evalstate's picture
download
raw
61.4 kB
<!-- Copyright 2026 Anthropic PBC · SPDX-License-Identifier: Apache-2.0 -->
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Rendering checker pipeline — module explainer</title>
<style data-birch-system>
/*
Birch HTML Design System
Opinionated single-option stylesheet for reliable LLM-generated artifacts.
Model:
1. Design tokens as CSS variables.
2. Layout primitives as classes.
3. Semantic components as classes with data-* variants.
*/
/* ==================== 1. TOKENS ==================== */
:root {
color-scheme: light;
/* Palette */
--color-ivory: #FAF9F5;
--color-slate: #141413;
--color-clay: #D97757;
--color-clay-dark: #B85C3E;
--color-oat: #E3DACC;
--color-olive: #788C5D;
--color-rust: #B04A3F;
--color-sky: #6A8CAF;
--color-white: #FFFFFF;
--color-gray-50: #F7F5EE;
--color-gray-100: #F0EEE6;
--color-gray-150: #E8E4DA;
--color-gray-200: #DED9CD;
--color-gray-300: #D1CFC5;
--color-gray-500: #87867F;
--color-gray-700: #3D3D3A;
--color-gray-800: #242421;
/* Semantic colors */
--bg: var(--color-ivory);
--surface: var(--color-white);
--surface-tint: var(--color-gray-100);
--surface-warm: var(--color-oat);
--text: var(--color-slate);
--text-muted: var(--color-gray-500);
--text-soft: var(--color-gray-700);
--muted: var(--text-muted);
--border-color: var(--color-gray-300);
--accent: var(--color-clay);
--accent-strong: var(--color-clay-dark);
--success: var(--color-olive);
--warning: #C78E3F;
--danger: var(--color-rust);
--info: var(--color-sky);
/* Typography */
--font-serif: ui-serif, Georgia, "Times New Roman", serif;
--font-sans: system-ui, -apple-system, "Segoe UI", Roboto, sans-serif;
--font-mono: ui-monospace, "SF Mono", Menlo, Consolas, monospace;
--text-xs: 11px;
--text-sm: 12px;
--text-md: 14px;
--text-base: 15px;
--text-lg: 17px;
--text-xl: 21px;
--text-2xl: 26px;
--text-3xl: 36px;
--text-4xl: 44px;
--leading-tight: 1.15;
--leading-title: 1.25;
--leading-body: 1.55;
--leading-loose: 1.7;
/* Spacing */
--space-1: 4px;
--space-2: 8px;
--space-3: 12px;
--space-4: 16px;
--space-5: 24px;
--space-6: 32px;
--space-7: 48px;
--space-8: 64px;
--space-9: 96px;
/* Shape */
--radius-xs: 4px;
--radius-sm: 8px;
--radius-md: 12px;
--radius-lg: 16px;
--radius-xl: 20px;
--radius-pill: 999px;
--border: 1.5px solid var(--border-color);
--border-thin: 1px solid var(--border-color);
--shadow-sm: 0 1px 2px rgba(20, 20, 19, 0.06);
--shadow-md: 0 4px 14px rgba(20, 20, 19, 0.08);
--shadow-lg: 0 12px 28px rgba(20, 20, 19, 0.12);
/* Motion */
--ease: cubic-bezier(0.2, 0, 0, 1);
--ease-spring: cubic-bezier(0.34, 1.56, 0.64, 1);
--dur-fast: 120ms;
--dur: 180ms;
--hover-y: -2px;
/* Page sizing */
--page-max: 1100px;
--page-pad: 28px;
--page-pad-block: 48px;
/* Legacy aliases used by source examples */
--ivory: var(--color-ivory);
--slate: var(--color-slate);
--clay: var(--color-clay);
--clay-d: var(--color-clay-dark);
--oat: var(--color-oat);
--olive: var(--color-olive);
--rust: var(--color-rust);
--sky: var(--color-sky);
--white: var(--color-white);
--gray-50: var(--color-gray-50);
--gray-100: var(--color-gray-100);
--gray-150: var(--color-gray-150);
--gray-200: var(--color-gray-200);
--gray-300: var(--color-gray-300);
--gray-500: var(--color-gray-500);
--gray-700: var(--color-gray-700);
--gray-800: var(--color-gray-800);
--serif: var(--font-serif);
--sans: var(--font-sans);
--mono: var(--font-mono);
}
/* ==================== BASE ==================== */
*,
*::before,
*::after {
box-sizing: border-box;
}
html {
background: var(--bg);
color: var(--text);
font-family: var(--font-sans);
-webkit-text-size-adjust: 100%;
}
body {
margin: 0;
background: var(--bg);
color: var(--text);
font-family: var(--font-sans);
font-size: var(--text-base);
line-height: var(--leading-body);
-webkit-font-smoothing: antialiased;
text-rendering: optimizeLegibility;
}
img,
svg {
max-width: 100%;
}
a {
color: inherit;
text-decoration-color: rgba(217, 119, 87, 0.45);
text-underline-offset: 3px;
