v0.5.0 - Ballistics-accurate metering

What's new

Input and output meters now show real ballistics instead of a raw peak read:

• Peak — instant attack, ~20dB/s exponential release
• VU — one-pole RMS integrator, ~300ms to settle on a step (classic VU-style ballistics)
• True peak — 4x-oversampled estimate (Catmull-Rom interpolation) catching inter-sample peaks a plain sample read misses
• Clip indicator — held ~1.5s, triggered at a small headroom margin below 0dBFS

All of it runs on a read-only side path off a copy of the buffer — it can't add latency to the actual signal chain, keeping the plugin's zero-added-latency guarantee intact.

Bugs found and fixed during verification

Both caught by the project's real-audio test harness, not by pluginval or GUI-only checks:

• Clip threshold never triggered. An exact >= 1.0f check missed genuinely full-scale samples because normalized-parameter round-tripping through the "0dB" gain stage doesn't always land on bit-exact 1.0f. Fixed with a -0.1dBFS headroom threshold — also just how real clip indicators are normally built.
• Clip hold timing was wrong. The hold counter was tracked in "blocks remaining," decremented once per process() call, but the hold duration was computed assuming a specific block size. Different block sizes on later calls threw off the real elapsed hold time. Fixed by tracking samples remaining instead.

Verified

• pluginval strictness level 5: clean on both VST3 and AU
• Real-audio test harness: peak attack/release timing, VU ballistics, true-peak inter-sample detection, and clip-hold duration all checked against known test signals
• Hosted successfully in REAPER

Known limitations

• Dynamic EQ: no external sidechain input yet (internal detection only)
• Not yet tested against real (non-silent) audio hardware in a live signal chain
• No linear-phase mode — intentionally out of scope (zero latency is the point)

Full Changelog: v0.4.0...v0.5.0