[repeat98]

POLY8 0.3.3-experimental: on card builds with POLY8 and VECTOR, POLY8's shared SRC SETUP hooks now ask VECTOR for its page 2, as the base does over USB. This is the POLY8 half of #1119 (also #842/#843). The VECTOR half is #1141.

Related to #1119

What was wrong

In a card composition, POLY8 replaces VECTOR's SRC SETUP hooks at 0x4003a52e and 0x4003cd98 with mr_setup_edit6/mr_setup_draw6. Those map every module row to its backing pool's stock setup and never call VECTOR. So a double tap on SRC on a VECTOR track shows LOOP … TSTR AUTO instead of SEED/SPAN, which is the reporter's "TSTR is in AUTO".

What changes

• mr_setup_edit6/mr_setup_draw6 ask mr_setup_knob/mr_setup_page first, mirroring the base's mw_setup_edit6/mw_setup_draw6 (same registers and stack slots). This happens only with SRC SETUP's cursor on VECTOR's row and the current track a VECTOR track. The page is then st_setup_page's, and encoders A-F go to st_setup_knob.
• The encoders follow only a page VECTOR actually drew (setup_shown). With VECTOR 0.2.5, whose st_setup_page still wants row 5, everything stays stock, so the two PRs can merge in either order.
• USB: nothing changes. The package drops the machine_registry unit (shared-machine.s) and stays 0.3.1's; the lag note is updated.
• Pins updated: machine-ports.json, machine-recipes.json. Packages are rebuilt in the isolated toolchain and imported, and POLY8's build gates pass (addressing, loaded state, 256 signature masks).

How I checked it

Emulator only (ot_emu from upstream octabam, MKII panel, DSP on, empty scratch card). Images were composed by the site's builder and driven with scripts/capture-module-ui.py. Details are in TESTING.md.

• main (POLY8 0.3.2 + VECTOR 0.2.5 + FM Synth) reproduces the report.
• POLY8 0.3.3 + VECTOR 0.2.6 + FM Synth: VECTOR's page 2 shows. Encoder A gives SEED 4, encoder E gives RPT 2, and the cursor on POLY8 shows the stock setup.
• POLY8 0.3.3 + VECTOR 0.2.5: every screen is pixel-identical to main's. The stock encoders change nothing visible on that row in either build, so the claim that encoders stay stock comes from reading the code, not from observation.
• Not tested: a unit, MKI, a set with Analog BD (VECTOR on row 7), reload, power cycle.

Refreshed 6 Nov 2026

Merged main (about 500 commits). Packages rebuilt in the isolated toolchain and imported; Analog BD proofs re-exported (all 212 unchanged, only the source-tree hash moves); Tape Echo and Euclid recipes re-pinned; the machine recipes name the merged machine-ports.json and the new shared-machine.s (images unchanged). evidence/performance.json now names 0.3.3 (no new measurement: the render and its limits are 0.3.1's, as 0.3.2 did), which moves the native source hash. Two fixes the old head needed and CI never reached (it stopped at modules:check):

• the 0.3.3 release-note line was 206 characters (limit 200); split in two;
• machine_composition.py (the VECTOR/FM bridge) pins shared-machine.c and shared-machine-hooks.s, which 0.3.3 changes, so sdk/tests refused its proofs. Re-pinned; its tests pass. Analog BD's hash-only analog_bd_native_observations.json pins that file for the FLAGS it imports, which are unchanged, so that one pin moves with it. Say if you'd rather that observation record stay untouched.

Checked locally: lint, typecheck, sdk/tests (563, all pass with Node 24), elekloader/compat/pairs, board. Every remaining red is downstream of the record: modules:check, the catalogue still at 0.3.2 (the composer tests: "compiled module versions differ from the catalog"), the release notes being newer than the catalogue, generated module media for the bundle and page tests, and POLY8's native comparison.

Needs you: the approval record

sdk/poly8-build-approval.json binds POLY8's exact source, so modules:generate stops until you approve 0.3.3. I didn't write it. Same shape as 0.3.2's: version 0.3.3-experimental, sourceSha256 06ed886b4ca22c146fcf2436b9794657335888e78718e03cd6ae7a33c86c6c48, the same four waived items (they also cover the doctor's three performance rows), retainedUiVersion 0.2.7-experimental.

Then: npm run modules:generate, npm run module:verify -- poly8 --os <1.40C> (the native builds for this tree are cached on your Mac since 6 Nov; the browser side refuses until the catalogue is at 0.3.3), commit the record, the catalogue and sdk/native-comparisons/poly8.json, and npm run pr:ready.

Whichever of this and #1141 merges second needs a refresh: both rebuild the shared package files and pins.

What the guides lacked: same as #1141, fixed in #1144. Nothing new in the 6 Nov refresh.

🤖 Generated with Claude Code

[repeat98]

Closing this as superseded. POLY8 0.3.3 changed the card build only (VECTOR's second SRC SETUP page beside POLY8), and since today modwerk.app builds no card firmware (#1451, #1453). On the Modwerk base the base itself opens VECTOR's second page, so there is nothing to carry over.