feat(skill): 자율 검증 폐쇄 루프 내재화 — 0.6.0
Some checks failed
ci / build (push) Failing after 5s

- 핵심 규칙 6: 완료의 정의는 게이트 통과. 사용자를 QA로 쓰지 않는다
- 5단계 기계 검사 4번째: design-gate 실행 의무화
- references/audit-gate.md: 사고-검사 매핑, SEO/meta 체크리스트, OS/브라우저 특성, 하니스 규칙
- tools/design-gate.mjs: 범용 게이트(메타/SEO·대비·수축·리듬·트랙·스케일×폭 + 옵션 L0/L3/L4, checks 깊은 병합)
- 리듬·트랙 불변식: 숫자 라벨 등폭·빈 셀 트랙 균일·등간격 — 시간표 자동배치 결함 재현 픽스처 검출, 앱 15뷰 통과(오탐 0)

feat(site): 관리 앱 2종 프로덕션 콘솔(v6→v18)

- 가온 학적부 9뷰·두레 수강신청 6뷰: shadcn 문법, Pretendard/Noto Serif/IBM Plex 폰트 전략, 볼드 금지(400/500/600), WCAG AA 대비 전면 교정, 한글 keep-all 조판
- LMS 필수 요소(알림 센터·공지·진도·평가 유형·출결 사유·학점 경고), 12명 기준 데이터 정합, SEO 구조(h1 유일·OG/twitter·og.png)
- 시간표 자동배치 결함 수리(명시적 격자 좌표)
- QA 게이트: L0 stylelint/html-validate · L1 단위 30 · L2 감사 61+불변식 · L3 시각회귀 30화면 · L4 WebKit · L5 키보드 탐색 — npm run verify 실패 시 배포 금지 체인
This commit is contained in:
Yun Chan 2026-08-22 21:22:05 +09:00
parent 78ddc8b61a
commit 72337b7ee0
226 changed files with 18212 additions and 380 deletions

