Output Matrix 0.1.1-experimental: always-on LEV label and smaller core-0 DSP, on the USB base

[npp1993]

Output Matrix 0.1.1-experimental, on the USB beta base: the always-on LEV label and a smaller core-0 DSP. Builds on #682 and contains its commit until that merges.

What changes for users

• In MATRIX the LEV box is always labelled with the track's destination (MN, PHN, M+C, …), not only while CUE is held. New detour lev_box_lev at 0x4004dcfa, sharing dest_label with the CUE-held path.
• Core-0 DSP code shrinks from 727 + 42 to 657 + 14 P words (671 in all). The output is identical sample for sample in 26 ot_emu scenarios. The worst case drops from 4,240 to 4,194 executed instructions per frame (275 per sample with the phones crossfade). Both are in the resident engine the beta base places on demand: 675 words instead of 773.

USB base and package

• Re-pinned manifest.py, matrix.s and matrix_mix.asm in resident_dsp.py, stock_hook_controllers.py and test_analog_bd_part_fence.py.
• Controller re-port: the label moves lev_box_cue's stock-equal span from offset 58 to 38, and lev_box_lev repeats the same 14 bytes. That makes 20 sites and six stock copies, with new trust rows for package e2df2fbf…. The package's stock-free scan passes (10, fails at 12).
• Base rebuilt with the beta flags: identity 3e3e3de1…, browser .bin = build_core.py .bin ca8b457f…. On this machine the same command first reproduced main's committed base byte for byte. The lease entries didn't move, so Analog BD's and Sidechain's package bytes are unchanged; the other base-bound packages only change requiredBase.

Checked

• MKII over USB, 16 Oct 2026. I ran the panel checks; the base was RAM-booted and the package installed over USB by Claude Code. Base MAIN.raw 5ca0afbf…, package e2df2fbf…. Five checks, all passed:

  • switch into MATRIX;
  • always-on label;
  • CUE + LEVEL to MAIN, CUE, PHONES, a mono jack and OFF on a playing track, with audio where the label says;
  • MIX as the PHONES level;
  • back to NORMAL.

  Details are in evidence/usb-hardware.md. The first 10 s trial's rollback was refused once with needs-recovery and completed by the base without a further command, which is the case octatrack-link.ts retries.

• module:verify output-matrix: 52 images match, 74 matching refusals, 0 mismatches of 126. Without Fix: native builder places a floating cave in the overflow run before refusing #682 two 11-module selections refused for different reasons, which is that PR's bug.

• Packages: rebuilt in the isolated image and imported (--development). Tape Echo's and Euclid's standalone recipes name the four rebuilt shared asset rows' hashes (provenance only, as in Fix: write Formatter.PLAIN's zero formatter words without a stepped slot (#626) #634 and BBD Delay 0.1.1-experimental: over USB for beta testers, stop kill, cheaper worst case #645).

• Analog BD composition proofs: re-exported natively; all 212 and builderSources identical, only moduleSourceTreeSha256 changes.

• OT UI captures: re-run on a static 0.1.1 image; pixel-identical to 0.1.0's.

• npm run pr:ready: Ready.

Limits

• No power cycle, project or Part reload, MASTER TRACK, STUDIO, or other modules beside it on this version on the unit, because the base was RAM-booted.
• The 32 s emulator stress run in evidence/performance.json was made on 0.1.0 and not repeated.
• module:verify rewrote chooser-metadata.json with POLY8's entry in another position and no content change. I kept main's order so the committed base stays the one tested on the unit.
• These sources are pinned in your resident-engine preparation. This PR moves those pins; tell me if you'd rather re-pin them yourself.

🤖 Generated with Claude Code

[repeat98]

Thanks @npp1993. This is merged as #705: your commits with current main merged in, including E-Verb 0.1.2, because I couldn't push to your fork. E-Verb and this PR changed different rows of the same compiled package files. Each merged file is main's plus your rows, and the build record, the Analog BD proofs and the Tape Echo and Euclid recipes carry the merged hashes.

I also corrected three things in the release notes:

• The DSP code has run on hardware over USB, inside the base; on a card it has run only in the emulator.
• This release publishes base 3e3e3de1.
• The 0.1.1 card image, Part/project reload, MASTER TRACK and STUDIO weren't checked on the unit.

Closing in favour of #705. I'll follow the publish run.

---

Generated by Claude Code