[stigmagit]

Improvement: a C runtime package can raise an event with no site (build.py --raise)

What was missing. docs/module-guides/machines.md asks a machine that starts its own note from a trig to ask the
base's raised hook ev_trig_muted first, so a mute module's dropped trigs start no note. An octabam module does that
with "raise" in its runtime.json; a C runtime package (the sources rows of octatrack-module-packages.json) had
no way to: build.py took --subscribe only for C sources. SYNTIZ (#1028), the first native machine written in C,
needs it, and so will the next ones.

What changes.

• build.py takes --raise EVENT for C sources (repeatable). It goes through the same subscribe(..., raises=...)
  that runtime.json's raise already uses: the module defines its zero word raise_EVENT in initialized data,
  and the event's hook word takes RAISE and that word's offset. A package that only raises is ABI 8, as one that
  subscribes.
• scripts/octatrack-module-packages.py passes a row's "raise" list to build.py, beside "subscribe".
• sdk/runtime/loader/README.md and machines.md say how; a changelog entry.
• The reviewed recipes name build.py's new source hash, as Fix: enforce ColdFire safety checks for C runtime modules #949 did: the images are unchanged and the recipes still
  check every compiled byte. The beta base is regenerated (build.py is one of its inputs) and the base-bound
  packages are bound to it, their bytes unchanged.

How I checked it.

• New test_c_sources_raise_an_event_as_a_runtime_json_does (sdk/tests/test_runtime_loader.py): a C source that
  raises ev_trig_muted builds, its hook word at index 35 holds RAISE and the offset of a zero word; --raise ev_voice_bind (a site event) is refused. It and the existing raise test pass with m68k-elf-gcc.
• npm run octatrack:module-packages -- --check: every package identical. npm run pr:ready: Ready.
• Existing C packages (Beat Slicer's row) build as before: they pass no --raise.

What the guides lacked. machines.md told a machine to raise ev_trig_muted "in runtime.json", which a C
runtime package cannot have; it now names --raise and the package row's raise list.

Also in this PR: twin lessons for docs/OCTATRACK_DRIVING.md

The USB bridge fix this PR first carried is #1024's, already in main: the owner's refresh (005e517) took the bridge
files from main and dropped the duplicate changelog entry. The driving-guide edits below stay.

What the guides lacked. docs/OCTATRACK_DRIVING.md warned about Keeps on a twin, not that a boot can come back
short of a machine without any Keep: the boot's restore refused SYNTIZ as busy while the project played, then every
submission was refused until a RAM boot of the flashed base's own image with the transport stopped. The guide now
says so and to check inventory after any twin and any power cycle.

Two more (30 October 2026, the author's MKII, twin 04e94075): the guide did not say that npm run device -- lifecycle keeps twice: it
accepts a test module, then its removal. Run on a twin as a read, it rewrote both kept-set files, and the flashed
base booted with no modules until they were tried and kept again. Nor did it say how to listen to a machine on a twin:
a project loaded while the machine was absent keeps its tracks drawn as FLEX until a trig or a Part change, and a
machine's trial that ends while tracks name it answers needs-recovery, with broken audio afterwards. The guide now says
to start the try first, load the project after, and load a project without the machine before the trial ends.

The branch also merges current main: machines.md's muted-trig line keeps main's new START bullet and adds the
--raise sentence; the recipes name the merged build.py's hash, and the beta base and package rows are regenerated.

🤖 Generated with Claude Code

[repeat98]

Thanks, --raise fits well: it goes through the same subscribe(..., raises=) path that runtime.json's raise uses, and the new test covers it.

One thing: the USB bridge fix here (usb_bridge.py, scripts/octatrack-usb-bridge.py, its changelog entry and the driving-guide line) is the same fix #1024 has carried since 27 October, for the same Rytm-plugged-in case. Please drop it from this PR so it stays one change. Both PRs also regenerate the beta base, so whichever merges second needs main merged in and npm run pr:ready run again. If you'd rather I push that refresh to your branch, just say so.

[repeat98]

I've pushed the refresh to your branch (005e517): main merged in, the bridge files taken from main (#1024's fix, now merged) and the duplicate changelog entry dropped, your driving-guide edit kept, and the beta base regenerated. npm run pr:ready says Ready. Please pull before your next push.