}
a:hover {
color: var(--accent);
}
h1,
h2,
h3,
p {
margin-block: 0;
}
h1,
h2,
h3 {
font-family: var(--font-serif);
font-weight: 500;
letter-spacing: -0.01em;
line-height: var(--leading-title);
}
h1 {
font-size: clamp(32px, 5vw, var(--text-4xl));
}
h2 {
font-size: var(--text-2xl);
}
h3 {
font-size: var(--text-xl);
}
code,
kbd,
pre {
font-family: var(--font-mono);
}
code {
font-size: 0.92em;
white-space: normal;
word-break: normal;
}
h1 code,
h2 code,
h3 code {
overflow-wrap: anywhere;
word-break: break-word;
}
button,
input,
textarea,
select {
font: inherit;
}
button {
cursor: pointer;
}
::selection {
background: rgba(217, 119, 87, 0.24);
}
/* ==================== 2. LAYOUT PRIMITIVES ==================== */
.page {
width: min(100% - (var(--page-pad) * 2), var(--page-max));
margin-inline: auto;
padding-block: var(--page-pad-block) var(--space-9);
}
.page[data-size="narrow"] {
--page-max: 820px;
}
.page[data-size="wide"] {
--page-max: 1280px;
}
.page[data-size="full"] {
--page-max: 1440px;
}
.section {
margin-block: var(--section-gap, var(--space-5));
}
.section[data-gap="lg"] {
--section-gap: var(--space-6);
}
.section[data-gap="xl"] {
--section-gap: var(--space-7);
}
.section:first-child {
margin-block-start: 0;
}
.stack > .section {
margin-block: 0;
}
.rule {
border: 0;
border-top: var(--border-thin);
margin: var(--space-3) 0 var(--space-5);
}
.stack {
display: flex;
flex-direction: column;
gap: var(--stack-gap, var(--space-4));
}
.stack[data-gap="xs"] { --stack-gap: var(--space-2); }
.stack[data-gap="sm"] { --stack-gap: var(--space-3); }
.stack[data-gap="md"] { --stack-gap: var(--space-4); }
.stack[data-gap="lg"] { --stack-gap: var(--space-6); }
.stack[data-gap="xl"] { --stack-gap: var(--space-8); }
.stack > :is(.button, .btn, .chip, .badge) {
align-self: flex-start;
}
.cluster {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: var(--cluster-gap, var(--space-3));
}
.cluster[data-align="start"] { align-items: flex-start; }
.cluster[data-align="end"] { align-items: flex-end; }
.cluster[data-justify="between"] { justify-content: space-between; }
.cluster[data-justify="end"] { justify-content: flex-end; }
.grid {
display: grid;
gap: var(--grid-gap, var(--space-5));
}
.grid[data-cols="2"] {
grid-template-columns: repeat(2, minmax(0, 1fr));
}
.grid[data-cols="3"] {
grid-template-columns: repeat(3, minmax(0, 1fr));
}
.grid[data-cols="4"] {
grid-template-columns: repeat(4, minmax(0, 1fr));
}
.auto-grid {
display: grid;
grid-template-columns: repeat(auto-fit, minmax(min(100%, var(--grid-min, 240px)), 1fr));
gap: var(--grid-gap, var(--space-5));
}
.auto-grid > * {
min-width: 0;
}
.split {
display: grid;
grid-template-columns: minmax(0, 1fr) minmax(280px, var(--split-side, 360px));
gap: var(--split-gap, var(--space-6));
align-items: start;
}
.split > * {
min-width: 0;
}
.section-head {
min-width: 0;
}
.section-rail {
display: grid;
grid-template-columns: minmax(0, 1fr) minmax(280px, var(--rail-width, 360px));
column-gap: var(--rail-gap, var(--space-6));
row-gap: var(--space-5);
align-items: start;
}
.section-rail > .section-head {
grid-column: 1 / -1;
}
.section-rail > .reference-panel,
.section-rail > aside {
align-self: start;
}
.sidebar-layout {
display: grid;
grid-template-columns: var(--sidebar-width, 240px) minmax(0, 1fr);
gap: var(--sidebar-gap, var(--space-7));
align-items: start;
}
.sidebar-layout > :first-child,
.sticky {
position: sticky;
top: var(--sticky-top, var(--space-5));
}
.bleed {
margin-inline: calc(var(--page-pad) * -1);
}
.scroll-x {
overflow-x: auto;
overscroll-behavior-inline: contain;
}
.center {
display: grid;
place-items: center;
}
.spacer {
flex: 1;
}
@media (max-width: 860px) {
:root {
--page-pad: 20px;
--page-pad-block: 40px;
}
.grid[data-cols],
.split,
.section-rail,
.sidebar-layout {
grid-template-columns: 1fr;
}
.sidebar-layout > :first-child,
.sticky {
position: static;
}
}
/* ==================== TYPOGRAPHY UTILITIES ==================== */
