feat(firmware): bump memllib to current upstream and dissolve the fork

The §7.5 decision was "rebase our commits onto upstream, then vendor". On
inspection there was no rebase to do, and the inventory that recommended it
(7a30da9) was wrong on two counts — both corrected in this commit.

All three fork commits touch only `examples/`. That directory is not in the
sketch symlink forest (firmware/MEMLNaut-NISPS/src/ links exactly audio,
hardware, interface, synth, utils, PicoDefs.hpp), so it is never compiled.
They existed to let the RL code build against nisps/core *while it was being
ported*, and that port is done — nisps/ml/{jolt,ou_noise,feedback,geo_push}.hpp
cite the upstream sources directly. Two of the 31 commits I flagged as
"work we are missing", `9fcd459 jolts` and `d0d8a72 noise`, are precisely the
ones already absorbed.

So the question was never rebase-or-not but which snapshot to vendor. Current
upstream, because the staleness was already costing us:

  - upstream has DisplayDriver::NavigateToView; the pin does not. The SelfTest
    variant called it and had been failing to compile (worked around in
    b953681). It was written against a newer memllib, not against nothing.
  - e291192 "l r input swap" is a hardware bug fix: the physical L/R input
    sockets are wired to opposite codec ADC channels. Every mode on the old
    pin sees its stereo input backwards.

Verified by building all three variants with arduino-cli:

  SLPWorkshop  145348 flash (+320)  87388 RAM (+4)
  PAFSynth     145300 flash (+312) 107060 RAM (+4)
  SelfTest     141840 flash (+320)  12028 RAM (+4)

Exactly one compile error: the .ino used kSampleRate in a constexpr, and
upstream 1997699 made it a runtime `extern size_t` so a mode can pick its own
rate. constexpr -> const; it is a once-per-second diagnostic print. The uniform
+316 is the AudioDriver/DisplayDriver deltas — the bulky new upstream code
(GrainDelayI16, ReverbI16, ModFXI16, CCSelectView, RLView, VUMeterView,
PSRAMManager) is header-only and unreferenced, so the linker drops all of it.

.gitmodules points at upstream again: with the pin on an upstream commit, the
fork holds nothing the firmware compiles. Phase 0 pointed it at the fork only
because b37fc53 existed on no remote. The fork's feat/nisps-core-swap branch
stays pushed; nothing is destroyed.

Refs: plan §5, §7.5; ALIGNMENT Q4.

.gitmodules

@@ -1,3 +1,10 @@
 [submodule "src/memllib"]
 	path = src/memllib
-	url = https://github.com/monkey-w1n5t0n/memllib.git
+	# Upstream (the lab's shared library), not the monkey-w1n5t0n fork.
+	# Phase 0 pointed this at the fork because the pin b37fc53 existed on no
+	# remote. Those three fork commits touch only `examples/`, which the
+	# firmware never compiles (it is not in the sketch symlink forest) and whose
+	# content is already ported into nisps/ml/{jolt,ou_noise,feedback,geo_push}.
+	# With the pin moved to an upstream commit, upstream is the correct source.
+	# The fork's feat/nisps-core-swap branch is still pushed; nothing is lost.
+	url = https://github.com/MusicallyEmbodiedML/memllib.git

ALIGNMENT.md

@@ -84,9 +84,17 @@ Schemas declare per-mode dims and since P5.3 both targets honour them. Is the mi
 
 Legacy a-immersive was mobile-first; Manifold is desktop-first. Defer until user data exists.
 
