[npp1993]

Requested by Nathaniel

#943 (base flags SETUP_ALL and LIST) landed on main in #1017; this branch carries main at 4690b6eb.

Analog BD 0.1.6-experimental

What changes for players:

• SRC SETUP layout: TYPE, FIXD and ACNT on the top row, LOW, HIGH and LPF below.
  • TYPE (808 left, 909 right) replaces the engine browser. It uses the byte the browser edited, so the 808/909 choice and the 808 default are kept.
  • TYPE and FIXD use stock's two-state switch, as on AMP SETUP's SYNC.
• FIXD, on both models: OFF, the default, varies each hit slightly; ON makes every hit open identically.
  • 909: ON fixes the click strength at its mean (0.96) and restarts the noise from its seed. OFF is the 0.1.5 sound.
  • 808: OFF (new) scales each hit's trigger transient by 1 ± 4 %. ON is the 0.1.5 sound.
• LPF: 500 Hz at 0, ORIG at 63, 18 kHz at 126, and OFF fully clockwise (127), the new default.
• HIGH: can now be edited over USB (Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943's SETUP_ALL).
• Double tap: a double tap on TRACK opens the machine list at once (Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943's LIST).
• Labels: four letters: PTCH, DCAY, SWEP, TYPE, ACNT, FIXD.
• DSP: both engines are rescheduled with bit-identical audio and state.
  • On a MKII (dev base meter, one voice, 16 trigs a bar), one voice loads its DSP core about 26 % (909) and 21 % (808) less than 0.1.5; peaks −18 % / −20 %.
  • Emulator count (dsp_host, 52 scenarios): about 15 % fewer executed instructions on average; peaks about 8 % lower on the 909 (6534 → 5999 per 16-sample block) and 5 % lower on the 808 (4260 → 4034).
  • The declared price (411 per voice) is unchanged, so no voice gains room.
  • Resident engine: 1009 P words per core (998 in 0.1.5): −26 from the rescheduling, +37 for the FIXD tests and the 808 variation. Card build: 1008 (997).
• 0.1.5 projects are not converted. ACNT, LPF and FIXD moved slots, so README and the release notes say to set them again on each Analog BD track. Until then a 0.1.5 track loads:
  • its old LPF value as ACNT (127 by default);
  • the old hidden slot's 0 as LPF (500 Hz);
  • FIXD OFF.

Review follow-up (repeat98's 26 October review)

1. 0.1.5 projects. There is no AB/2 migration. AB/1 is also in the base (known[], machine_contributors.c, the card composition, the Part-restore tests). A new signature needs base code so that an absent Analog BD's tracks stay protected through the boot restore (finding 24). With the base in flux, I'd rather add no migration code to the base or the module for an experimental module whose re-set is documented. What changed instead:
   • FIXD now plays ON only when its stored byte is 1. A 0.1.5 track's accent byte in that slot reads OFF on every track, so unselected tracks no longer play FIXD ON.
   • The selected-track clamp to 0/1 stays.
   • README and the release notes say plainly what to set again.
2. Captures: fresh ot_emu captures of this source (SETUP, 808 page, 909 page). They come from a native card build with Analog BD alone, media/capture.json records the plan, and every image was opened. No retainedUiVersion or release-documentation waiver is needed.
3. Docs: FIXD is on both models and the 808 varies with FIXD OFF (README, TESTING.md, captions). The P-word count is corrected. TESTING.md has one current 0.1.6 section.
4. DSP claim: reworded as above in the changelog, README, TESTING.md and the manifest's resource text, with the chip measurement kept apart from the emulator count.
5. Small things: the changelog is dated 2026-10-27 and says "HIGH can now be edited over USB". if(1) return; in ab_engine_open has a comment saying why the browser code stays compiled (the card build's list hooks and the CF verifier name its symbols).

Compatibility: conflicts unchanged (synth). Base-bound: this release needs the regenerated beta base 6483508a, which carries 0.1.6's trust row (rebuilt on main after #1017, #1026 and #1030). Analog BD's package is f83c7f81. The trust row, the recipe and the packages are in this PR. The Analog BD package is a2def725.

Attribution: Analog BD is repeat98's module, and repeat98 is also the owner reviewing this PR; I have no separate author approval to link.

What the guides lacked: how to make a SETUP control a two-state switch, and the encoder-E rule. Both were fixed in #943, now on main (docs/module-guides/machines.md). This release carries no guide edit.

