 dB point, adjust the reverb decay — up to three passes — until the tail lands at the window. Stage two then runs the same applyEdit chain for every destination. Hear equals print equals export equals drag.
The top plate is the real-time strip. Left side: the TIME knob, the FEEL box, PREVIEW LIVE, and the LIVE PREVERB toggle. Centre: the Evidence Scope with its LIVE VIEW toggle. Right side: WET, SHAPE, PITCH and BLEND, plus the octave −/+ buttons. A status label reads the current mode and the latency being reported to the host.
Engages the real-time preverb engine. Switching it on triggers a kernel build and moves the processor from passthrough onto the convolution path. Reported latency jumps from zero to the full pre-roll length — the host pulls the plugin early, and the swell arrives in front.
Engage it while the transport is stopped, then play. Hosts apply delay compensation on playback start. See FILE 10 and FILE 11.
The stepped pre-roll length — how far in front of the note the swell begins. The note value is converted through the host BPM (120 if the host reports none) and clamped between 0.05 s and 5.0 s; anything longer means impractical latency and heavy convolution.
The panel tags the range you are working in: MICRO at 1/16 and below, TIGHT from 1/8 to 1/4, PREVERB at 1/2 and above.
Multiplies the note length by 1, 1.5, or 2/3 before the conversion to samples. Dotted pre-rolls run longer; triplets run tighter.
Trim on the convolved swell. The kernel itself is energy-bounded at an L2 norm of 0.40, so 150% is a usable push, not a blast.
The exponential build-up envelope on the kernel. 50% is the by-ear-confirmed default. Below it the rise is gentler and energy is present earlier; above it the energy arrives later, for a more dramatic bloom.
The extremes are deliberate effects. Above 0.8 a smoothstep gate silences up to the first 25% of the pre-roll, then surges — hold, then bloom. Below 0.2 an early floor makes the swell drag in audibly from the very start.
In Live mode this is a dedicated real-time shifter on the wet feed only — the dry signal is never pitched. The shifter stays in-line at all times, so wet-path latency is constant across zero.
In the Vault, PITCH is a real offline pitch-shift of the source region, applied before the FX tail is generated — the reverb hears the transposed material.
How much of the wet feed is pitched versus left at original pitch. The shifter keeps running regardless of the setting, so blend moves never click. Real-time — no kernel rebuild.
At 0% the transpose is inaudible by design. If PITCH seems dead, look here first.
Renders the exact live kernel offline onto the active source slot — linear convolution, same build — and auditions the result. This is how you hear Live-mode sound in the standalone app, whose live input is muted, and how you A/B a kernel without rolling the transport.
Switches the Evidence Scope to a 512-point decimated peak stream of the final output across a window of roughly 5.5 seconds. The tap sits after the output filter, so the scope draws exactly what you hear.
The left plate is the reverb — the engine that builds the main swell and the live preverb kernel. FLAVOR and BLEND sit on top; below them, up to ten per-flavor knobs in five columns. These knobs are internal, not host parameters: their names and ranges change with the active flavor, so publishing them to the host would scramble automation. The flavor owns its full sound.
Selects the algorithmic reverb. The flavor box is also the on/off switch: on Off, an OFF badge rings the box so a silent module is never a mystery.
The main reverb layer's share of the swell. An armed flavor with BLEND at zero is inaudible — the panel flags the state with a NO BLEND badge ringing the slider.
Each flavor is relabeled and re-ranged from its own layout and owns its complete character. Decay and PreDly also feed the engine's estimate of tail length, which sizes the render window.
On GHOST SHIMMER the Duck slot becomes SHIMMER (0–1, default 0.35) — the octave bloom. In the live kernel, shimmer is forced off: its octave shift is exactly the metallic artifact. There, TONE scales the kernel's fixed low-pass instead, 3–15 kHz.
Pre-delay into the tank. Inside the live kernel it is scaled hard — 5% to 45% of the dialed value — so the swell still fits the pre-roll window.
Per-flavor output trim on the reverb. Level the flavors against each other here, not with BLEND.
