Changelog

For full release notes, binaries, and platform-specific archives, see the GitHub Releases page.

v2.0.0 — Unreleased

EQ filter topology: TPT state-variable filters (breaking sound change)

Every EQ stage in the plugin now runs on a trapezoidal-integrated (TPT / zero-delay-feedback) state-variable filter core instead of the RBJ-cookbook direct-form biquads used in v1.0. This covers API5500 (all five bands and the HPF/LPF), Pultec (all five sections), Dynamic EQ (all four bands plus the sidechain and solo filters), and Sheen (BODY, PRESENCE, AIR and both WIDTH filters). Tracked in #15; rationale in docs/V2_ROADMAP.md §3.2.

What changes for existing sessions. Parameter IDs are unchanged, so v1.0 sessions load and play without any relinking or missing-parameter warnings. What you hear is slightly different:

Static curves are the same. The TPT core is the bilinear transform of the same analog prototype the biquads used, so at fixed settings the new filters null against the old ones to floating-point precision at every frequency, gain and Q the modules expose. A session that was not automating EQ parameters will sound identical.
Moving parameters behave differently, for the better. Direct-form biquad state is a set of delayed output samples that only make sense for the old coefficients, so sweeping a frequency or Q (or the Dynamic EQ’s gain-reduction path recomputing coefficients every couple of samples) produced small transient bursts. The SVF state variables are the integrators’ outputs, so a coefficient change lands on the new response immediately. Automated EQ moves, Dynamic EQ under heavy gain reduction, and high-Q settings are the places you may hear a cleaner result.
No biquad fallback. The migration is hard: there is no “v1.0 filter” toggle. This is the same compatibility stance taken for Sheen in v1.0.

Pultec: Linear Phase mode

A new Linear Phase switch on the Pultec (parameter pultec_linear_phase, off by default). When engaged, the five EQ sections are replaced by a single zero-phase FIR whose magnitude is designed from the very same filter coefficients the minimum-phase chain uses, so the two modes measure the same and differ only in phase. The tube stage still runs after the FIR.

Latency: 512 samples at any sample rate (10.7 ms at 48 kHz), reported to the host for delay compensation. Latency is reported only while the switch is on and the Pultec is in the module chain; the module’s bypass and the global bypass both preserve the delay so toggling never shifts the track against its neighbours.
Low-frequency resolution. The 513-tap kernel resolves roughly fs/513 (≈ 94 Hz at 48 kHz). HF sections reproduce within a fraction of a dB; the LF boost shelf and its resonant bump are delivered but with a softer corner than minimum-phase mode below ~200 Hz. A longer-kernel “high resolution” option is a planned follow-up.
Defaults off, so v1.0 sessions keep their zero-latency behaviour on load.

Transformer + Sheen WARMTH: hysteresis-modeled saturation (bigger, noticeably-warmer sound change)

A single-cell play-operator hysteresis model — a Preisach-lite approximation of the memory effect in real magnetic tape and transformer cores — is now part of Transformer’s saturation core, and available as an optional “tape” sub-mode on Sheen’s WARMTH stage. Unlike the plugin’s previous memoryless waveshapers, output now depends on recent input history, not just the present sample. Tracked in #16; rationale in docs/adr/0012-single-cell-hysteresis-saturation.md.

Transformer: on by default. This is a bigger sound change than the TPT filter migration above — sessions with Transformer enabled will sound noticeably warmer than in v1.0. A new transformer_hysteresis_bypass switch (default off, meaning hysteresis is ON) restores bit-identical v1.0 saturation for anyone who needs it.
Sheen WARMTH: opt-in. A new “tape” sub-mode toggle (default off) routes WARMTH through the same hysteresis cell before its existing Inflator-style polynomial. WARMTH’s factory-default sound is unchanged unless you turn it on.

v1.0.0 — 2026-04

Two major workstreams shipped together: Sheen (a hidden master-end polish coat) and a full multi-fx rack UX redesign that finally makes module reordering feel native.

Sheen — master-end polish coat

A new pinned DSP stage at the end of the chain (post-Punch, pre-master-gain) that makes the chassis sound finished out of the box. The brushed-brass API brand plate in the chassis header is the only front-panel surface — click it to flip into a hidden back view with five sliders.