.eyebrow {
display: inline-flex;
align-items: center;
gap: var(--space-2);
color: var(--accent);
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 700;
letter-spacing: 0.08em;
text-transform: uppercase;
}
.eyebrow::before {
content: "";
width: 24px;
height: 1.5px;
background: currentColor;
}
.lede {
max-width: 720px;
color: var(--text-soft);
font-size: var(--text-lg);
line-height: 1.6;
}
.muted {
color: var(--text-muted);
}
.caption {
color: var(--text-muted);
font-family: var(--font-mono);
font-size: var(--text-sm);
}
.stat-value {
display: block;
color: var(--text);
font-family: var(--font-serif);
font-size: clamp(30px, 3.2vw, 42px);
font-variant-numeric: tabular-nums;
font-weight: 550;
letter-spacing: -0.025em;
line-height: 1;
}
.stat-card {
gap: var(--space-2);
}
.stat-card .caption {
letter-spacing: 0.08em;
text-transform: uppercase;
}
.stat-card .muted {
max-width: 34ch;
font-size: var(--text-sm);
}
.mono {
font-family: var(--font-mono);
}
/* ==================== 3. SEMANTIC COMPONENTS ==================== */
.panel,
.card {
background: var(--surface);
border: var(--border);
border-radius: var(--radius-md);
}
.panel {
padding: var(--panel-pad, var(--space-5));
}
.card {
padding: var(--card-pad, var(--space-5));
transition:
transform var(--dur-fast) var(--ease),
border-color var(--dur-fast) var(--ease),
box-shadow var(--dur-fast) var(--ease),
background-color var(--dur-fast) var(--ease);
}
.card[data-hover="lift"]:hover,
.card[data-clickable]:hover {
transform: translateY(var(--hover-y));
border-color: rgba(217, 119, 87, 0.55);
box-shadow: var(--shadow-md);
}
.card[data-clickable] {
cursor: pointer;
}
.card[data-variant="flat"] {
background: transparent;
border-color: transparent;
box-shadow: none;
}
.card[data-variant="outlined"] {
background: var(--surface);
border: var(--border);
}
.card[data-variant="elevated"] {
border-color: transparent;
box-shadow: var(--shadow-md);
}
.card[data-variant="filled"],
.card[data-tone="oat"] {
background: var(--surface-warm);
}
.card[data-tone],
.panel[data-tone],
.callout[data-tone] {
background: var(--surface);
}
.card[data-tone="danger"],
.panel[data-tone="danger"],
.callout[data-tone="danger"] {
border-color: rgba(176, 74, 63, 0.36);
}
.card[data-tone="warning"],
.panel[data-tone="warning"],
.callout[data-tone="warning"] {
border-color: rgba(199, 142, 63, 0.38);
}
.card[data-tone="success"],
.panel[data-tone="success"],
.callout[data-tone="success"] {
border-color: rgba(120, 140, 93, 0.38);
}
.card[data-tone="info"],
.panel[data-tone="info"],
.callout[data-tone="info"] {
border-color: rgba(106, 140, 175, 0.36);
}
.card[data-accent] {
position: relative;
overflow: hidden;
}
.card[data-accent]::before {
content: "";
position: absolute;
inset: 0 auto 0 0;
width: 5px;
background: var(--accent);
}
.card-head {
display: flex;
align-items: center;
gap: var(--space-3);
min-width: 0;
max-width: 100%;
}
.card-head > :first-child {
flex: 0 0 auto;
}
.card-titles {
flex: 1 1 0;
min-width: 0;
max-width: 100%;
}
.card-title {
max-width: 100%;
margin: 0;
font-family: var(--font-serif);
font-size: var(--text-xl);
font-weight: 550;
line-height: var(--leading-title);
overflow-wrap: anywhere;
}
.card-sub {
margin: 2px 0 0;
color: var(--text-muted);
font-size: var(--text-sm);
overflow-wrap: anywhere;
}
.toolbar {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: var(--space-4);
padding: var(--space-4) var(--space-5);
background: var(--surface);
border: var(--border);
border-radius: var(--radius-md);
}
.toolbar[data-sticky] {
position: sticky;
top: 0;
z-index: 10;
}
.chip,
.badge {
display: inline-flex;
align-items: center;
justify-content: center;
gap: 6px;
border-radius: var(--radius-pill);
white-space: nowrap;
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 650;
line-height: 1;
}
.chip {
min-height: 24px;
padding: 0 10px;
color: var(--text-soft);
background: var(--surface-tint);
border: 1px solid transparent;
}
.badge {