The right plate is the delay. Get the hierarchy straight: the reverb builds the MAIN swell; the delay is a SECOND layer with its own bloom timeline, mixed in on top. FLAVOR, SYNC and DIVISION sit up top, the layer controls BLEND / SWELL / DECAY and the STEREO box below, then up to eight per-flavor knobs in two rows of four. Like the reverb knobs, the per-flavor knobs are internal — not host parameters.
Selects the delay machine. Off bypasses the module and rings the FLAVOR box with an OFF badge.
Locks the delay time to host tempo through DIVISION instead of the flavor's Time knob in milliseconds.
The tempo division used while SYNC is on — quarter-note multiples from 4.0 down to 0.25.
How much reversed delay is mixed into the swell. At zero the layer is skipped entirely — the classic reverb-only swell is bit-identical. Raise it and pre-echoes start riding the rise.
WHEN the delay layer blooms. This is an onset envelope across the rise, not feedback. Low values bring the echoes in early, alongside the reverb. High values hold them back so they arrive late and bloom into the landing. Below 2% the envelope goes flat and the repeats run across the whole window.
Echo sustain inside the pre-swell window. The kernel is a finite offline render, so this control safely pushes feedback toward ~0.995 — impossible in a real-time loop — plus a sustain-lift of up to ~12x on the earliest pre-echoes, so the flutter survives all the way back to the start of the window.
Reshapes the delay kernel's image along its time axis. PING-PONG bounces pre-echoes left/right per repeat — the flip period is the effective delay time. ORBIT rotates the image roughly one turn per bar, tempo-locked. A side high-pass keeps the lows centred; every mode is mono-safe.
Echo time when SYNC is off. In the live kernel, Time is clamped so at least five repeats always fit the pre-roll window; a long user Time only gets clamped when the window is shorter than it.
Repeat count. It also feeds the estimate of the ring-out tail budget, so heavier feedback buys the aftertail more room.
Each flavor applies its own coloration — tape wow, drum motor, PCM grit, sampler crush — and returns its own dry/wet-mixed stereo. Learn the flavors by ear; the knobs mean what the machine says they mean.
Eleven knobs in one row across the bottom: the eight gold Dust macros, a divider, then the output group, plus the ALL/WET filter-mode button and the OPEN DUST VAULT button. LEVEL reads in dB; RINGOUT reads in percent, with 0 shown as Off.
Two kinds of knob live here. TAIL, GHOST and RINGOUT change how the tail is generated — they mark the print stale and, in Live mode, rebuild the kernel. Everything else is a live final-edit and needs no re-render.
The reverse-rise curve of the printed swell. 50% is the natural reverse-reverb rise. Lower is more even; higher is a dramatic late suck into the landing. It re-shapes the cached raw bloom live — no re-render.
Overall delay/reverb trail length and intensity. A tail-generator parameter: in the Vault, press Create Swell after moving it.
Diffusion, modulation, shimmer influence, blur. Haunted atmosphere, not just more reverb. Re-render required.
Dust-Vault age. A tanh saturation stage (drive 1 + 3x) is crossfaded in, plus 2% signal-gated dust noise. The noise runs from a fixed seed, so renders are reproducible — the same swell prints the same twice.
Global openness. The same Harrison-style resonant 24 dB/oct TPT-SVF low-pass engine as the Vault's Final Filter, swept from 650 Hz to 20 kHz in log space. At 97% and above it is an exact no-op — leave it at 100% and your explicit HPF/LPF endpoints win untouched.
The Dust 12.47 mono-safe two-stage widener, applied to the final swell. It scales the existing side signal and adds decorrelated side derived from the mid, with a 150 Hz side high-pass. All the width lives in the Side channel, so a mono fold-down never hollows out.
Output trim on the finished swell. It applies identically to Play Swell, Print, Export and Drag — never to raw source playback — and sits after loudness compensation and WIDTH, before the safety ceiling. This is the knob that tames a hot snare print.