-### Q4: Who owns memllib? (2026-07-21)
+### Q4: Who owns memllib? — DECIDED, half-executed (2026-07-21)
 
-Fork-pin (PIO `lib_deps` on `monkey-w1n5t0n/memllib`), vendor the actually-used subset, or upstream the nisps-swap to MusicallyEmbodiedML? Requires the load-bearing-surface inventory (plan §5). Phase 0 pushes the branch either way.
+Operator decision: **vendor**, self-contained in this repo. The inventory
+(`docs/specs/recon/memllib-usage-inventory.md`) settled the shape: there is no small load-bearing
+subset — it is all of memllib bar `examples/` (~1.8 MB, 24/24 compiled TUs link). The fork is
+dissolved: its three commits touch only `examples/`, which the firmware never compiles and whose
+content already lives in `nisps/ml/{jolt,ou_noise,feedback,geo_push}.hpp`, so the submodule now
+points at upstream and is pinned to current `main` (verified: all three variants build, +316 bytes
+flash, and it brings the `l r input swap` hardware fix plus the `NavigateToView` the SelfTest
+variant was already written against). **Remaining: the vendoring copy itself**, which lands with the
+PlatformIO cut (plan §5). Delete this entry when it does.
 
 ### Q5: Legacy feedback modes — delete or keep for A/B? (2026-07-21)
 

docs/specs/plans/simplification-plan.md

@@ -100,15 +100,17 @@ DESCRIPTIVE — it documents that engine voice spaces already square the value i
 One cut, no dual path. `firmware/useq-celium/` already proves the PIO + arduino-pico pattern in-repo.
 
 - `platformio.ini` with one `[env]` per firmware variant passing `-DMEMLNAUT_MODE_TYPE=<alias>`; selftest becomes a plain `-DNISPS_SELFTEST=1` env. Deletes: the sed/python machinery in `firmware-common.sh` that **mutates the committed .ino**, the `.ino` comment-registry, the entire `NISPS_ST_*` token-paste table (already silently missing the currently-active SLPWorkshop variant — L12), the sketch-tree symlink forest, and the global TFT_eSPI library mutation (handled via PIO lib config/build flags instead).
-- **memllib consumption decision** (§7.5): the prerequisite inventory is DONE —
+- **memllib consumption decision** (§7.5): **SETTLED — the submodule bump landed; the vendoring copy
-  `../recon/memllib-usage-inventory.md`, built from the firmware linker map rather than grep. Result:
+  is what remains.** See `../recon/memllib-usage-inventory.md`. Result: all 24 compiled TUs link, so
-  all 24 compiled TUs are linked, so "the load-bearing subset" is ~1.8 MB / 84 files — essentially
+  the vendoring surface is ~1.8 MB / 84 files — all of memllib bar `examples/`; there is no small
-  all of memllib minus `examples/`. There is no small subset to lift. It also surfaced a
+  subset to lift. The operator chose "rebase then vendor", but on inspection **there was no rebase to
-  consequence that was not visible when §7.5 was decided: memllib is the LAB's shared library and
+  do**: all three fork commits touch only `examples/`, which is not in the sketch symlink forest and
-  our fork is 3 commits ahead but **31 behind** (+2034/−153 across the files we link, including
+  is never compiled, and whose content is already ported into `nisps/ml/`. So the fork is dissolved
-  `new staticmlp`, `jolts`, `verb`). Vendoring freezes that gap permanently. Recommended: rebase the
+  and the submodule is repointed at upstream, pinned to current `main` — verified by building all
-  three NISPS commits onto upstream `main` first, THEN vendor — it is the only moment that merge is
+  three variants (+316 bytes flash, one `constexpr`→`const` fix in the `.ino` because upstream made
-  cheap. Operator call before the cut.
+  `kSampleRate` runtime-settable). That bump brings the `l r input swap` hardware fix and the
+  `NavigateToView` the SelfTest variant had already been written against. **Vendor from this
+  snapshot**, recording the upstream commit so a re-sync stays a documented diff.
 - Then **S9**: a CI job compiling 2–3 representative envs with cached toolchain — firmware enters an automated gate for the first time.
 