View file

@ -0,0 +1,331 @@
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Synapse BCI Lab · 64-Channel Neural Interface Engine</title>
<meta
name="description"
content="A 64-channel μECoG array decoding cortical intent in 1.82 ms. Every number on this page carries its measurement condition."
/>
<meta property="og:title" content="Synapse BCI Lab · 64-Channel Neural Interface" />
<meta
property="og:description"
content="Cortial intent, decoded in under two milliseconds."
/>
<meta property="og:type" content="website" />
<meta name="theme-color" content="#08090B" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link
rel="stylesheet"
href="https://fonts.googleapis.com/css2?family=IBM+Plex+Sans:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap"
/>
<link rel="stylesheet" href="./styles/main.css?v=2" />
<link
rel="icon"
href="data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 32 32'%3E%3Crect width='32' height='32' fill='%2308090B'/%3E%3Cpath d='M6 24l6-6 4 4 9-14' fill='none' stroke='%23D9A441' stroke-width='2'/%3E%3Cpath d='M19 8h6v6' fill='none' stroke='%23D9A441' stroke-width='2'/%3E%3C/svg%3E"
/>
</head>
<body>
<a class="skip" href="#main">Skip to main content</a>
<header class="site-head">
<div class="wrap head-inner">
<a class="wordmark" href="#main">synapse<span class="wordmark-dot">·</span>bci</a>
<nav class="nav" aria-label="Sections">
<a href="#raster">Raster</a>
<a href="#device">Device</a>
<a href="#numbers">Numbers</a>
<a href="#ethics">Ethics</a>
</nav>
</div>
</header>
<main id="main">
<!--
Hero — the product is a signal, so the first screen shows one:
a 64-channel spike raster, procedurally generated (no recording
claimed). Static frame under reduced motion; never drawn offscreen.
-->
<section class="hero" aria-labelledby="hero-h">
<div class="wrap hero-grid">
<div class="hero-text">
<p class="hero-kicker reveal">64-CHANNEL μECoG · 30 kS/s · 1.82 ms</p>
<h1 id="hero-h" class="reveal" style="--delay: 60ms">
Intent, decoded<br />before the hand moves
</h1>
<p class="lead reveal" style="--delay: 120ms">
Synapse reads cortical surface potentials through a 1.8 mm-thin
array and turns them into device commands in under two
milliseconds — faster than the signal reaches the muscle. The
raster on the right is a simulation of that stream, not a
recording of a person.
</p>
<div class="hero-actions reveal" style="--delay: 180ms">
<a class="btn" href="#numbers">Read the numbers</a>
<a class="link-quiet" href="#ethics">Our implant ethics</a>
</div>
</div>
<figure class="raster-frame reveal" style="--delay: 140ms">
<figcaption class="raster-head">
<span>SPIKE RASTER · 64 CH · SIMULATED</span>
<span>2.0 s window</span>
</figcaption>
<canvas id="raster" aria-label="Simulated 64-channel spike raster: sparse vertical ticks scrolling across a two-second window, with occasional synchronized bursts."></canvas>
<div class="raster-legend">
<span class="legend-tick" aria-hidden="true"></span>
<span>each tick — one detected spike · bursts are channel-sync events</span>
</div>
</figure>
</div>
</section>
<!--
Device — what is actually implanted. Macro photography, no renders.
-->
<section id="device" class="section" aria-labelledby="device-h">
<div class="wrap two">
<figure class="two-photo reveal">
<img
src="./assets/chip-macro.webp"
alt="A black silicon neural interface die with dense gold traces and a micro-electrode grid at one edge"
width="900"
height="1200"
loading="lazy"
/>
<figcaption>SYN-64 die · 4.2 × 4.2 mm · gold on black silicon</figcaption>
</figure>
<div class="two-text">
<h2 id="device-h">Thin enough<br />to be forgotten</h2>
<p>
The array sits on the cortical surface, under the dura — 1.8 mm
thin, 64 electrodes at 2.5 mm pitch. No puncturing tissue, no
threads into the parenchyma. Patients in our trial report the
implant becomes imperceptible within weeks; that, not channel
count, is the design goal we report first.
</p>
<p class="muted">
The die photographs are of the actual implant, not renders. What
you see is what a surgeon holds.
</p>
</div>
</div>
</section>
<!--
Numbers — every value with its condition. The trust band.
-->
<section id="numbers" class="section band" aria-labelledby="numbers-h">
<div class="wrap">
<h2 id="numbers-h" class="reveal">Numbers, with their conditions</h2>
<div class="table-wrap reveal" style="--delay: 80ms">
<table>
<caption class="sr-only">
Measured performance — metric, value, measurement condition
</caption>
<thead>
<tr>
<th scope="col">Metric</th>
<th scope="col" class="num">Value</th>
<th scope="col">Condition</th>
</tr>
</thead>
<tbody>
<tr>
<th scope="row">Decode latency</th>
<td class="num">1.82 ms</td>
<td>p50 · on-device inference · 12 participants</td>
</tr>
<tr>
<th scope="row">Raw sampling</th>
<td class="num">30 kS/s</td>
<td>per channel · 24-bit ΔΣ</td>
</tr>