The aftertail: up to 3.0 seconds of the natural forward FX decay, printed after the landing so the swell does not dead-stop. The landing itself never moves. At zero, only a release fade prevents a click.
The global output high-pass — 12 dB/oct TPT SVF, Q 0.707, attenuate-only. It runs in real time on everything you hear and is baked into prints and exports at write time, so what you hear is what prints. Smoothed cutoff, zero latency.
The global output low-pass. Same engine, same hear-equals-print rule.
The global dry/wet blend. In Live mode, wet gain is MIX × WET and dry gain is (1 − MIX) × Dry Amount, gain-ramped per block so moves are zipper-free.
What the output filters touch. ALL filters everything, the dry blend included. WET filters only the effect path — the live wet feed and auditions — leaving dry untouched. The filters are linear, so filtering the wet feed is exactly filtering the swell.
OPEN DUST VAULT lifts the full overlay page: import, capture, edit, print, drag to DAW. Three plates. EVIDENCE BANKS on the left. SOURCE / SWELL in the centre. TAIL / FILTER on the right. Every Vault setting is session and preset state — none of it is a host parameter.
The source bank feeds the reverse-swell engine. The print bank stores rendered swells as WAV files with a .btmeta.json sidecar. Import Audio or drag material in — audio files, and Cubase vst-xml text drags both land. Rename, Discard Evidence, and a right-click bulk menu keep the banks in order.
A DAW-sync state machine gates capture against the host transport and snapshots the input pre-output-stage — the take is clean of everything downstream. Finishing a capture stores the take's BPM and metre; those numbers are what make tempo-fit swells possible later.
The target swell length in bars at the captured tempo. Create Swell kicks a background render — the UI stays responsive, and controls that touch the render are disabled until it lands.
On: pad-don't-stretch. If the natural bloom cannot fill the window, the real reversed material is placed landing-aligned at the end, with silence before it — never a phase-vocoder smear. Off: a deliberate varispeed fit, the tape-style reverse drop.
TAIL TYPE chooses which FX chain generates the forward wet tail that gets reversed into the swell. Every type renders forward and then reverses; only the generator differs. Reverb Swell is the classic producer move. Legacy preset routing names migrate automatically.
The locator marks where the source feed ends and the tail begins. The reverse is built from the wet tail after the marker; everything before it is source-feed region and is skipped — the classic reverse-FX move. Audition Tail plays the forward wet tail so you can vet the FX before it is reversed.
Every audition, print, export and drag runs the same applyEdit chain. What you hear is what prints, exports, and drags.
Applied in stage two of the edit chain. Right-click a fade for the reset menu. The 1 ms floor means even "no fade" never clicks.
A Harrison 32C-inspired resonant high-pass / low-pass pair. Engage MOTION and the cutoffs sweep from start to end in log-frequency space across the swell: RISE ONLY opens toward the landing and holds; RISE+FALL falls back over the tail; FALL ONLY starts open and closes. CHAR adds resonance and analog-style drive.
Editing aids for the trim and tail locators on the source lane, with a bar/beat readout derived from the captured tempo.
Thirty parameters are declared in the automation layout; counted from the host side, the total shows as 31. The IDs are a compatibility contract — never renumbered — so automation lanes written today recall correctly later.
Every GUI knob writes through the same bridge: knob to audio parameter, parameter change to the applied value, into the atomics the DSP reads. A knob move and an automation move take the same road, so they can never diverge. Non-finite automation values — NaN, infinity — are sanitized to defaults before they touch anything.
The per-flavor delay and reverb knobs are deliberately not host parameters. Their meaning changes with the active flavor, and a lane whose meaning changes when you switch flavors is worse than no lane.
The final monitor gain, applied sample-smoothed over everything: live output, auditions, passthrough. There is no knob for it on the panel — reach it from the host's parameter list or automation.