Owner approval needed

modules:check (and with it modules:generate in pr:ready) fails until sdk/analog-bassdrum-build-approval.json covers 0.1.6. I did not write it. The fields it needs:

• version: 0.1.6-experimental
• sourceSha256: 0ddc82a295d320300a0e03c0c9cf836d19d887dfd2d340ff442d5185a4c23d43
• waived: current-build-hardware, chip-worst-case-cycles, complete-memory-bounds. No release-documentation or retainedUiVersion, now that the captures are fresh.
• approvedBy, approvedOn, ownerStatement and reason: yours

With a stand-in record committed locally (never pushed), npm run pr:ready says Ready (28 October, on main at 4690b6eb, check passed in 238 s).

Hardware results

The rebuild on base 6483508a (package f83c7f81) differs from the tested build only in the base link (the lease entries moved with main's base); it is not yet tested on hardware.

Tested build: base 64ce4825, package a2def725 (MKII, reported by me, npp1993, 27 October):

• Base 64ce4825 RAM-booted, Analog BD a2def725 kept: self-test passed. SRC SETUP reads TYPE FIXD ACNT / LOW HIGH LPF with 4-letter labels; TYPE flips 808/909 live; FIXD shows and switches on both models; encoder E moves HIGH; LPF reads OFF fully clockwise; the double tap opens the machine list at once with no flicker.
• The same package flashed in my personal base 5afa623e (this branch plus a separate base fix, not in this PR) with 13 other modules: self-test passed (14/14, 9 DSP effects together, cores 77 % / 80 %); after a power cycle it started from that slot and reloaded all 14.
• DSP load on the chip: dev twins of Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943's base with 0.1.5 and of this base with 0.1.6, one voice on T1, 16 trigs a bar, the same settings, 12 readings of the base's core meter:

	0.1.5	0.1.6	change
909, mean	545 (+141 over 404)	508 (+104)	−26 %
808, mean	495 (+91)	476 (+72)	−21 %
909, peak	554 (+142 over 412)	529 (+117)	−18 %
808, peak	506 (+94)	487 (+75)	−20 %

• On the flashed personal base later that evening: T2's window switched to T1 shows the machine list, and a double tap on a FLEX and on a STATIC track opens their pools.
• Not tested on the unit: loading a project saved with 0.1.5 (none at hand). There is no migration; the expected behaviour is the documented one (set ACNT, LPF and FIXD again).
• Base limits seen in the same run, not caused by this module: a FLEX ↔ Analog BD switch kept the old machine's live parameters until SOURCE CLEAR (fixed on its own base branch), and right after a boot Analog BD is silent if PLAY comes before its engine is placed, which happens only while stopped (a separate base fix is planned).

Earlier 0.1.6 builds, reported by me (npp1993) on a MKII. The last was flashed base 9976a5e8 with Analog BD dfcb89a1 and 13 other modules. These worked:

• the layout and labels, TYPE flipping live, FIXD on both models
• HIGH moves; LPF OFF fully clockwise and at SOURCE CLEAR
• save/reload keeps every SETUP value
• the double tap opens the machine list instantly with no flicker; T2 → T1 shows the list
• the 14-module set survives a power cycle

Validation (this build)

• Card-build packages: rebuilt in the toolchain image (build-modules-isolated.sh, modules:import). The requested GNU link proofs were refreshed (export-requested-link-proofs.py).

• Recipe pins: the reviewed Analog BD source pins were moved for control.c/control.s. Analog BD's recipe was recaptured against the base's lease symbols. The other machine recipes re-pin machine-ports.json, and the Tape Echo and Euclid standalone recipes re-pin the card-build assets.

• module:verify: 164 selections, 0 mismatches (rerun on base 06bdc59e; module:doctor confirms it still matches on 6483508a).

• Composition proofs: Analog BD's 212 proofs were re-exported natively on base 06bdc59e (146 builds, 66 refusals; the refusals are the same set as on main). verify-analog-bd-native passes.

• dsp_host and exact render: unchanged from the previous push, since the DSP code is unchanged. exact-render.json holds 0.1.6's hashes.

• npm run pr:ready: Ready with the stand-in approval (check passed in 189 s). Without it, only the approval record is missing.

• Changing a module someone else wrote: author said yes (the author is the owner reviewing this)

• I read the machines guide, and TESTING.md lists what I did not test.

• Source and media are mine or properly licensed; nothing comes from Elektron firmware.

• Screenshots are real captures of the current module UI (fresh ot_emu captures of this source).