min-height: 22px;
padding: 0 9px;
color: var(--text-muted);
background: var(--surface);
border: var(--border-thin);
text-transform: uppercase;
letter-spacing: 0.06em;
}
.chip[data-tone="clay"],
.badge[data-tone="clay"] {
color: var(--accent);
background: rgba(217, 119, 87, 0.14);
}
.chip[data-tone="olive"],
.badge[data-tone="olive"],
.chip[data-tone="success"],
.badge[data-tone="success"] {
color: var(--success);
background: rgba(120, 140, 93, 0.15);
}
.chip[data-tone="warning"],
.badge[data-tone="warning"] {
color: var(--warning);
background: rgba(199, 142, 63, 0.16);
}
.chip[data-tone="rust"],
.badge[data-tone="rust"],
.chip[data-tone="danger"],
.badge[data-tone="danger"] {
color: var(--danger);
background: rgba(176, 74, 63, 0.13);
}
.chip[data-tone="sky"],
.badge[data-tone="sky"],
.chip[data-tone="info"],
.badge[data-tone="info"] {
color: var(--info);
background: rgba(106, 140, 175, 0.14);
}
.finding {
border-left: 6px solid var(--border-color);
}
.finding[data-severity="blocker"],
.finding[data-severity="critical"],
.finding[data-severity="danger"],
.finding[data-tone="danger"] {
border-left-color: var(--danger);
}
.finding[data-severity="high"],
.finding[data-severity="warning"],
.finding[data-tone="warning"] {
border-left-color: var(--warning);
}
.finding[data-severity="medium"],
.finding[data-severity="info"],
.finding[data-tone="info"] {
border-left-color: var(--info);
}
.finding[data-severity="low"],
.finding[data-severity="success"],
.finding[data-tone="success"] {
border-left-color: var(--success);
}
.button,
.btn {
appearance: none;
display: inline-flex;
align-items: center;
justify-content: center;
min-height: 36px;
padding: 0 14px;
border: 1.5px solid transparent;
border-radius: var(--radius-sm);
font-family: var(--font-sans);
font-size: var(--text-md);
font-weight: 650;
line-height: 1;
text-decoration: none;
transition:
transform var(--dur-fast) var(--ease),
background-color var(--dur-fast) var(--ease),
border-color var(--dur-fast) var(--ease),
box-shadow var(--dur-fast) var(--ease);
}
.button:hover,
.btn:hover {
transform: translateY(-1px);
}
.button[data-variant="primary"],
.btn-primary {
color: var(--color-white);
background: var(--color-slate);
}
.button[data-variant="primary"]:hover,
.btn-primary:hover {
background: var(--color-gray-700);
}
.button[data-variant="secondary"],
.btn-secondary {
color: var(--text);
background: var(--surface-tint);
border-color: var(--border-color);
}
.button[data-variant="ghost"],
.btn-ghost {
color: var(--text-soft);
background: transparent;
}
.button[data-variant="ghost"]:hover,
.btn-ghost:hover {
color: var(--text);
background: var(--surface-tint);
}
.button[data-variant="danger"],
.btn-danger {
color: var(--color-white);
background: var(--danger);
}
.input {
width: 100%;
min-height: 38px;
padding: 0 12px;
color: var(--text);
background: var(--surface);
border: var(--border);
border-radius: var(--radius-sm);
outline: none;
}
.input:focus {
border-color: var(--accent);
box-shadow: 0 0 0 3px rgba(217, 119, 87, 0.16);
}
.avatar {
display: inline-flex;
align-items: center;
justify-content: center;
width: var(--avatar-size, 38px);
height: var(--avatar-size, 38px);
border-radius: 50%;
color: var(--text-soft);
background: var(--surface-warm);
font-family: var(--font-mono);
font-size: var(--text-sm);
font-weight: 700;
letter-spacing: 0.02em;
flex: 0 0 auto;
}
.code-block {
margin: 0;
padding: var(--space-4) var(--space-5);
overflow-x: auto;
color: #E8E6DC;
background: var(--color-slate);
border-radius: var(--radius-md);
font-family: var(--font-mono);
font-size: var(--text-sm);
line-height: var(--leading-loose);
}
.code-block[data-kind="command"],
.code-block[data-wrap="true"],
.command-block {
max-width: 100%;
min-width: 0;
box-sizing: border-box;
white-space: pre-wrap;
overflow-wrap: anywhere;
word-break: break-word;
overflow-x: hidden;
}
.code-block[data-wrap="true"] code,
.code-block[data-kind="command"] code {
display: block;