The dry level in Live mode. At 0 the dry is muted and Backtrace becomes a wet-only preverb — useful on a send. Host parameter only.
| PARAMETER | RANGE | DEFAULT | PANEL FACE |
|---|---|---|---|
| output | −24 to +24 dB (step 0.01) | 0 dB | none — host only |
| mix | 0–1 (step 0.001) | 0.25 | MIX, bottom strip |
| live_mode | on/off | off | LIVE PREVERB toggle |
| live_wet | 0–1.5 (step 0.001) | 1.0 | WET, top plate |
| live_dry | 0–1 | 1.0 | none — host only |
| live_pitch_blend | 0–1 | 1.0 | BLEND, top plate |
| live_time | 1/32 to 4 Bars (8 steps) | 1/4 | TIME, top plate |
| live_feel | Straight / Dotted / Triplet | Straight | FEEL box |
| live_shape | 0–1 (step 0.001) | 0.5 | SHAPE, top plate |
| pitch_semis | −24 to +24 st (step 0.01) | 0 | PITCH + oct −/+ |
| tail_type | 6 choices | Reverb Swell | TAIL TYPE box (Vault) |
| delay_flavor | Off + 6 machines | Off | FLAVOR, delay plate |
| delay_sync | on/off | off | SYNC, delay plate |
| delay_division | 9 divisions | 1/4 | DIVISION, delay plate |
| delay_blend | 0–1 | 0 | BLEND, delay plate |
| reverb_flavor | Off + 5 rooms | Off | FLAVOR, reverb plate |
| reverb_blend | 0–1 | 1.0 | BLEND, reverb plate |
| delay_swell_pos | 0–1 | 0.5 | SWELL, delay plate |
| delay_decay | 0–1 | 0.4 | DECAY, delay plate |
| delay_stereo | 4 modes | Stereo | STEREO box |
| pull | 0–1 | 0.5 | PULL macro |
| tail | 0–1 | 0.35 | TAIL macro |
| ghost | 0–1 | 0 | GHOST macro |
| damage | 0–1 | 0 | DAMAGE macro |
| reveal | 0–1 | 1.0 | REVEAL macro |
| width | 0–1 | 0.7 | WIDTH macro |
| ringout | 0–1 (0 = Off) | 0.35 | RINGOUT macro |
| swell_level | 0–1 (reads −24 to +6 dB) | 0.8 (~0 dB) | LEVEL macro |
| out_hpf | 20–2000 Hz | 20 Hz | HPF, output group |
| out_lpf | 200 Hz–20 kHz | 20 kHz | LPF, output group |
The header carries the branding art, the preset browser, and the utility hamburger menu: Open Dust Vault, Audition Forward Tail, Reveal Library Folder.
< and > cycle the bank. Save As writes a user preset file to disk; Del removes user presets; Reset restores the default FX state without clearing Vault audio. A dirty dot tracks edits since load.
Presets are settings-only. They never carry captured audio, so loading one cannot destroy evidence. Live Preverb category presets additively enable Live mode with tuned Time, Feel, Shape and Mix.
Fifty-eight starting points, filed by category. Treat every preset as a conversation opener, not a verdict.
