AC Native Story Refresh — Touchpoint Tracker #

Goal: The description of AC Native (and the broader AC system) currently appears in many places — grant proposals, papers, the platter, API docs, marketing copy. Several of these are stale. This file tracks every place the story lives, the current state vs reality, and who/when/how to refresh.

Priority right now: LACMA 2026 proposal (due 2026-04-22, 11:59 PM PST). Items marked 🔥 are LACMA-critical.

Started: 2026-04-21 Driver: @jeffrey (+ Claude Code)

1. AC Native — current feature audit (authoritative) #

Reality check from an exhaustive pass over fedac/native/, git log (last 30 days), PROGRESS.md, SCORE.md, BOOT-REGRESSION-REPORT.md, and the piece inventory. Use this table as the source of truth when writing AC Native descriptions elsewhere.

Claim	Status	Reality (April 2026)
"Boots from USB in under 2 seconds"	❌ stale	Currently ~7.3s cold boot. <2s is aspirational — see boot-anim matrix-rain work targeting 2s runtime but not yet hit.
"Custom C runtime as PID 1"	✅ true	`ac-native` binary is PID 1 (not systemd). See `src/ac-native.c:1535`.
"No desktop, no window manager, no browser"	⚠️ mostly true	True in default path. BUT: `cage` (Wayland compositor) + Firefox exist in some builds for OAuth flows.
"Renders directly to the framebuffer"	⚠️ misleading	Uses DRM dumb buffers via Linux display subsystem. Not raw `/dev/fb0`. Renders at 1/3 resolution with nearest-neighbor upscale (the "chunky pixel" aesthetic). Optional SDL3 GPU backend compiled in, loaded via dlopen.
"Reads input from raw device events"	✅ true	evdev (`/dev/input/event*`). See `src/input.c:315-374`.
"ALSA sample-level synthesis at 192 kHz"	✅ true	Confirmed. `AUDIO_SAMPLE_RATE=192000`, `AUDIO_PERIOD_SIZE=192` (~1 ms latency).
"Polyphonic synthesizer"	✅ true, stronger than claimed	32-voice polyphony (`AUDIO_MAX_VOICES=32`).
"Waveform selection"	✅ true, richer	8 waveforms in notepat: sine, triangle, sawtooth, square, composite, harp, whistle, sample. Plus gun (digital waveguide + biquad body modes), noise, flute (STK model).
"Room reverb"	✅ true	3-tap delay, 0.55 feedback, wet-mix slider.
"Time-of-day-responsive visuals"	⚠️ not verified for AC Native	True for the web notepat piece tint; not confirmed as a native-side feature. Check `pieces/notepat.mjs` on AC Native.
"`code` command launches Claude Code directly on the framebuffer"	❌ misleading	`code` is an alias that jumps to `terminal:claude`. Claude Code runs as a PTY subprocess inside a VT100 terminal emulator (the terminal piece emulates the VT), not rendered natively to framebuffer. The terminal emulator IS native.
"8,466-line musical composition tool `notepat`"	⚠️ outdated	Current `notepat.mjs` is 6,006 lines on AC Native (`pieces/notepat.mjs`). The 8,466 figure is the web version (`system/public/aesthetic.computer/disks/notepat.mjs`). Be precise which one we cite.
"Any x86 laptop"	⚠️ x86_64 UEFI only	No 32-bit. No BIOS-only. No ARM. Primary target: ThinkPad X1 Nano Gen 2. Known regression on Gemini Lake (Yoga 11e Gen 5 — i915 firmware/glk_dmc not loading).

Features worth adding to the story (currently under-represented) #

USB MIDI — notepat sends/receives via USB MIDI devices.
UDP MIDI broadcast — notepat broadcasts MIDI over UDP; status badge in HUD.
Camera + QR scanning — V4L2 + quirc. Used for OAuth flow (scans login URL from screen).
Flite TTS — male + female voices, integrated into audio ring buffer.
Sample recording & playback — 12 simultaneous sample voices, up to 10 s per sample.
SSH via Dropbear — auto-starts on WiFi connect (for dev, can be disabled).
OTA updates — ac-os upload rebuilds and uploads a signed EFI payload; devices verify byte-match before flashing /mnt/EFI/BOOT/BOOTX64.EFI.
Build naming — each build is tagged with an adjective-animal name (e.g. "funny-chatterjee"), embedded in the kernel via AC_GIT_HASH + AC_BUILD_NAME.
20+ shipped pieces on AC Native: notepat, terminal, prompt, os, claude, dj, tapes, chat, machine, lisp, samples, speed, split, power, printing, theme, login, clock, geo, painting, secrets, list, wifi, voice, cc, roz, laer-klokken.
Dual audio outputs — laptop speakers + HDMI secondary.
Master volume + drive FX — tanh soft saturation on the master bus (recent).

