Merge branch 'flow/m3te/P10.7' into m3te-plan

assets/audio/LICENSE.txt

@@ -8,10 +8,14 @@ domain), chosen and processed to EVOKE a sweet, juicy, tactile candy-game feel.
   >>> independent CC0 assets that merely evoke a similar candy/jelly character.
 
 All files are delivered in exactly the format iOS System Sound Services loads
-directly via audio.sx: WAVE / mono / 44100 Hz / signed-16-bit PCM. The whole
-bank is peak-limited to a quiet -9 dBFS (System Sound Services has no runtime
-volume, so loudness is baked into the file); the lose "aww" sits a touch lower
-at -12 dBFS.
+directly via audio.sx: WAVE / mono / 44100 Hz / signed-16-bit PCM.
+
+The bank is deliberately SOFT and GENTLE (P10.7 softening pass, see "Softening"
+below): every cue eases in instead of snapping, the brittle top end is rolled
+off for a warm rounded tone, and loudness is dialed well down. System Sound
+Services has no runtime volume, so loudness is baked into the file: the bank
+peaks at a quiet -15.5 dBFS, with the swap blip and the lose "aww" softer still
+at -17.5 dBFS.
 
 Licenses
 --------
@@ -56,7 +60,7 @@ lose.wav    — soft, gentle, descending "aww" (not a buzzer).
               github.com/sgossner/VCSL .../Idiophones/Struck Idiophones/Marimba
               The two real marimba notes are sequenced into a gentle descending
               B4->G4 gesture (ffmpeg adelay/amix), ~490 ms, faded, normalized to
-              a softer -12 dBFS.
+              the softer -17.5 dBFS.
 
 clear.wav   — "confirmation_001" from Kenney "Interface Sounds" (CC0).
               https://kenney.nl/assets/interface-sounds
@@ -71,3 +75,26 @@ trimmed to its transient, optionally pitch-shifted / sequenced with real ffmpeg
 DSP (asetrate/aresample, adelay/amix, afade), peak-normalized, and re-wrapped to
 the canonical WAVE/LEI16/44100/mono container with afconvert. The shipped game
 never runs any build tool; it only loads the finished WAVs.
+
+Softening (P10.7)
+-----------------
+
+The candy CHARACTER of the P10.6 bank was right but still hit too hard, so each
+finished cue was re-processed with real ffmpeg DSP — no re-synthesis, same CC0
+sources — to be SOFT and GENTLE:
+
+  1. Onset eased in   — a short qsin fade-in (8-14 ms) tames the attack transient
+                        so pops "bloop" instead of "snap".
+  2. Highs rolled off — a warm two-pole low-pass (2.6 kHz swap/lose, 3.0 kHz
+                        match/combo, 3.6 kHz win) removes the brittle/edgy top
+                        end. Spectral centroid drops roughly 40-60% per cue; the
+                        > 4 kHz energy fraction collapses (e.g. win 76% -> 28%,
+                        combos ~10% -> <1%). The win keeps a little glockenspiel
+                        sparkle by design; everything else is rounded right down.
+  3. Loudness lowered — re-normalized to the soft -15.5 / -17.5 dBFS targets
+                        above (down from ~-9 / -12 dBFS), lowering BOTH peak and
+                        RMS on every cue, then re-wrapped with afconvert.
+
+The cascade ladder is preserved: combo1..5 still ascend in fundamental pitch
+(measured dominant partials unchanged at 1045 < 1173 < 1317 < 1566 < 1758 Hz,
+all below the 3.0 kHz low-pass corner). A short qsin fade-out rounds each tail.