Always-on by default at research-grounded factory tuning:

BODY — low shelf @ 100 Hz, default +1.0 dB
PRESENCE — peak EQ @ 3 kHz, Q=1.0, default 0.0 dB (transparent at default)
AIR — high shelf @ 14 kHz, Q=0.5, default +1.8 dB
WARMTH — Sonnox Inflator polynomial @ Curve=0, 2× oversampled, default 20% mix
WIDTH — M/S side-only HPF @ 150 Hz + shelf @ 500 Hz, default 50%

Defaults are anchored in three parallel research reports — classic console-bus measurements (SSL G, Neve 33609, API 2500, Studer, Trident), polish-plugin teardowns (Slate Revival, Kush AR-1, Maag EQ4, Sonnox Inflator, Pensado, Vitamin, Ozone, PSP Vintage Warmer, bx_console), and tape + transformer harmonic profiles (Studer A800, Ampex ATR-102, Jensen, Lundahl, Carnhill). See docs/adr/0006-sheen-pinned-master-end-polish.md for citations and stage-by-stage rationale.

Sheen is excluded from global_auto_gain — auto-comp on a polish stage defeats its purpose. See the Sheen module page for the full reference.

Multi-fx rack redesign

The seven-slot rack got a ground-up UX overhaul:

Native vizia drag-drop replaces the previous hand-rolled mouse-capture state machine that was silently failing under baseview’s Win32 SetCapture lifecycle.
Swap-or-insert semantics decided by cursor X within the target slot at release time: left third = insert before, middle = swap, right third = insert after.
Live drop-position preview while dragging — bright cyan bar pinned to the target slot’s left or right edge (insert) or full yellow ring around the slot (swap), so you see the resolved drop intent before releasing.
Floating drag-ghost label tracks the cursor showing the dragged module’s tag.
Empty slots collapse by default to narrow dashed tabs; the library sidebar is now the sole “add module” affordance.
Focus mode — press 1..7 to focus a real-module slot (collapses every other slot to a tab); Esc exits focus, cancels drag, or closes any open back view.
Chain mini-map appears as a band above the chassis only when a slot is focused, hidden otherwise.
Brushed-brass brand plate is the entry point to the Sheen back view; mutually exclusive with the DynEQ back view.
Click-lag fix — the chassis-level MouseUp listener that previously routed every click through the broken drag state machine has been removed.

Compatibility

11 new automation params (Sheen) on top of the existing ~75 — no reused IDs. Existing DAW sessions load with Sheen ON at factory defaults; playback will subtly differ from pre-1.0 sessions. For bit-identical playback, flip the back-panel master Sheen bypass.

Other improvements

justfile FEATURES and CORE_FEATURES lists corrected (sheen and haas were missing from the bundle commands)
vizia event-routing documentation: clickable HStacks need on_press on each child Label because vizia’s on_press is leaf-targeted, not bubbling
Module-level cached params now use .value() instead of .smoothed.next() — the latter advances 1 sample per call

Test coverage

8 new Sheen unit tests
159 lib tests passing
Verified in Reaper across drum / bass / vocal / full-mix material; rack drag-drop verified with swap, insert-before, insert-after, drag-to-empty-slot, and drag-cancel-by-leaving-window

v0.5.0 — 2026-04

Pultec EQ — authentic LCR resonance and bandwidth control

The LF section is completely remodeled to match the sound of the original EQP-1A hardware.

LCR resonant bump: A PeakingEQ at the shelf corner frequency (45% of shelf gain, Q=1.8) models the inductor resonance of the original hardware. This resonant peak is why Pultec LF boosts sound focused and punchy rather than soft and woolly.
18 dB range for both LF Boost and LF Cut (extended from ~8 dB).
LF Boost Freq: now spans 20–300 Hz with skewed scaling (was stepped 20/30/60/100 Hz).
LF Cut Freq: now spans 20–400 Hz with skewed scaling (was 20–200 Hz).
New: LF Boost Bandwidth — controls shelf width from narrow (Q=1.0) to wide (Q=0.25). At the default of 0.67, the shelf spreads through the musical 100–300 Hz range. This is the difference between a Pultec that sounds sub-only and one that sounds full and present on guitars, keys, and mix buses.
New: LF Cut Bandwidth — same Q mapping as LF Boost BW.
All filter math now routes through shaping::biquad_coeffs to apply the correct Nyquist normalization (f0 * 2 / sr), fixing the biquad 0.5.0 frequency normalization bug that was placing every filter at 1/4 its intended corner frequency.