Known issues / not-yet-shipped (be honest in grant copy) #

Boot time target of <2s not reached (currently ~7.3s).
Multi-touch (MT protocol B) — planned, not shipped.
Firmware blobs for common WiFi chipsets (RTW88/89, MT7921/25, Intel SOF) — SCORE.md:188 flags as open.
Audio tearing at high echo + many voices (buffer underrun edge case).
Camera preview is Y-channel grayscale only (YUYV), not full color.
No regulatory.db for WiFi → defaults to US 5 GHz channels.

2. LACMA 2026 proposal — 🔥 due tomorrow #

🔥 LACMA-critical copy fixes #

Location	Current	Should be
lacma-2026.tex:118	"boots from a USB stick on any x86 laptop in under two seconds"	"boots from a USB stick on x86_64 UEFI laptops in under ten seconds" (honest, still fast for a full OS)
lacma-2026.tex:118	"renders directly to the framebuffer"	"renders directly via DRM (no compositor, no browser)" (accurate)
lacma-2026.tex:118	"A built-in `code` command launches Anthropic's Claude Code directly on the framebuffer"	"A built-in `code` command drops into a native terminal emulator running Anthropic's Claude Code as a PTY subprocess, so artists can ask an AI coding partner to modify a piece without ever leaving the OS." (accurate, still compelling)
lacma-2026.tex:118	"an 8,466-line musical composition tool called notepat"	Keep "8,466-line" — that's the full web notepat, which ALSO ships to AC Native with some native-specific paring. Or qualify: "an 8,000+ line musical composition tool."
lacma-2026.tex:118	Mentions: room reverb, waveform selection, time-of-day visuals	Add: 32-voice polyphony, USB & UDP MIDI, sample recording, Flite TTS
LACMA-2026-APPLICATION-DRAFT.md	Same stale phrasing as tex	Sync after tex is updated

Status: ☑ drafted ☑ applied ☑ PDF recompiled (492 words, under 500 cap) ☐ submitted

🔥 LACMA-critical content additions #

Figures — already refreshed with sosoft cards (Fig 3, Fig 4) ✅
Video demo — script in video-script.md; filming tomorrow
Fig 1 caption — "600+ pieces across 2,800 registered users" — check this is current (we verified 371 built-in + 265 user ≈ 636)
Images must be JPEG for Submittable — convert PNG figures before upload
Submittable account → paste draft → submit

LACMA nice-to-have (if time) #

Mention ~20 shipped pieces on AC Native (adds weight to "it really runs")
Mention NELA Computer Club as existing public-engagement evidence (done in bio, could echo in Public Engagement statement)
Note OTA update system as a credibility signal (we can field-update installed stations)

3. Platter — refresh #

The whistlegraph-platter/ and papers/sync-platter.mjs imply a "platter" concept (curated presentation/deck). Status of anything that serves as a current AC Native pitch:

Check what's in whistlegraph-platter/ — is it AC Native relevant or just Whistlegraph?
papers/sync-platter.mjs — does it publish a current platter artifact?
Where does the current pitch "platter" live? If not yet assembled, we may want to create one at grants/lacma-2026/platter.md or platter/ac-native.md that bundles: one-line, hero image, feature list, stats, short bio, 3 links.
Identify authoritative source: does any README or landing page serve as the AC Native "platter" for outside audiences?

4. API surface — refresh #

The disk API is the contract pieces program against. Documentation state:

system/public/aesthetic.computer/lib/disk.mjs — ~572 KB, the API itself. Reality is the code.
kidlisp/README.md — KidLisp language docs; says "118 built-in functions across 12 categories." Check count is still right.
Papers that cite API counts: papers/arxiv-kidlisp/, papers/arxiv-kidlisp-cards/, papers/joss-ac/, papers/joss-kidlisp/ — all claim specific function counts. Need sweep.
AC Native API delta — does AC Native expose the full disk API, or a subset? Enumerate what's NOT available native (e.g. WebGL, WebRTC, browser-only hooks). This is grant-relevant: describes what the native runtime trades for directness.