| CATEGORY | PRESET | EVIDENCE |
|---|---|---|
| Utility / Init | INIT - Clean Backtrace | Neutral start (always preset 0): Reverb Swell, Velvet Hall at 30% mix, pitch 0 |
| Live Preverb | Vocal Lift | Ghost Shimmer, 1/4 straight, 40% mix |
| Guitar Swell | Velvet Hall, 1/4, 40% mix | |
| Snare Halo | Phantom Plate (bright tone), 1/8, gentler shape | |
| 808 Sub Bloom | Dark Velvet Hall, 1/2-note pre-roll | |
| Big Riser | Ghost Shimmer, 1 bar, late-bloom shape 0.82, pitch +12 | |
| Tight Transition | Modern Space, 1/8 | |
| Cathedral Rise | Dark Velvet Hall, 2 bars, shape 0.70 | |
| Dotted Pulse | Ghost Shimmer, 1/8 dotted | |
| Micro Shimmer | Ghost Shimmer, 1/16 micro smear | |
| Plate Vocal Air | Airy Phantom Plate, 1/4 | |
| Vocal Throws | Ghost Vocal Rise | Ghost Shimmer reverb swell, pitch +7 |
| Dirty Vocal Throw | Tape Witness into Velvet Hall | |
| Cold Rack Wide Throw | Cold Rack + Modern Space in parallel, wide | |
| Tape Witness Slapback | 140 ms tape slap into hall | |
| Velvet Hall Vocal Bloom | Hall-only delay-into-reverb bloom | |
| Magnetic Noir Vocal | Dark drum echo into dark hall | |
| Reverse Tail Vocal Pull | Plate tail reversed (legacy reverse-tail routing) | |
| Reverse Blooms | Reel Echo Reverse | Multi-head tape echo reversed |
| Velvet Hall Reverse Bloom | Hall decay 0.5 reversed | |
| Reverse Tail Pull | Plate reverse tail | |
| Full Reverse Evidence | Tape Witness + Modern Space full-reverse print | |
| Parallel Dream Throw | Digital Pedal parallel with shimmer | |
| Modern Space Reverse Pull | Clean modern room reversed | |
| Ghost Shimmer Lift | Shimmer 0.5, decay 0.6, pitch +12 | |
| Drum FX | Snare Suckback | Bright plate, short decay, reverse tail |
| Reverse Crash Builder | Long hall decay reversed | |
| Dirty Drum Pull | Vault Delay grit into Modern Space | |
| Crushed Ping-Pong | Vault Delay character 0.65, wide, no reverb | |
| Room Hit Reverse | Modern Space reverse tail | |
| Rust Spring Snare Throw | Spring tank, 8 ms predelay | |
| Guitar Swells | Psychedelic Guitar Pull | Magnetic Drum + shimmer reversed |
| Tape Witness Guitar Swell | Dark tape into hall reversed | |
| Magnetic Drum Ghost Lead | Drum echo into hall | |
| Reel Echo Guitar Bloom | Reel echo into hall | |
| Phantom Plate Guitar Wash | Plate wash, delay-into-reverb | |
| Synth / Cinematic | Cinematic Backtrace | Wide shimmer, decay 0.7, reversed |
| Dark Chamber Pull | Dark hall reverse tail | |
| Deep Space Rise | Wide Modern Space, pitch +7 | |
| Ghost Shimmer Pad | Wide shimmer pad bed | |
| Modern Space Trailer Pull | Decay 0.7, width 0.9, reversed | |
| Wide Ghost Hall | Width 0.9 hall | |
| Dirty Vault | Vault Delay Broken Clock | Sampler delay, character + movement 0.6 |
| Dust Loop Transition | Vault Delay into Modern Space | |
| 8-Bit Evidence | Character 0.85 crush, delay only | |
| Haunted Buffer | Crushed delay + shimmer reversed | |
| Dirty Reverse Artifact | Crushed + movement, reversed, no reverb | |
| Broken Converter Throw | Character 0.75 into hall | |
| Clean Studio | Cold Rack Clean Throw | PCM delay into Modern Space |
| Modern Space Vocal Depth | 45 ms predelay, ducked | |
| Velvet Hall Studio Throw | Subtle reel echo into hall | |
| Phantom Plate Vocal Sheen | Bright ducked plate | |
| Parallel Clean Bloom | Digital Pedal parallel with Modern Space | |
| Experimental | Feedback Evidence Loop | Reel echo feedback 0.45, verb-to-delay feedback routing |
| Full Reverse Mutation | Crush + shimmer 0.5 full-reverse, pitch +12 | |
| Redacted Memory | Crushed delay + dark hall reversed | |
| Pitch Drop Witness | Tape into hall, pitch −12 | |
| Wide Reverse Glitch | Wide crushed delay + wide Modern Space reversed |
Operational procedures developed during the build. They are not the only ways to run the unit. They are evidence of what it will do.
TECHNIQUE 01 — ENGAGE BEFORE THE TAPE ROLLS
Engage LIVE PREVERB while the transport is stopped, then play. The swell only lands in front of the sound because the host compensates the reported latency, and hosts apply that compensation on playback start. This is a mix and transition effect by design — not a zero-latency tracking effect.