 ## §6 Phase 5 — Vision-facing architecture (each item spec-first, own session)

docs/specs/recon/memllib-usage-inventory.md

@@ -46,39 +46,67 @@ Firmware's direct include surface is just six entry points — `audio/AudioDrive
 of memllib minus `examples/`.** There is no small subset to lift. `synth/` alone is 1.3 MB (mostly
 `maximilian.cpp`) and all three of its TUs link.
 
-## The consequence that was not visible when §7.5 was decided
+## The fork divergence — and the correction to this document's first version
 
 `src/memllib` is not a third-party dependency. It is **the lab's shared library**
-(`MusicallyEmbodiedML/memllib`), and our fork is:
+(`MusicallyEmbodiedML/memllib`), and our fork was 3 commits ahead / 31 behind.
 
-- **3 commits ahead** — `bf9691c` (swap memlp includes for nisps/core, inline RL utilities),
+**The first version of this document (commit `7a30da9`) drew the wrong conclusion from that, and
-  `32cc831` (newlib RNG fix), `b37fc53` (seed-helper rename). These are the NISPS-specific changes.
+recommended a rebase-then-vendor on the strength of it. Two facts checked afterwards overturn it:**
-- **31 commits behind** — including `770a990 new staticmlp`, `9fcd459 jolts`, `d0d8a72 noise`,
-  `616b8e7 cc select`, `a8bf255 verb`, `671abbe screen ordering`, `e291192 l r input swap`.
-  Diffed against the five subdirs we link: **30 files changed, +2034 / −153**.
 
-Vendoring freezes that gap permanently and converts "we are behind the lab" into "we have our own
+1. **All three of our commits touch only `examples/`** — `bf9691c` (swap memlp includes for
-divergent copy". That may be exactly what is wanted — self-containment is a real goal, and the lab's
+   nisps/core, inline RL utilities), `32cc831` (newlib RNG fix), `b37fc53` (seed-helper rename).
-`main` is not obviously a branch we track — but it is a fork of a colleague-shared library, not a
+   `examples/` is **not in the sketch symlink forest** (`firmware/MEMLNaut-NISPS/src/` symlinks
-snapshot of a vendor drop, and the 31 commits contain work (`staticmlp`, `jolts`) that sounds
+   exactly `audio hardware interface synth utils PicoDefs.hpp`), so it is never compiled. Those
-directly relevant to this project.
+   commits existed to let the RL code build against `nisps/core` **while it was being ported**, and
+   that port is finished: `nisps/ml/{jolt,ou_noise,feedback,geo_push}.hpp` cite the upstream sources
+   directly. So there is nothing of ours to carry forward, and **no rebase to perform** — vendoring
+   the five linked subdirs drops `examples/` and our three commits with it.
+2. **Two of the "31 commits of work we are missing" were already absorbed.** `nisps/ml/jolt.hpp:3`
+   cites `9fcd459 "jolts"` and `nisps/ml/ou_noise.hpp:3` cites `d0d8a72 "noise"` as their sources.
+   The alarm in the first version — "the 31 commits contain work that sounds directly relevant" —
+   was therefore overstated for exactly the two commits it named.
 
-## Options, with what each costs
+So the real question was never "rebase or not" but **"which snapshot do we vendor: the pinned
+`188496d` we build today, or current upstream?"**
 
-1. **Vendor as decided.** Copy the five subdirs into the repo, drop the submodule and the symlink
+## The staleness was already costing us
-   forest, delete `examples/`. Repo grows ~1.8 MB. Self-contained, no submodule init, PlatformIO
-   gets a plain `lib/` — this materially simplifies the §5 migration. Cost: the 31 upstream commits
-   become a manual merge, forever.
-2. **Vendor, but first rebase our 3 commits onto upstream `main`.** Same end state, except the
-   snapshot is current rather than 31 commits stale. Costs one merge now (the three commits are
-   small and mechanical), and it is the only moment when that merge is cheap.
-3. **Fork-pin (status quo + Phase 0's fix).** Already works: the pin is reachable, CI is green,
-   fresh clones build. Not self-contained, and keeps the submodule friction PlatformIO would rather
-   not have.
 