5. Architectural image — refresh #

The three-layer story (OS / language / network) is the canonical pitch. Refresh targets:

Visual diagram — is there a current architectural diagram anywhere? (Check system/public/, fedac/native/internals.md, paper figures.) If not, we may want to draw one for grants, in the AC palette.
Describe the module boundaries honestly: boot.mjs → bios.mjs → disk.mjs on web; ac-native.c → display.c → audio.c → pieces/*.mjs on native.
Data flow: where state lives (client memory, MongoDB, Redis, Digital Ocean Spaces, on-chain Tezos), and what pieces can touch what.
The code command loop — this deserves its own mini-diagram: piece → jump("terminal:claude") → PTY → Claude Code → writes to /mnt/ac-repo → next piece load picks up change.

6. Other places the AC Native story appears (sweep needed) #

7. Open questions (flag for Jeffrey) #

Boot time claim — do we lean into "~10 seconds, a full creative OS from cold USB" (which is honest + still impressive for what it does), or do we aggressively target <2s before grant decisions come in?
Framebuffer vs DRM — "bare-metal" is the artistic claim. Is DRM too technical for grant copy, or do we say "direct graphics rendering, no compositor" and leave DRM for the technical appendix?
Claude Code framing — "AI coding partner in the terminal" vs "framebuffer-native AI"? The first is honest; the second overclaims.
notepat line count — which version do we cite? (Web: 8,466; Native: 6,006.) Or total AC lines that make notepat possible?
Platter — does this exist as a concrete artifact, or should we assemble one during the refresh?

Working notes / log #

2026-04-21 14:00 — First pass on LACMA proposal: stats refreshed (371 pieces, notepat 8466), Anthropic hook added, bio front-loaded, sosoft cards replace Fig 3/4. PDF recompiled.
2026-04-21 14:30 — Audit done against current fedac/native/. Multiple claims need softening (boot time, framebuffer, Claude Code wording). File created to track long-running refresh.
2026-04-21 14:30 — LACMA-critical copy fixes applied to lacma-2026.tex:127 and LACMA-2026-APPLICATION-DRAFT.md:
- Removed "in under two seconds" boot claim (was ~7.3s actual)
- Changed "renders directly to the framebuffer" → "renders graphics through DRM without a compositor"
- Changed "reads input from raw device events" → "reads input from raw evdev streams"
- Strengthened "polyphonic" → "32-voice polyphony"
- Changed "launches Anthropic's Claude Code directly on the framebuffer" → "drops into a native terminal running Anthropic's Claude Code" (honest, still has the Anthropic punch)
- Added explicit features that were under-represented: eight waveforms, sample recording, USB + UDP MIDI
- Added "twenty other pieces ship alongside it" (previously only notepat was mentioned)
- Updated Fig 5 caption — removed "in under 2 seconds" and "any x86 machine" (now "x86_64 UEFI machines")
- Added caveat "without replacing the factory OS" — accurate to how AC Native boots from USB
- x86_64 escaped as x86_64 in tex (underscore bug)
- Final word count: 492 (was 522 pre-trim) — tightened last clause from feature-list to just "twenty other pieces ship alongside it"
2026-04-21 14:38 — Swept the three 100-word statements. Found & fixed one stale claim in Technology & Culture: "boots in two seconds" → "boots directly into a single piece of art software." Artistic Merit's parenthetical "(custom kernel, framebuffer rendering, sample-level audio synthesis)" — kept as lay-language since it's philosophical, not technical. Public Engagement statement was clean.
2026-04-21 14:44 — Converted all figures PNG → JPEG via sips (macOS). Built a 4-card composite card-gallery.jpg via a tiny LaTeX stub so the 4 cards compress into a single image slot. Final 5 JPEGs staged in grants/lacma-2026/jpegs/submit/: platform-screenshot, kidlisp-featured, card-gallery, card-berz (close-up), hardware-yoga.
2026-04-21 : — (next: video demo filming by @jeffrey; then submit)

How to use this file #

When you edit the LACMA tex, cross off items in §2.
When you touch another paper / landing page, add it in §6 and record what you changed.
When audit reality changes (we reach <2s boot, we add a new feature), update §1 — it's the source of truth.
When ambiguity bites you, add a question to §7.
Don't delete items — mark them done with ✅ or ❌ (won't do) so the history stays.