max-width: 100%;
white-space: pre-wrap;
overflow-wrap: anywhere;
word-break: break-word;
}
.copyable {
position: relative;
}
.copyable > .code-block {
padding-inline-end: 88px;
}
.copy-button {
appearance: none;
position: absolute;
inset-block-start: var(--space-2);
inset-inline-end: var(--space-2);
display: inline-flex;
align-items: center;
justify-content: center;
min-height: 26px;
padding: 0 9px;
border: 1px solid rgba(250, 249, 245, 0.18);
border-radius: var(--radius-sm);
color: var(--color-gray-100);
background: rgba(255, 255, 255, 0.08);
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 700;
letter-spacing: 0.04em;
text-transform: uppercase;
transition:
background-color var(--dur-fast) var(--ease),
border-color var(--dur-fast) var(--ease),
transform var(--dur-fast) var(--ease);
}
.copy-button:hover {
transform: translateY(-1px);
border-color: rgba(250, 249, 245, 0.35);
background: rgba(255, 255, 255, 0.14);
}
.copy-button[data-copied="true"] {
color: var(--color-white);
background: rgba(120, 140, 93, 0.55);
border-color: rgba(120, 140, 93, 0.75);
}
.reference-panel {
--panel-pad: var(--space-5);
}
.callout {
padding: var(--space-5);
color: var(--text-soft);
background: var(--surface-tint);
border: var(--border);
border-inline-start: 4px solid var(--accent);
border-radius: var(--radius-md);
}
.callout[data-tone="success"] {
border-inline-start-color: var(--success);
background: rgba(120, 140, 93, 0.10);
}
.callout[data-tone="warning"] {
border-inline-start-color: var(--warning);
background: rgba(199, 142, 63, 0.11);
}
.callout[data-tone="danger"] {
border-inline-start-color: var(--danger);
background: rgba(176, 74, 63, 0.10);
}
.callout[data-tone="info"] {
border-inline-start-color: var(--info);
background: rgba(106, 140, 175, 0.11);
}
.callout-label {
display: block;
margin-bottom: var(--space-2);
color: var(--text-muted);
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 700;
letter-spacing: 0.09em;
text-transform: uppercase;
}
.callout > :last-child {
margin-bottom: 0;
}
.checklist {
list-style: none;
margin: 0;
padding: 0;
display: grid;
gap: var(--space-3);
}
.checklist li {
display: block !important;
position: relative !important;
min-width: 0;
padding-inline-start: calc(18px + var(--space-3));
overflow-wrap: anywhere;
}
.checklist li::before {
content: "✓";
position: absolute !important;
inset-block-start: 2px;
inset-inline-start: 0;
display: inline-flex;
align-items: center;
justify-content: center;
width: 18px;
height: 18px;
border-radius: 50%;
color: var(--success);
background: rgba(120, 140, 93, 0.12);
font-family: var(--font-mono);
font-size: 12px;
font-weight: 700;
line-height: 1;
}
.checklist code {
white-space: normal;
overflow-wrap: anywhere;
word-break: break-word;
}
.plain-list {
margin: 0;
padding-inline-start: 1.15em;
color: var(--text-soft);
}
.plain-list li {
overflow-wrap: anywhere;
}
.plain-list li + li {
margin-top: var(--space-2);
}
.plain-list li::marker {
color: var(--text-muted);
}
.insight-list,
.takeaway-list {
list-style: none;
margin: 0;
padding: 0;
display: grid;
gap: var(--space-3);
}
.insight-list li,
.takeaway-list li {
position: relative;
display: block;
padding-inline-start: calc(10px + var(--space-3));
color: var(--text-soft);
}
.insight-list li::before,
.takeaway-list li::before {
content: "";
position: absolute;
inset-block-start: 0.65em;
inset-inline-start: 0;
width: 7px;
height: 7px;
border-radius: 50%;
background: var(--accent);
}
.insight-list li > *,
.takeaway-list li > * {
min-width: 0;
}
.insight-list[data-tone="success"] li::before,
.takeaway-list[data-tone="success"] li::before {
background: var(--success);
}
.insight-list[data-tone="info"] li::before,
.takeaway-list[data-tone="info"] li::before {
background: var(--info);
}
.metric-list {
--metric-label: 132px;
--metric-value: 64px;
display: grid;
gap: var(--space-3);
}
.metric-row {
display: grid;
grid-template-columns: var(--metric-label) minmax(120px, 1fr) var(--metric-value);
