fix: timeline waveforms of near-silent clips no longer drawn full scale

Waveforms are normalised to the source's peak so quiet recordings stay readable, but the
boost was unbounded: an empty audio stream at -52 dBFS (silent archive film) was drawn as
loud as dialogue. The normalisation is now capped at +12 dB.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

crates/ui-egui/src/panels/timeline.rs

@@ -509,8 +509,8 @@ fn draw_waveform(app: &mut FilmcraftApp, p: &egui::Painter, body: Rect, it: &Tra
     if clip.width() <= 0.0 {
         return;
     }
-    let peak = peaks.iter().fold(0f32, |m, (a, b)| m.max(a.abs()).max(b.abs())).max(1e-4);
-    let gain = filmcraft_render::audio::db_to_gain(it.gain_db) / peak;
+    let peak = peaks.iter().fold(0f32, |m, (a, b)| m.max(a.abs()).max(b.abs()));
+    let gain = waveform_display_gain(peak, it.gain_db);
     let mut mesh = egui::Mesh::default();
     let dur_px = body.width().max(1.0);
     for x in (clip.min.x.floor() as i32)..(clip.max.x.ceil() as i32) {
@@ -550,6 +550,15 @@ fn draw_waveform(app: &mut FilmcraftApp, p: &egui::Painter, body: Rect, it: &Tra
     }
 }
 
+/// Display gain for a clip's waveform: clip gain, plus normalisation of the source to full scale
+/// so quiet recordings stay readable, but by at most +12 dB, so a near-silent track (room tone at
+/// −50 dBFS, a silent film's empty audio stream) does not look like a loud one.
+fn waveform_display_gain(source_peak: f32, clip_gain_db: f64) -> f32 {
+    const MAX_BOOST: f32 = 3.981_072; // +12 dB
+    let boost = (1.0 / source_peak.max(1e-9)).min(MAX_BOOST);
+    filmcraft_render::audio::db_to_gain(clip_gain_db) * boost
+}
+
 fn request_peaks(app: &mut FilmcraftApp, item: ItemId) -> Option<Arc<Vec<(f32, f32)>>> {
     if let Some(p) = app.tl.peaks.lock().unwrap_or_else(|e| e.into_inner()).get(&item) {
         return Some(p.clone());
@@ -1611,3 +1620,26 @@ fn interact(app: &mut FilmcraftApp, ui: &mut egui::Ui, seq: &Sequence, layout: &
     }
     app.auto.add("timeline.tracks", layout.content, "tracks");
 }
+
+#[cfg(test)]
+mod waveform_tests {
+    use super::waveform_display_gain;
+
+    fn db(g: f32) -> f32 {
+        20.0 * g.log10()
+    }
+
+    #[test]
+    fn normalisation_is_capped_for_near_silent_sources() {
+        // a −14 dBFS recording is normalised to full scale (+12 dB cap reached: −2 dBFS)
+        assert!((db(0.2 * waveform_display_gain(0.2, 0.0)) - (-2.0)).abs() < 0.1);
+        // a −3 dBFS recording is normalised exactly
+        assert!((db(0.708 * waveform_display_gain(0.708, 0.0))).abs() < 0.1);
+        // a −52 dBFS "silent" stream stays near the floor of the −48 dB display range
+        assert!(db(0.0025 * waveform_display_gain(0.0025, 0.0)) < -39.0);
+        // digital silence does not blow up
+        assert!(waveform_display_gain(0.0, 0.0).is_finite());
+        // clip gain still applies
+        assert!((db(waveform_display_gain(1.0, -6.0)) - (-6.0)).abs() < 0.1);
+    }
+}