🤖 Generated with Claude Code

[repeat98]

Thanks for this, and for the careful DSP work. The red checks are expected: modules:check fails until I write the 0.1.6 approval record. Before I do, a few things need changing:

1. 0.1.5 projects change sound when they load. ACNT and LPF moved slots, and the Part signature stays AB/1, so nothing can migrate the bytes. A track left on 0.1.5's defaults loads with ACNT 127 (old LPF byte), LPF 0, which is now 500 Hz (old hidden slot 11), and FIXD ON (old ACCNT byte). ab_clamp_switches (control.c:129) only fixes the selected track, and the DSP reads any nonzero FIXD as ON (bd909.asm, zq06), so the other tracks play FIXD ON until you select them. Octabam's trap list warns about exactly this ("A PART SAVED UNDER AN OLDER SLOT LAYOUT…"). Please migrate them: for example a new signature AB/2 (sig_lo 1), with AB/1 tracks rewritten once in every Part copy when a project loads, so a 0.1.5 project keeps its sound. README line 34 should then say so; today it says saved patches keep their control bytes.

2. Fresh captures instead of the waiver. You already have ot_emu captures of the new layout. Please replace media/ot-setup.png, ot-909.png and ot-808.png with 0.1.6 captures (scripts/capture-module-ui.py), so the approval doesn't have to waive release documentation or retainedUiVersion.

3. Make the docs say what 0.1.6 does. A few places still describe earlier versions:

   • FIXD is on both models, but README lines 3 and 30, TESTING.md lines 8 and 33, and the ot-setup caption say "909 only".
   • The ot-808 caption says the 808 engine is "unchanged", and TESTING.md line 8 says the 808 has no random source.
   • TESTING.md line 46 says 972 P words, while the manifest says 1008 and line 205 credits the 11 words to "the FIXD switch". It's −26 from rescheduling plus 37 for the 808 variation.

   Please merge the 0.1.6 sections of TESTING.md into one current one.

4. The "15% less DSP" claim. That's the mean count of instructions the emulator executes. The peaks fell about 8 % (909) and 5 % (808), there's no chip measurement, and the declared price of 411 per voice is unchanged, so players don't get more room. Please word the changelog line and the title along those lines ("about 15 % fewer DSP instructions on average, emulator count").

5. Small things: the changelog date is 2026-10-24, the version was never released then, and "HIGH can be edited over USB again" (it never could). Also the if(1) return; in ab_engine_open (control.c:268) leaves the whole engine-browser body as dead code. Fine for now if the UI fixtures need the symbols, but say so in a comment.

The hardware results match the build this PR publishes today (base 9976a5e8, package dfcb89a1). The changes above mean a new build anyway, and other base pull requests are ahead of #943 in the queue, so please hold the next unit run until #943 is rebuilt on current main. I'll merge #943 and this back to back after that.

[repeat98]

Thanks, this is careful work, and the bit-exact reschedule is a nice find. Two things on my side before the approval.

1. I have an unreleased 0.1.6 of my own (the 909 refit: body, PITCH/DECAY ranges, hit-to-hit variation). Yours ships as 0.1.6, and mine follows as 0.1.7 on top of it.
2. This changes the controller, so Analog BD's trust row in the base changes, and every unit would need a base update to take the module. I decided on 22 October that a module update must never need a base rebuild; that's why the engine moved into the package. Could the changes stay inside what the current controller admits?

It's stacked on #943, so it waits for that one, and it now conflicts with main in the base files after #857.

[npp1993]

Thanks for the thorough review. I'll work through all of the items (AB/2 with a one-time rewrite of 0.1.5 tracks, fresh 0.1.6 captures, the doc fixes, the reworded DSP claim, the changelog date and wording, and a comment on if(1) return;) once #943 is rebuilt on main.

On the controller: the point of 0.1.6 is the steady-hits switch (FIXD) on SRC SETUP, with TYPE replacing the engine browser. Those are controller changes by nature, and so is the AB/2 migration you asked for. Only the leaner DSP fits inside the current controller. Since #943 already needs a base rebuild that every unit will take, could Analog BD 0.1.6's new trust row ride that same base update if the two merge back to back, as you planned in your review? If you'd rather keep it separate, I can split this: ship the DSP-only part now and hold the switch until Phase 2 admits controllers without trust rows. Happy to go either way, and 0.1.7 for your 909 refit sounds right.

---

Generated by Claude Code

[npp1993]

I've pushed the review fixes, except one: there's no AB/2 migration. Here's why, and what I did instead.

Why no AB/2. AB/1 is wired into the base as well as the module: known[] in machine_registry.c, machine_contributors.c, machine_composition.py and the Part-restore tests. A new signature would need base code (the signature range) so an absent Analog BD's AB/2 tracks are still protected during the boot restore (finding 24). With the base in flux I'd rather not add migration code to it, or to the module, for an experimental module whose re-set is documented.

What changed instead

• FIXD plays ON only at its stored 1. A 0.1.5 track's accent byte in that slot now reads OFF on every track, so unselected tracks no longer play FIXD ON. The selected-track clamp to 0/1 stays.
• README and the release notes say plainly that 0.1.5 projects keep TYPE but need ACNT, LPF and FIXD set again, and what each one loads as in the meantime.

