From 1f38f1a7fbfd8cd3e2f899c7e2dd85003c40f07b Mon Sep 17 00:00:00 2001
From: monkey-w1n5t0n <w1n5t0n@lnfinitemonkeys.org>
Date: Tue, 21 Jul 2026 17:04:35 +0200
Subject: [PATCH] fix(engines): restore MEMLCelium's feedback tap (L7)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

feedback_/fbzm1_/fb_smooth_alpha_ were read every process() call but nothing
ever wrote feedback_, so the path was silently inert. Git history shows this
was a *muted* feature, not dead code: `feedback = v0 * feedbackGain` with
feedbackGain 0.1f (MEMLCeliumAudioApp.hpp @ d095688), then "0; //0.1f" —
zeroed with the value kept in the comment — then the write was dropped
entirely when the engine was ported to nisps/engines/ at the 2026-04-29
rewrite (8d0d47b). Operator decision: port it back, live at 0.1f.

The tap is voice 0's own enveloped output, post-envelope and pre voice-1/mix/
shape — NOT the final output. This differs from PAFSynth's feedback shape;
where the sibling and the reference disagreed, the reference won.

Not schema-exposed: the reference hardcoded it, and MEMLCelium has no voice
spaces to vary it by. If feedback depth becomes a research axis it wants to be
a runtime parameter — noted, not built.

Verification, and a gap worth recording: parity-check.sh PASSES (max delta
2.38e-7) but that is NOT evidence for this change — its harness only exercises
PAFSynth and ChannelStrip, and the impulse baseline runs all-params-0.5, under
which voice 0's sequencer never triggers, so v0_env stays 0 and the new write
is 0 either way. That baseline is blind to this path, not a check on it. So it
was cross-checked directly instead: sequencer params forced to trigger early,
4800 samples native vs WASM through the nisps_engine_* C ABI —

  native vs WASM max abs diff  1.19e-7   (tolerance 1e-5)
  all 4800 samples nonzero, finite, max amplitude ~0.99 (tanh-bounded)
  same run with gain forced to 0 diverges by up to 1.49 — not a no-op

No fixture regenerated. nisps.wasm ships here because it is a tracked artifact
under the Phase 0 freshness gate.

Follow-up: the parity harness covers 2 of 8 engines. "Firmware and WASM share
the same engines" is asserted repo-wide but tested narrowly.
---
 manifold/public/nisps.wasm   | Bin 130040 -> 130068 bytes
 nisps/engines/memlcelium.hpp |  14 ++++++++++++++
 2 files changed, 14 insertions(+)

diff --git a/manifold/public/nisps.wasm b/manifold/public/nisps.wasm
index 7dab364393dbe0d04884655de4cda71619404a67..9eb04a87f08e9e7d4a4978b86579f109f31a5c3d 100755
GIT binary patch
delta 795
zcmY+CO=uHA6vubo>~=TFt_9J02-2BdLFqv#l)4JiMw>KglAaXAV4Aa_z4TBD)nL6U
zxNi{!^(5lKL$E?F0X=x|5TRfM1%(`nryP1HG>05~v)hvNGH>Sn-^Xv>O#i?P`)1=0
zu%Fw1VKc+tVP!#{gX@}B{~QPq(QdAmb}b)vIJq@Vf?!in$GzNXJ><f|j|*L9B4%&V
z+RIjG&F3o9Nfq@CWJ=jzAWtaU0GUvBAEd2p8>FLb7o>@u+zhiYv3Sw9YK|tj5<c+2
zHH1~u=LM(6hU7L&2}>AmMA0CbuQ*ypi`=-n_+&g8QC8jY46Mz<0O#}aLXV<F>X^_`
zE=OH-j+E7tTUSDnz#DmwIk=ubYnicR<Li7i4Yp4AP(ZuUI$(hdJ`9)uoG5rV!RCiL
zxhSf~@JLt3Jw}vShdO5XEId}7d6Z3evf)_f{xB1t7KB48h10wtlsYyFzwz191~bqu
z`ix_?csZ@AQ9xH0I)xg72SMnLg+6q2d{}fx*gr*;+7Pj~>2a}D9AzgL6>34mu5@+$
z0;BBF)QFwbQT90V>o5~%N+PLlHU&FC;MLNjGsbWP0|jmP^bdrA2c=4ihs5o2dUTF=
z%HD)Yqk6&a#-HG=9v({1%ibx2q}Gx2K9=x%`Mftqz<L-z8})sqyMR={6b4Ns-fVfx
pO>dbkO|qbQ{~l}DY_~bN!Pa@|TD@NH|9H9m@9L%In@x86!GH8osw)5h