TECHNIQUE 02 — THE SHAPE GATE
Automate SHAPE past its validated middle. Above 0.8, an onset gate silences up to the first 25% of the pre-roll and then surges — hold, then bloom, ideal before a drop. Below 0.2, an early floor makes the swell drag in audibly from the very start. The 0.2–0.8 span, including the confirmed-good 0.5, is untouched by either extreme.
TECHNIQUE 03 — PRE-ECHOES THAT BOUNCE INTO THE HIT
Delay BLEND defaults to 0, so the classic reverb-only swell stays bit-identical until you raise it. For reverse-delay pre-echoes that bounce into the hit: raise Delay BLEND, push SWELL high so the echoes bloom late into the landing, and pick PING-PONG — the flip period is locked to the effective delay time, so one flip lands per repeat. Want rotation instead? ORBIT turns roughly once per bar, tempo-synced.
TECHNIQUE 04 — IMPOSSIBLE FEEDBACK
Delay DECAY is a superpower of the offline kernel. Because the pre-swell is a finite render, feedback can safely approach ~0.995 — impossible in a real-time loop — and a sustain-lift keeps the earliest pre-echoes audible. Set DECAY to 100% for a long flutter that fills the entire window.
TECHNIQUE 05 — THE TAPE DROP
Keep Pitch OFF is a deliberate effect, not a fallback. The reversed tail is varispeed-resampled to the window — the tape-style reverse drop. Keep Pitch ON never stretches: a short phrase in a long window gets silence first, then the true swell landing exactly on the grid.
TECHNIQUE 06 — BALANCE IS NOT LOUDNESS
The blend knobs shape balance, never level. After the delay and reverb layers are mixed, the render is loudness-compensated to a fixed 0.60 peak before WIDTH. To tame a hot source — a snare, say — use LEVEL (−24 to +6 dB, default ~0 dB). It applies identically to Play, Print, Export and Drag.
TECHNIQUE 07 — THE REVEAL SWEEP
REVEAL is the same Harrison-style resonant 24 dB/oct engine as the Vault's Final Filter, swept 650 Hz to 20 kHz. It is an exact no-op at 100%. Automate it for dark-to-open reveals without disturbing your explicit HPF/LPF endpoints or a Motion sweep already printed into the swell.
TECHNIQUE 08 — LET IT RING
RINGOUT prints up to 3 seconds of the reverb or delay's real forward decay after the landing — the render window is sized to include it, so no tail is ever hard-cut at the swell end. At 0 only a release fade prevents a click. The landing position never moves; only the exhale after it.
Seven cold cases, solved. If the unit seems wrong, it is probably doing exactly what it was told.
CASE 01 — THE MODULE THAT CHANGES NOTHING
SYMPTOM: The delay (or reverb) module is set up but changes nothing.
FINDING: Check the badge next to the module title. OFF means the FLAVOR box is on Off — flavor is the on/off switch, and the box gets ringed in gold. NO BLEND means the engine is armed but its Blend is 0, so it cannot reach the swell; the Blend slider is ringed instead. And the delay layer only exists when a reverb builds the main swell — with the reverb Off, the delay already is the tail, and Delay BLEND does nothing.
CASE 02 — THE PRINT THAT WILL NOT UPDATE
SYMPTOM: You turned TAIL, GHOST, RINGOUT, TAIL TYPE or PITCH and the printed swell did not change.
FINDING: Those are tail-generator parameters. They mark the print stale but never rebuild it on their own — so Audition, Print and Export always use the same cached render. The UI shows an out-of-date warning; press Create Swell. PULL, DAMAGE, REVEAL, WIDTH, LEVEL, the fades and the Final Filter are live and need no re-render.
CASE 03 — THE SWELL THAT ARRIVES LATE
SYMPTOM: In Live mode the swell lands late, or on top of the sound instead of before it.
