feat: v0.2.0 deep-dive voicing pass, M2 presets, DE localisation, app icon

[yves-vogl]

Summary

Research-driven rework of Overture's Drive -> Clipper -> Tone chain (docs/design-brief.md, docs/research-notes.md), the M2 preset system, a German i18n frame, and housekeeping (app icon wiring, release workflow fix).

DSP rework

• biteAmount ("Bite", 0-100%, default 65%): frequency-dependent gain inside the drive-to-clipper path via a fixed ~700 Hz low-shelf (anchored to the sourced ~720 Hz reference-circuit feedback corner), applied to the oversampled signal immediately before the voicing dispatch. Bass is clipped less than treble - the reference circuit's actual tightening mechanism, not an approximation via a separate pre-clip filter (Tight keeps doing that unchanged job). Skipped entirely (not unity-gain) at 0% - bit-identical to v0.1's flat-gain clipper.
• kneeSoften ("Knee Soften", 0-100%, default 40%): drive-dependent knee softening (src/dsp/KneeSoftening.h), blending each voicing's raw output toward tanh(raw), scaled by (kneeSoften/100) * driveIntensity. Applies to all three voicings including Hard Clip (zero knee at 0%, any Drive). Algebraically always 0 at kneeSoften = 0, so Drive-invariant, matching v0.1.
• asymmetryAmount ("Asymmetry", 0-100%, default 40%): exposes the Asymmetric voicing's previously-fixed bias (a = 0.2) as a 0.0-0.5 mapped control. Default reproduces v0.1's fixed bias exactly.
• biteTilt ("Bite Tilt", -100%..+100%, default 0%) replaces tone: a single 2nd-order RBJ high-shelf around a fixed ~3 kHz corner, bidirectional (v0.1's Tone could only ever darken). Flat is a true no-op (filter skipped, not unity-gain). A v0.1 session's tone value is lossily migrated to an equivalent biteTilt position on load (OvertureAudioProcessor::setStateInformation()), detected before apvts.replaceState() runs.
• Defaults: tight 130 -> 100 Hz (midpoint of the sourced 80-120 Hz workflow sweet spot), drive 8 -> 3 dB (best-documented "near-zero drive, Level does the pushing" workflow). Voicing enum indices remain frozen.
• Every new IIR filter (Bite shelf, Bite Tilt shelf) uses the same allocation-free ArrayCoefficients + ovtr::applyBiquadCoefficients pattern main already established for Tight/Tone (issue Unconditional per-block heap allocation in OvertureEngine::process() via IIR::Coefficients::make{High,Low}Pass #12) - tests/AllocationTests.cpp extended accordingly.

M2 preset system

Copied verbatim from the Nave pilot implementation (src/presets/PresetManager.{h,cpp}, src/presets/PresetBar.{h,cpp}) per .scaffold/specs/preset-system-m2.md. Nine factory presets (docs/presets.md): a certified Default (= Classic Boost's settings, the shipped defaults) plus the brief's eight named presets (Clean Push, Classic Boost, Drop-Tune Tight, Smooth Push, Own Distortion, Fuzz-Adjacent Lead, Parallel Grit, De-Fizz Cleanup).

i18n

German localisation of the preset-bar frame (resources/i18n/de.txt, src/presets/Localisation.{h,cpp}), auto-selected via SystemStats::getUserLanguage(). Core/DSP parameter names are never translated (verified by tests/LocalisationTests.cpp).

Housekeeping

• ICON_BIG "docs/assets/icon.png" added to juce_add_plugin() - previously missing (no patch release had shipped since the icon was added to docs/README).
• .github/workflows/release.yml: added an idempotent create-release job (both platform jobs now needs: create-release) - suite-wide fix ported from basilica-audio/crypta main, folded into this PR per the orchestrator's flow addition (nothing else had touched this file).
• docs/architecture.md, docs/manual.md, README.md, CHANGELOG.md, CLAUDE.md updated for v0.2.0.

Scope note on the "0% = v0.1-identical" guarantee

The brief's guarantee 1 ("bite_amount = 0 and knee_soften = 0 reproduce v0.1's exact transfer function bit-for-bit... for all three voicings, at every Drive level") is tested at the clipper-transfer-function level (tests/ClipperVoicingTests.cpp's new backward-compatibility test, comparing the composed dispatch against the unmodified ClipperVoicings::processSample primitive bit-for-bit) rather than as a whole-engine snapshot diff against v0.1's actual binary output. Tight/Drive/Oversample/Level/Mix are structurally unchanged code (only Tight/Drive's default values moved), so a whole-engine null test on those specifically would be redundant; Bite/Bite Tilt's own "skip when off" behaviour is separately verified by their own dedicated spectral/no-op tests.

Test plan

• Full local suite green: cmake --build build --target Tests Overture_Standalone && ctest --test-dir build --output-on-failure - 96/96 test cases, 21278/21278 assertions passed (up from 54 on main).
• No audio-thread allocations while any v0.2.0 control (Bite/Knee Soften/Asymmetry/Bite Tilt) is being automated (tests/AllocationTests.cpp, extended from issue Unconditional per-block heap allocation in OvertureEngine::process() via IIR::Coefficients::make{High,Low}Pass #12's original Tight/Tone guard).
• Backward-compatible null cases: bite_amount = 0/knee_soften = 0 reproduce v0.1's clipper dispatch bit-for-bit, all three voicings (tests/ClipperVoicingTests.cpp).
• Frequency-dependent gain proof (Bite): low/high-frequency clip-onset peak gap grows monotonically with biteAmount at a fixed Drive (tests/EngineTests.cpp).
• Drive-dependent knee-softening proof: waveform-shape change from softening is measurably larger at high Drive than low Drive at a fixed kneeSoften; zero at kneeSoften = 0 regardless of Drive (tests/EngineTests.cpp).
• Asymmetry sweep: 0-100% maps monotonically to bias 0.0-0.5; default reproduces v0.1's a = 0.2 (tests/ClipperVoicingTests.cpp).
• Bite Tilt bidirectionality: flat is a true no-op, negative/positive monotonically darken/brighten, -100% is at least as dark as v0.1's fully-closed Tone at a matched test frequency (measured directly against v0.1's own retired formula, not assumed) (tests/EngineTests.cpp).
• State migration: v0.1-only session loads without crashing, tone maps to the documented biteTilt equivalent, new parameters fall back to their own defaults (tests/StateTests.cpp).
• Preset system: save/load round-trip, forward/backward-tolerant import, wrong-plugin/format refusal, every factory preset parses/loads/produces no NaN/Inf/silence, default resolution order, dirty-flag lifecycle, prev/next wrap-around, rename/delete guards, single-file and bank export/import (tests/PresetManagerTests.cpp, 16 cases).
• i18n: de.txt parses, every TRANS() key used by the preset frame is present, parameter names are verifiably absent from the mapping (tests/LocalisationTests.cpp).
• NaN/Inf robustness: all four new controls swept to extremes combined with extreme Drive/Tight/Level, every voicing (tests/RobustnessTests.cpp).

🤖 Generated with Claude Code