delta 767
zcmY*XO=uHQ5YC%@n@x5%si;IDgub^CgdP+@YP5&CG-;BiX-`5R(B`V(&%uL;)!2&1
z9(J~<Rq$XD4;}&vp;A!9g9i^icu=g6gBQVrV!#}7aNcGc>*dXx_kHutw=?VajNTuk
zwF7MP#aCEz*(%oa@~}1)i8Nn(LKDGlvXOR-9Bga&FrQ%tZst2IhU<kIoxclpI_IbB
zC*p#Q!F!OVil0CxRBVAvs@Mf-srUh;tzrjc48KgDB(;6TW723OBGiq5=bGo}!f5cb
zLX#kLsoBgD1}+v;#2qah!S&+#TAU0<A`8xl3s=XX4}u9xd7c#UsmuFU&iEqXulWH<
zjHGk$Lov+%iax3*dr)MCMM<oevdqR?rA#a~fGpf7Eu^{4$-eUUEjatkb9fH=OmNJU
zvy7u#KH!EgiuRK^uvkb?4_6z9JE|81e30CRP>afm+GRM@S}P0S7QQWij*dZ-W`z%C
zZm<NVDmj+K`O0aU^^=N|)`xyxlC&BtO)W=3=-8=b&*`=n_Mj)iw*Yr>mA-+!?t)2>
zJEDTrhwcWI@BUR{ZC03iu*PkWQCyx~KD6f}gDsFumt-(qty|y=vPBu?x2xF{{TS|j
zc7bxTu)+NhdR5IHBvE&37kV~;@K^QdaV~7yrAT@r=J`SwQMFWBGbudc4j}zBrm#H1
wg3HRpTzg`U&5W|u_T7uDWwTEE)OB`EGu>vh*`rnL@OzK1u<z}sH`%47zp6*5Hvj+t

diff --git a/nisps/engines/memlcelium.hpp b/nisps/engines/memlcelium.hpp
index ef32e22..5cade08 100644
--- a/nisps/engines/memlcelium.hpp
+++ b/nisps/engines/memlcelium.hpp
@@ -185,6 +185,10 @@ class MEMLCeliumEngine {
         const float p2 = v0_paf2_.play(freq2, freq2 + (v0_paf2_cf_ * freq2),
                                        v0_paf2_bw_, v0_paf_vib_, v0_paf_vfr_, v0_paf2_shift_, true);
         float v0 = (p0 + p1 + p2) * v0_env;
+        // Feedback tap: voice 0's own enveloped output, one block behind (see
+        // fbsmooth above) — matches firmware's `feedback = v0 * feedbackGain;`
+        // taken at this same point (post-envelope, pre voice-1/mix/shape).
+        feedback_ = v0 * feedback_gain_;
 
         // ----- Voice 1 -----
         const float v1_env = v1_amp_env_.play();
@@ -293,6 +297,16 @@ class MEMLCeliumEngine {
     float feedback_       = 0.f;
     float fbzm1_          = 0.f;
     float fb_smooth_alpha_ = 0.5f;
+    // Restored 2026-07-21 (finding L7): firmware history shows this went
+    // 0.1f (live, `modes/AudioApps/MEMLCeliumAudioApp.hpp` @ d095688) -> "0;
+    // //0.1f" (explicitly muted, value preserved in comment) -> the write
+    // was dropped entirely when the engine was ported to nisps/engines/ at
+    // the 2026-04-29 rewrite (8d0d47b) — feedback_/fbzm1_/fb_smooth_alpha_
+    // kept being read in process() but nothing wrote feedback_, so the path
+    // was silently inert. Operator decision: port it back live at 0.1f, not
+    // muted. Not schema-exposed — a single fixed value (MEMLCelium has no
+    // voice spaces to vary it by, unlike PAFSynth's feedback_gain_).
+    float feedback_gain_   = 0.1f;
 };
 
 static_assert(AudioEngine<MEMLCeliumEngine>, "MEMLCeliumEngine must satisfy AudioEngine");