FINDING: The preverb depends on host latency compensation: Backtrace delays the dry by the pre-roll and reports it as latency. Some hosts do not compensate live-monitored input; on playback the swell should be pulled early automatically. Latency is latched to the loaded kernel — after a tempo change it rebuilds once, only after the BPM has drifted more than 2% and stayed stable for about 500 ms. That is deliberate: it avoids host PDC re-sync storms. TIME is also capped at 5 seconds of pre-roll regardless of the chosen note value at slow tempos.
CASE 04 — THE SILENT STANDALONE
SYMPTOM: LIVE PREVERB makes no sound in the standalone app.
FINDING: The standalone's live input is muted by design. Use PREVIEW LIVE — it renders the exact same kernel offline onto the active source slot and auditions it, so you can judge the live sound without a DAW.
CASE 05 — THE PANEL THAT WENT QUIET
SYMPTOM: Knob moves during a swell render seem ignored; parts of the panel briefly disable.
FINDING: Create Swell runs on a background worker. Controls that touch the render are disabled while it is busy, and queued renders coalesce — the latest parameters win. In Live mode, every tail-affecting knob move rebuilds the kernel with a 100 ms debounce and an ~85 ms crossfade into the new sound. A short settle before you hear the change is normal, not a stuck control.
CASE 06 — THE PITCH THAT IS NOT THERE
SYMPTOM: PITCH does nothing in Live mode.
FINDING: Check BLEND. At 0 the wet feed stays unpitched and the transpose is inaudible by design. The dry signal is never pitched in Live mode — only the swell feed is.
CASE 07 — THE PLUGIN THAT NEVER SLEEPS
SYMPTOM: The host shows the plugin always active and never suspends it.
FINDING: Intentional. Backtrace reports an infinite tail because its RAM auditions — Play Swell, Preview Live — sound without any input. A finite tail let hosts park the audio processing on silent, stopped tracks and kill playback mid-audition. The unit stays awake so the evidence keeps playing.
| PARAMETER | SPECIFICATION |
|---|---|
| Formats | VST3, AU (AudioUnit effect), Standalone. AAX and CLAP noted as future additions — not built. |
| Host coverage | VST3: Live, Cubase, Nuendo, Studio One, REAPER, FL Studio, Bitwig, Waveform. AU: Logic, GarageBand, MainStage. |
| Platform | macOS (AU + Developer ID signing). Only macOS targets are built. |
| Channels | Mono in to mono out, or stereo in to stereo out — input and output layouts must match. |
| Latency | 0 samples in Capture/Print mode. Live Preverb: 2048 (convolution head) + kernel length − 1 + pitch-shifter latency, reported deterministically; kernel clamped 0.05–5.0 s. Latency latched to the loaded kernel; tempo drift over 2%, stable ~500 ms, triggers one rebuild. |
| GUI Size | 1320 × 752 fixed design canvas, scaled; resizable 660 × 376 to 2640 × 1504 at fixed aspect. |
| Presets | 58 factory in 10 categories, plus user presets on disk. |
| Host parameters | 30 declared ("output" + 29 bridged); 31 as counted from the host side. IDs frozen. |
| Tail reporting | Infinite — the host can never suspend the plugin on input silence and kill a RAM audition. |
| Capture | Up to 60 s per take; 8 source + 8 print slots; modes Manual / DAW Locators / Next Cycle with 1/2/4/8-bar fallback. |
| Identifiers | Manufacturer SDet · plugin code Bktr · bundle com.sounddetective.backtrace · VST3 category Fx|Delay|Reverb · product "Detective 47s Backtrace". |
| Safety | NaN/inf armour on every numeric entry; DC block and transparent-below-unity soft ceiling on every offline path; verified by the SafetyTest, RenderTest and TortureTest console targets. |
| Manufacturer | Sound Detective / Detective 47 |
| Version | 1.0.0 — single-sourced from the build system, shown on-panel as DETECTIVE 47 // BACKTRACE v1.0.0. |
Evidence reversed. Case closed before it opened. The swell is yours now.