New module: Haas stereo widener

A psychoacoustic stereo widener using M/S encoding and Haas effect comb filtering.

Two modes: Side Comb (WOW-Thing style, mono-compatible) and Wide Comb (diffuse, L-R delay injection).
Delay time: 1–20 ms, Hermite-interpolated with one-pole smoothing (τ=20 ms) to prevent zipper noise on automation sweeps.
RMS-safe output trim: automatic gain compensation based on mid/side gains and comb depth.
Anti-denormal protection: Airwindows-style 1e-20 alternating dither in the delay line.
Default position: before Punch — so the clipper catches any widener-induced peaks before they hit the ceiling.
Feature flag: haas (enabled by default).

Plugin integration tests

New src/plugin_integration_tests.rs file exercises the full plugin pipeline rather than isolated module tests. Catches parameter wiring failures, bypass default regressions, and coefficient initialization bugs. Currently covers Pultec module end-to-end.

CI fix

src/haas.rs was missing from the repository (untracked locally). Added to git. The haas feature is in [features] default, so all CI targets were failing with E0583: file not found for module 'haas'.

v0.4.0 — 2026-03

ButterComp2 — new compression models

Three new compressor models join the original Airwindows Classic algorithm. Select the model from the dropdown at the top of the ButterComp2 module panel; the controls below update to match.

VCA — Hard-knee voltage-controlled amplifier compression. Fast, precise transient response. Threshold (dB), Ratio (1–20), Attack (ms), Release (ms), Character % (0–100, controls 1176-style color), Mix. Use on drum buses and any source where you want predictable, controllable gain reduction.
Optical (Opt) — Soft-knee program-dependent compression. Threshold (dB), Character % (0–100, controls program-dependent release behavior and tube warmth), Attack (ms), Release (ms), Mix. Use on vocal, acoustic instrument, and bass stems.
FET — Field-effect transistor compression driven by an input gain stage rather than a threshold control. Input (dB), Output (dB), Ratio (1–20), Attack (ms), Release (ms), Auto Release (toggle — enables program-dependent release), Mix. Use on drum buses and any stem where you want a forward, saturated character.
Classic — Airwindows ButterComp2 (original, unchanged). Bipolar interleaved compression with Compress, Output, and Dry/Wet. VCA, Optical, and FET are native Rust implementations; Classic compiles from Airwindows C++ via Rust FFI.

Parameter display — integer formatting

All non-percentage float parameters now display as integers throughout the UI, automation lanes, and parameter tooltips. This eliminates long decimal strings on thresholds, ratios, EQ gains, and timing values. Existing automation data is unaffected — the underlying parameter ranges are unchanged.

UI — API5500 EQ layout

LF and HF shelves moved to a two-column layout, freeing vertical space in the module slot
Parametric bands now ordered low-to-high: LMF, MF, HMF

UI — Pultec EQ

HF boost bandwidth control now visible in the panel
HF cut frequency and gain controls now visible

UI — Transformer

Output saturation control exposed
Panel sections structured as INPUT / OUTPUT / TONE for clarity

UI — ButterComp2 model panel

Model selector dropdown at top of module
Per-model control panel updates via Binding::new pattern — controls swap without resizing the module slot
Fixed-height panel prevents adjacent modules from shifting when switching models

DSP — global bypass

Global bypass now passes audio through with zero processing overhead via early return from process()
Previously, bypass still evaluated the module order dispatch loop

DSP — global auto gain

New global Auto Gain compensation: RMS level measured pre- and post-processing chain
~5-second smoothing time constant; ±18 dB correction range
Compensates for level changes introduced by heavy compression or EQ boost without manual output trimming

Fix — NaN sentinel values

Replaced NaN-based dirty-check sentinels in the compressor model switching logic
Previously, switching models could leave NaN values in internal state, causing audio silence until the plugin was reset

v0.3.0 — 2026-03

Documentation site

Migrated documentation from MkDocs to Astro 5 + Starlight 0.32 for improved navigation, search, and theming
Added Architecture, Contributing, Changelog, and Parameter Reference pages
Live site deployed to GitHub Pages at fsecada01.github.io/bus_channel_strip