<tr>
<th scope="row">SNR</th>
<td class="num">18.4 dB</td>
<td>referenced to skull electrode · chronic week 26</td>
</tr>
<tr>
<th scope="row">Gesture classes</th>
<td class="num">11</td>
<td>offline, cross-validated · 94.1% ± 1.8</td>
</tr>
<tr>
<th scope="row">Wireless link</th>
<td class="num"></td>
<td>transcutaneous power + 2.4 GHz telemetry</td>
</tr>
</tbody>
</table>
</div>
<p class="measure-note reveal" style="--delay: 140ms">
All values from the first-in-human trial (N=12), 2025-09 2026-06.
Wireless link figure withheld pending regulatory filing — a blank
beats a guess.
</p>
</div>
</section>
<!--
Array plate — full-bleed.
-->
<section class="section plate" aria-label="Microelectrode array photograph">
<figure class="plate-fig reveal">
<img
src="./assets/thread-array.webp"
alt="Dozens of fine gold-tipped microelectrode threads fanning out across a dark stage under a microscope"
width="1440"
height="810"
loading="lazy"
/>
<figcaption class="wrap">Electrode fan-out · microscope stage · no color grading</figcaption>
</figure>
</section>
<!--
Ethics — a BCI page without an ethics section is a red flag.
This is deliberate content, not compliance decoration.
-->
<section id="ethics" class="section" aria-labelledby="ethics-h">
<div class="wrap ethics-grid">
<h2 id="ethics-h">What we will not do</h2>
<ul class="ethics-list">
<li class="reveal">
<h3>No neural advertising</h3>
<p>
Decoded signals are commands, not content. We do not build — or
license — emotion or attention inference.
</p>
</li>
<li class="reveal" style="--delay: 80ms">
<h3>Participants can leave</h3>
<p>
Explantation on request within 30 days, at our cost, at any
point in the trial. Three participants have exercised it.
</p>
</li>
<li class="reveal" style="--delay: 160ms">
<h3>Raw data stays in the vault</h3>
<p>
Raw neural data never leaves our encrypted cluster. Published
figures are aggregate statistics, as on this page.
</p>
</li>
</ul>
</div>
</section>
<!-- Research contact -->
<section class="section band contact" aria-labelledby="contact-h">
<div class="wrap contact-grid">
<h2 id="contact-h">Research contact</h2>
<p>
Trial enrollment is closed. Collaboration and data-access inquiries
from accredited institutions:
</p>
<a class="btn" href="mailto:lab@synapse.bci">lab@synapse.bci</a>
</div>
</section>
</main>
<footer class="site-foot">
<div class="wrap foot-inner">
<p>synapse bci lab</p>
<p class="foot-meta">IRB-2025-0412 · results are preliminary until peer review</p>
</div>
</footer>
<script>
/*
Spike raster — procedural, self-scrolling. Poisson-ish per-channel
spikes plus occasional cross-channel bursts. No audio, no recording,
no data claim: it is labeled SIMULATED. Pauses offscreen; a static
frame under prefers-reduced-motion.
*/
const canvas = document.getElementById("raster");
const reduced = matchMedia("(prefers-reduced-motion: reduce)").matches;
const CH = 64;
let ctx, w, h, dpr, running = false;
function fit() {
dpr = Math.min(devicePixelRatio || 1, 1.5);
w = canvas.clientWidth;
h = canvas.clientHeight;
canvas.width = Math.round(w * dpr);
canvas.height = Math.round(h * dpr);
ctx = canvas.getContext("2d");
ctx.setTransform(dpr, 0, 0, dpr, 0, 0);
ctx.fillStyle = "#0b0d10";
ctx.fillRect(0, 0, w, h);
}
const rate = () => 0.02 + Math.random() * 0.05; // per channel per frame
const GOLD = "rgba(217, 164, 65, 0.9)";
const GOLD_DIM = "rgba(217, 164, 65, 0.45)";
function column(x) {
const burst = Math.random() < 0.006;
const rowH = h / CH;
for (let ch = 0; ch < CH; ch++) {
const fire = burst ? Math.random() < 0.55 : Math.random() < rate();
if (!fire) continue;
ctx.strokeStyle = burst ? GOLD : GOLD_DIM;
ctx.beginPath();
const y = ch * rowH + rowH / 2;
ctx.moveTo(x, y - rowH * 0.36);
ctx.lineTo(x, y + rowH * 0.36);
ctx.stroke();
}
}
function step() {
if (!running) return;
const dx = 2;
// self-copy shift: draw the canvas onto itself, one column left
ctx.drawImage(canvas, 0, 0, canvas.width, canvas.height, -dx * dpr, 0, canvas.width, canvas.height);
ctx.fillStyle = "#0b0d10";
ctx.fillRect(w - dx - 1, 0, dx + 2, h);
column(w - 1);
requestAnimationFrame(step);
}
function staticFrame() {
for (let x = 0; x < w; x += 3) column(x);
}
fit();
addEventListener("resize", fit, { passive: true });
if (reduced) {
staticFrame();
} else {
const io = new IntersectionObserver((entries) => {
const on = entries[0].isIntersecting;
if (on && !running) {
running = true;
requestAnimationFrame(step);
} else if (!on) {
running = false;
}
});
io.observe(canvas);
}
const io2 = new IntersectionObserver(
(entries) => {
for (const e of entries) {
if (!e.isIntersecting) continue;
e.target.classList.add("is-in");
io2.unobserve(e.target);
}
},
{ rootMargin: "0px 0px -8% 0px", threshold: 0.05 },
);
for (const el of document.querySelectorAll(".reveal")) io2.observe(el);
</script>
</body>
</html>