gap: var(--space-3);
align-items: center;
}
.metric-row > :first-child {
min-width: 0;
}
.metric-row > :last-child {
justify-self: end;
font-family: var(--font-mono);
font-size: var(--text-sm);
}
.meter {
height: 9px;
overflow: hidden;
border-radius: var(--radius-pill);
background: var(--surface-tint);
}
.meter span {
display: block;
width: var(--value);
height: 100%;
background: var(--tone, var(--accent));
}
.flow-list {
list-style: none;
margin: 0;
padding: 0;
display: grid;
gap: var(--space-3);
}
.flow-step {
position: relative;
display: grid;
grid-template-columns: 42px minmax(0, 1fr);
gap: var(--space-3);
align-items: start;
padding: var(--space-4);
background: var(--surface);
border: var(--border);
border-radius: var(--radius-md);
}
.flow-step > .flow-num {
grid-column: 1;
}
.flow-step > :not(.flow-num) {
grid-column: 2;
min-width: 0;
}
.flow-step::after {
content: "";
position: absolute;
inset-block-start: calc(100% + 1px);
inset-inline-start: 36px;
width: 1.5px;
height: var(--space-3);
background: var(--border-color);
}
.flow-step:last-child::after {
display: none;
}
.flow-num {
display: inline-flex;
align-items: center;
justify-content: center;
width: 34px;
height: 34px;
border-radius: var(--radius-pill);
color: var(--text);
background: var(--surface-warm);
box-shadow: inset 0 0 0 1px rgba(20, 20, 19, 0.10);
font-family: var(--font-mono);
font-size: var(--text-sm);
font-weight: 700;
}
.flow-title {
margin: 0 0 var(--space-1);
font-family: var(--font-serif);
font-size: var(--text-xl);
font-weight: 550;
line-height: var(--leading-title);
}
.flow-detail {
margin: 0;
color: var(--text-soft);
}
.flow-branch {
display: grid;
grid-template-columns: repeat(2, minmax(0, 1fr));
gap: var(--space-3);
}
.flow-branch .flow-step::after {
display: none;
}
.flow-step[data-tone="success"] {
border-color: rgba(120, 140, 93, 0.38);
background: rgba(120, 140, 93, 0.08);
}
.flow-step[data-tone="warning"] {
border-color: rgba(199, 142, 63, 0.42);
background: rgba(199, 142, 63, 0.09);
}
@media (max-width: 620px) {
.metric-row {
grid-template-columns: 1fr max-content;
}
.metric-row .meter {
grid-column: 1 / -1;
order: 3;
}
.flow-branch {
grid-template-columns: 1fr;
}
}
/* ---------- Charts and numeric data ---------- */
.chart-panel {
max-width: 100%;
min-width: 0;
overflow: hidden;
}
.chart-panel > * {
min-width: 0;
}
.chart-panel svg,
svg.chart-svg {
display: block;
width: 100%;
height: auto;
}
.chart-caption {
color: var(--text-muted);
font-size: var(--text-sm);
}
.numeric-table-wrap {
overflow-x: auto;
background: var(--surface);
border: var(--border);
border-radius: var(--radius-md);
}
.numeric-table {
width: 100%;
min-width: 720px;
border-collapse: collapse;
}
.numeric-table th,
.numeric-table td {
padding: var(--space-3) var(--space-4);
border-bottom: 1px solid var(--color-gray-200);
vertical-align: middle;
}
.numeric-table th {
color: var(--text-muted);
background: var(--surface-tint);
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 700;
letter-spacing: 0.08em;
text-align: left;
text-transform: uppercase;
}
.numeric-table tr:last-child td {
border-bottom: 0;
}
.numeric-table .metric,
.numeric-table .num {
text-align: right;
font-variant-numeric: tabular-nums;
white-space: nowrap;
}
.numeric-table .entity,
.numeric-table .label-cell {
font-weight: 650;
}
.numeric-table .note {
color: var(--text-soft);
}
.numeric-table code {
white-space: normal;
overflow-wrap: anywhere;
word-break: break-word;
}
.numeric-table .subtle {
display: block;
margin-top: 2px;
color: var(--text-muted);
font-family: var(--font-mono);
font-size: var(--text-xs);
font-weight: 400;
}
@media (max-width: 620px) {
.numeric-table {
min-width: 0;
}
.numeric-table thead {
display: none;
}
.numeric-table tbody,
.numeric-table tr,
.numeric-table td {
display: block;
}
.numeric-table tr {
padding: var(--space-3) var(--space-4);
border-bottom: 1px solid var(--color-gray-200);
}
.numeric-table tr:last-child {