-**Recommendation: option 2.** The operator's goal (self-contained) is satisfied identically by 1 and
+Upstream `main` **has `DisplayDriver::NavigateToView`** (`display/DisplayDriver.hpp:54`); the pinned
-2, but 2 does not silently discard `staticmlp`/`jolts`/`verb`. Doing the rebase after vendoring
+commit does not. The SelfTest firmware variant called it and had been failing to compile — fixed in
-means doing it against a copy that no longer has upstream history — i.e. never.
+`b953681` by routing around the missing method. That variant was not written against broken code; it
+was written against a **newer memllib than the pin**. The gap is not theoretical.
 
-This decision gates the PlatformIO migration (plan §5), which otherwise has to keep the submodule
+Upstream also carries `e291192 "l r input swap"`, a **hardware bug fix**: the physical L/R input
-and its symlink workaround.
+sockets are wired to the opposite codec ADC channels, so `AudioDriver.cpp` now swaps them at the
+lowest level. Every mode on the pinned commit sees its stereo input backwards.
+
+## Verified: current upstream builds, and costs almost nothing
+
+Submodule moved to `e291192` (upstream `main`), all three variants built with `arduino-cli`:
+
+| variant | flash | Δ vs pin | RAM | Δ |
+|---|---|---|---|---|
+| SLPWorkshop | 145348 | +320 | 87388 | +4 |
+| PAFSynth | 145300 | +312 | 107060 | +4 |
+| SelfTest | 141840 | +320 | 12028 | +4 |
+
+Exactly **one** compile error had to be fixed: `MEMLNaut-NISPS.ino:169` used `kSampleRate` in a
+`constexpr`, and upstream `1997699 "mode sample rate"` made it a runtime `extern size_t` so a mode
+can choose its own rate. `constexpr` → `const`; it is a once-per-second diagnostic print.
+
+The +316-byte uniform delta is the AudioDriver/DisplayDriver changes. The bulky new upstream code
+(`GrainDelayI16`, `ReverbI16`, `ModFXI16`, `CCSelectView`, `NameInputView`, `RLView`, `VUMeterView`,
+`PSRAMManager`) is header-only and unreferenced, so the linker drops all of it.
+
+## Where this leaves the vendoring
+
+Vendor **from current upstream**, not from the old pin. Take the five linked subdirs +
+`PicoDefs.hpp`; drop `examples/` (~1.8 MB → the vendored surface). Record the exact upstream commit
+so a future re-sync is a documented diff rather than an archaeology exercise.
+
+The submodule now points at **upstream** rather than the fork: with the pin moved to an upstream
+commit, the fork has nothing the firmware compiles. The fork's `feat/nisps-core-swap` branch remains
+pushed, so nothing is destroyed.
+
+This unblocks the PlatformIO migration (plan §5).

firmware/MEMLNaut-NISPS/MEMLNaut-NISPS.ino

@@ -166,7 +166,10 @@ void loop() {
             blip_counter = 0;
             Serial.println(".");
             digitalWrite(33, HIGH);
-            constexpr float audioHeadroomMul = 1.0f / (1000000.f * 48.0f / kSampleRate);
+            // `const`, not `constexpr`: memllib made kSampleRate a runtime
+            // `extern size_t` so a mode can pick its own rate. This is a
+            // once-per-second diagnostic print, so the divide is free.
+            const float audioHeadroomMul = 1.0f / (1000000.f * 48.0f / kSampleRate);
             Serial.printf("ml: %d, aud: %d, q: %f\n",
                           PERF_GET_MEAN(MLSTATS),
                           AUDIOLOOP_MEAN,

src/memllib

@@ -1 +1 @@
-Subproject commit b37fc53864545f7acca3ce2fba734e00ee38b58d
+Subproject commit e291192d8e4f2fca7b79670c4df9c2ec8bdf03cd