The rest of your list

• Docs: FIXD is on both models, the 808 varies with FIXD OFF, and TESTING.md has one current 0.1.6 section.
• P words: 1009 resident (998 −26 from the rescheduling +37 for the 808 variation); 1008 in the card build.
• DSP wording: "about 15% fewer executed instructions on average (emulator count); peaks about −8% (909) and −5% (808); the declared 411 per voice is unchanged", in the title, changelog and docs.
• Changelog: dated 2026-10-27, and "HIGH can now be edited over USB".
• if(1) return;: commented with why the browser code stays (the card build's list hooks and the CF verifier name its symbols).
• Captures: fresh ot_emu captures of this source: SETUP, the 808 and 909 pages. No retainedUiVersion or release-documentation waiver needed now.

Both branches carry main at b5ac6142. #943 alone builds beta base b2810a87; with 0.1.6's trust row it's 64ce4825 (Analog BD package a2def725). module:verify: 0 mismatches over 164 selections; the 212 Analog BD proofs are re-exported and verified; pr:ready says Ready with a stand-in approval record (the fields are in the description).

Hardware (my MKII, today). On 64ce4825 (RAM boot) the SETUP layout and labels, TYPE flipping live, FIXD on both models, HIGH on encoder E, LPF OFF fully clockwise and the instant machine list on double tap all work, and the self-test passes. The same package in a flashed personal base, with 13 other modules, passes the self-test (14/14) and survives a power cycle. On the chip (dev twins of #943's and this base, one voice, 16 trigs a bar, the same settings) one voice loads its DSP core 26 % (909) and 21 % (808) less than 0.1.5, peaks −18 % / −20 %; that's separate from the emulator's ~15 % instruction count. Still to run: T2 → T1, the FLEX/STATIC pool double taps and a 0.1.5 project. Two base issues showed up that aren't this module's: a FLEX ↔ machine switch keeps the old live parameters until SOURCE CLEAR (I have a base fix on its own branch), and right after boot a machine is silent if PLAY comes before its engine is placed (separate fix to follow).

---

Generated by Claude Code

[repeat98]

Thanks for working through the whole list. Here's where I've landed on the two open points:

1. The trust row: OK. TYPE and FIXD can't fit inside the current controller, so as a one-time exception 0.1.6's trust row rides Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943's base rebuild. I'll merge Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943 and this one back to back, so b2810a87 never ships on its own.
2. No AB/2: OK too. The documented re-set of ACNT, LPF and FIXD is fine for an experimental module.

Two small things before the merge:

• README "Tests and measurements" (line 64) still describes 0.1.5: "808 output/counts remain unchanged" and "997 P words". Please bring it in line with 0.1.6, as you did in TESTING.md.
• Please merge Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943 into this branch again once the HOOKS.md link fix is in there.

I'll add the approval record for 0.1.6 myself once those are pushed.

[npp1993]

Thanks! Both points are in, and it's ready for your approval record.

• README "Tests and measurements" now describes 0.1.6: bit-identical to 0.1.5 where the settings mean the same (909 FIXD OFF, 808 FIXD ON), 1008 P words in the card build and 1009 resident, the emulator count and the chip measurement kept apart, and the MKII checks.
• Merged main (with Octatrack beta base: more DSP room (receiver in the shared window), FX2 swaps keep their block, with the fix/base-oct28 bundle and the Python firmware codec #1017, which carries Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943; now at d2cb12d6) instead of Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943 itself, since Fix: Octatrack base, SRC SETUP's encoder E and the double-tap window for module machines (SETUP_ALL, LIST flags) #943 is on main and I've closed it. Everything is regenerated with the producers on main's base:
  • beta base 06bdc59e with 0.1.6's trust row (main's is 6541d7e9);
  • Analog BD's recipe recaptured, because main's base moved the lease entries; the package is now f83c7f81 (was a2def725), and the module source is unchanged;
  • the 212 composition proofs re-exported (146 builds, 66 refusals, the same refusals as on main) and verify-analog-bd-native passing; module:verify gives 164 selections and 0 mismatches.
• pr:ready says Ready with a stand-in approval record (removed). The approval needs sourceSha256 0ddc82a295d320300a0e03c0c9cf836d19d887dfd2d340ff442d5185a4c23d43 and the waivers current-build-hardware, chip-worst-case-cycles, complete-memory-bounds. The three red checks are only that record.

The hardware results in TESTING.md and the description are from the 27 October build (base 64ce4825, package a2def725). The rebuild differs only in the base link and isn't on a unit yet.

---

Generated by Claude Code

Addressed: the requested changes are in, and the owner approved the 0.1.6 record.