border-bottom: 0;
}
.numeric-table td {
padding: var(--space-1) 0;
border-bottom: 0;
}
.numeric-table td[data-label] {
display: flex;
justify-content: space-between;
gap: var(--space-4);
}
.numeric-table td[data-label]::before {
content: attr(data-label);
color: var(--text-muted);
font-family: var(--font-mono);
font-size: var(--text-xs);
text-transform: uppercase;
}
}
/* ---------- Diff ---------- */
.diff {
overflow-x: auto;
color: #E8E6DC;
background: var(--color-slate);
border-radius: var(--radius-md);
font-family: var(--font-mono);
font-size: 12.5px;
line-height: var(--leading-loose);
}
.diff-row {
display: grid;
grid-template-columns: 48px 18px minmax(max-content, 1fr);
align-items: baseline;
min-width: max-content;
padding-inline-end: var(--space-4);
white-space: pre;
}
.diff-row .ln {
padding-inline-end: 14px;
color: var(--color-gray-500);
text-align: right;
user-select: none;
}
.diff-row .mark {
color: var(--color-gray-500);
text-align: center;
}
.diff-row .code {
color: #E8E6DC;
}
.diff-row.ctx .code,
.diff-row[data-kind="ctx"] .code {
color: #B8B6AC;
}
.diff-row.add,
.diff-row[data-kind="add"] {
background: rgba(120, 140, 93, 0.24);
box-shadow: inset 3px 0 0 rgba(120, 140, 93, 0.78);
}
.diff-row.add .mark,
.diff-row[data-kind="add"] .mark {
color: #B9D394;
font-weight: 800;
}
.diff-row.del,
.diff-row[data-kind="del"] {
background: rgba(176, 74, 63, 0.24);
box-shadow: inset 3px 0 0 rgba(176, 74, 63, 0.82);
}
.diff-row.del .mark,
.diff-row[data-kind="del"] .mark {
color: #F19A8D;
font-weight: 800;
}
.diff-row.hunk,
.diff-row[data-kind="hunk"] {
background: rgba(255, 255, 255, 0.045);
}
.diff-row.hunk .code,
.diff-row[data-kind="hunk"] .code {
color: var(--color-gray-500);
}
.diff[data-wrap="true"] {
overflow-x: hidden;
}
.diff[data-wrap="true"] .diff-row {
grid-template-columns: 42px 18px minmax(0, 1fr);
min-width: 0;
white-space: normal;
}
.diff[data-wrap="true"] .code {
white-space: pre-wrap;
overflow-wrap: anywhere;
}
/* ---------- Timeline ---------- */
.timeline {
position: relative;
display: grid;
gap: 0;
padding-inline-start: var(--space-6);
}
.timeline:is(ol, ul) {
margin: 0;
list-style: none;
}
.timeline::before {
content: "";
position: absolute;
inset-block: 4px 4px;
inset-inline-start: 9px;
width: 1.5px;
background: var(--border-color);
}
.timeline-item,
.tl-entry {
position: relative;
display: grid;
grid-template-columns: minmax(84px, max-content) minmax(0, 1fr);
gap: var(--space-4);
padding-block: 0 var(--space-5);
}
.timeline-item:last-child,
.tl-entry:last-child {
padding-block-end: 0;
}
.timeline-item::before,
.tl-entry::before {
content: "";
position: absolute;
inset-block-start: 4px;
inset-inline-start: calc(4.25px - var(--space-6));
width: 11px;
height: 11px;
border: 2px solid var(--surface);
border-radius: 50%;
background: var(--accent);
box-shadow: 0 0 0 1.5px var(--accent);
}
.timeline-item:not(:has(> .timeline-time, > .tl-time)),
.tl-entry:not(:has(> .timeline-time, > .tl-time)) {
display: block;
}
.timeline-time,
.tl-time {
color: var(--text-muted);
font-family: var(--font-mono);
font-size: var(--text-sm);
white-space: nowrap;
}
.timeline-body,
.tl-body {
min-width: 0;
color: var(--text-soft);
}
.timeline-body strong,
.tl-body strong {
color: var(--text);
}
@media (max-width: 620px) {
.timeline-item,
.tl-entry {
grid-template-columns: 1fr;
gap: var(--space-1);
}
}
/* ---------- Flowchart ---------- */
.flowchart {
display: block;
width: 100%;
height: auto;
}
svg.flow,
svg.flowchart {
display: block;
width: 100%;
height: auto;
}
.flow text,
.flowchart text {
fill: var(--text);
font-family: var(--font-sans);
font-size: 12px;
}
.flow .sub,
.flowchart .sub {
fill: var(--text-muted);
font-size: 10px;
}
.flow-edge,
.edge {
fill: none;
stroke: var(--text-muted);
stroke-width: 1.5;
}
.flow-edge[data-kind="yes"],
.edge.yes {
stroke: var(--success);
}
.flow-edge[data-kind="no"],
.edge.no {
stroke: var(--danger);
stroke-dasharray: 4 4;
}
/* SVG flow nodes */