CI/CD pipeline fixes

Replaced deprecated actions/create-release@v1 with gh release create and GITHUB_TOKEN permissions
Fixed Skia cross-compile failure on macOS ARM64 runners: Xcode’s clang on aarch64 blocks x86 SIMD intrinsic headers (mmintrin.h, emmintrin.h) even when targeting x86_64-apple-darwin. Fix: ARM64 builds include GUI, Intel builds compile core modules only without Skia
Removed gui from [features] default in Cargo.toml to prevent Skia from compiling on all CI targets regardless of --features flags passed to xtask bundle
Used actions/download-artifact@v4 with merge-multiple: true for flat artifact glob upload to releases

Release artifacts (4 platform targets)

windows-x86_64 — VST3 + CLAP with GUI (Skia built from source with LLVM 19)
linux-x86_64 — VST3 + CLAP with GUI
macos-aarch64 (Apple Silicon) — VST3 + CLAP with GUI
macos-x86_64 (Intel) — VST3 + CLAP, core modules only, no GUI (Skia cross-compile limitation, tracked as Issue #1)

v0.2.x

Punch module

New Punch module: clipper + transient shaper with 8x oversampling
Three clip modes: Hard, Soft (tanh), Cubic (polynomial soft knee)
Transient detection with configurable attack/release times and sensitivity
Pumping fix: moved transient detection + shaping to pre-clip stage. Previously, post-clip transient shaping created time-varying gain modulation that audibly pumped on every note attack
Downsample IIR pole reduced from 0.3 → 0.05, further eliminating pumping artifacts
Transient detector now uses native sample rate (not the oversampled rate)
Punch module is bypassed by default — user must enable it intentionally

Module reordering UI

Drag-to-swap module ordering implemented in the vizia GUI
Click the drag handle (≡) to select a slot as swap source
Click another slot’s handle to swap positions
Visual feedback: white border + yellow module name on selected slot
Handle label changes reactively: “MOVE” (self) / “SWAP HERE” (others) / “CANCEL”

vizia-plug GUI integration

Replaced placeholder GUI with full vizia-plug integration (September 2025)
Fixed dependency configuration: removed conflicting skia-safe direct dependency
Switched to nightly Rust toolchain (required by vizia-plug)
Skia builds from source on Windows x86_64 using LLVM 19 + MSVC STL
Window: 1820×820 px, six module slots with per-module color coding
Spectrum analyzer canvas with real-time FFT display, band tint overlays, and GR bars

Dynamic EQ — sidechain masking analysis

Optional stereo sidechain input (second audio I/O layout in CLAP/VST3)
Lock-free one-shot analysis pipeline: GUI arms analysis, audio thread runs FFT on sidechain snapshot, results delivered via AtomicBool + AnalysisResult
Spectral overlap detection: overlap[k] = main_fft_mag[k] * sc_fft_mag[k]
Suggests threshold and target frequency for masking band
Two-step UX: “ANALYZE SC” arms; “APPLY RESULT” sets DynEQ parameters

v0.1.x

Initial implementation

Core plugin skeleton with NIH-Plug framework
API5500 EQ: 5-band semi-parametric (LF shelf, LMF/MF/HMF parametric, HF shelf)
ButterComp2: Airwindows C++ module wrapped via extern "C" FFI, compiled with cc crate
Pultec EQ: EQP-1A style with simultaneous boost/cut and tube saturation (tanh shaping)
Transformer: 4 vintage transformer models (Vintage, Iron, Modern, Warm) with input/output saturation
Dynamic EQ: 4-band frequency-dependent compression with configurable mode (compress downward / expand upward)
Module reordering system: six EnumParam<ModuleType> parameters for runtime signal chain ordering
Lock-free audio thread: no allocations, no locks, no panics in process()
Migrated from biquad v0.4 to v0.5.0 API: filter constructors now require gain parameter (Type::PeakingEQ(gain_db), Type::LowShelf(gain_db), Type::HighShelf(gain_db)); .set_gain() removed
Sample-accurate automation enabled (SAMPLE_ACCURATE_AUTOMATION: bool = true)