.flow-node,
.flow .node {
cursor: pointer;
transition: transform var(--dur-fast) var(--ease);
}
.flow-node:hover,
.flow .node:hover {
transform: translateY(-1px);
}
.flow-node rect,
.flow .node rect {
fill: var(--surface);
stroke: var(--border-color);
stroke-width: 1.5;
rx: 8;
}
.flow-node path,
.flow .node.gate path {
fill: var(--surface);
stroke: var(--border-color);
stroke-width: 1.5;
}
.flow-node[data-shape="term"] rect,
.flow .node.term rect {
fill: var(--surface-tint);
rx: 22;
}
.flow-node[data-kind="ok"] rect,
.flow .node.ok rect {
fill: rgba(120, 140, 93, 0.12);
stroke: var(--success);
}
.flow-node[data-kind="bad"] rect,
.flow .node.bad rect {
fill: rgba(176, 74, 63, 0.10);
stroke: var(--danger);
}
.flow-node.active rect,
.flow-node.active path,
.flow .node.active rect,
.flow .node.active path {
stroke: var(--accent);
stroke-width: 2;
}
/* HTML flow nodes, for non-SVG diagrams */
div.flow-node {
padding: var(--space-3) var(--space-4);
background: var(--surface);
border: var(--border);
border-radius: var(--radius-sm);
transition:
transform var(--dur-fast) var(--ease),
border-color var(--dur-fast) var(--ease),
box-shadow var(--dur-fast) var(--ease);
}
div.flow-node:hover {
transform: translateY(var(--hover-y));
border-color: rgba(217, 119, 87, 0.55);
box-shadow: var(--shadow-sm);
}
div.flow-node[data-shape="term"] {
border-radius: var(--radius-pill);
background: var(--surface-tint);
}
div.flow-node[data-kind="ok"] {
border-color: var(--success);
background: rgba(120, 140, 93, 0.12);
}
div.flow-node[data-kind="bad"] {
border-color: var(--danger);
background: rgba(176, 74, 63, 0.10);
}
@media (prefers-reduced-motion: reduce) {
*,
*::before,
*::after {
scroll-behavior: auto !important;
transition-duration: 0.01ms !important;
animation-duration: 0.01ms !important;
animation-iteration-count: 1 !important;
}
}
/* Responsive safety net for generated artifacts. */
p,
li {
overflow-wrap: anywhere;
}
@media (max-width: 860px) {
.split,
.section-rail,
.sidebar-layout {
grid-template-columns: minmax(0, 1fr) !important;
}
.split > *,
.section-rail > *,
.sidebar-layout > *,
.chart-panel {
width: 100%;
max-width: 100%;
min-width: 0;
}
}
</style>
</head>
<body>
<main class="page stack" data-gap="lg">
<header class="stack" data-gap="sm">
<div class="eyebrow">Module explainer</div>
<h1>The rendering checker is a deterministic HTML/visual smoke pipeline that catches Birch contract violations before they ship.</h1>
<p class="lede">A three-stage pipeline — static parsing, headless screenshot capture, and browser geometry audit — turns a single HTML artifact into a structured JSON/Markdown report. Source-grounded from <code>scripts/check_birch_renderings.py</code>, <code>scripts/birch_mpl.py</code>, <code>evals/charts/</code>, and the Birch CSS contracts in <code>styles/birch-system.css</code>.</p>
<div class="cluster">
<span class="chip">Python</span>
<span class="chip" data-tone="clay">Chrome headless</span>
<span class="chip" data-tone="olive">Pillow</span>
<span class="chip" data-tone="sky">Static HTML parsing</span>
</div>
</header>
<section class="section stack" data-gap="lg">
<div class="section-head">
<span class="eyebrow">Mental model</span>
</div>
<div class="panel chart-panel stack" data-gap="sm">
<svg class="chart-svg" viewBox="0 0 680 340" role="img" aria-label="Rendering checker pipeline: artifact/HTML input flows through static checks, browser capture, and geometry audit to produce findings and reports">
<defs>
<marker id="arrow" viewBox="0 0 10 7" refX="10" refY="3.5" markerWidth="8" markerHeight="6" orient="auto"><polygon points="0 0, 10 3.5, 0 7" fill="#3D3D3A"/></marker>
</defs>
<!-- Stage 1: Input -->
<g class="flow-node" data-shape="term">
<rect x="30" y="20" width="140" height="48" rx="22"/>
<text x="100" y="42" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">HTML artifact</text>
<text x="100" y="57" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">--artifact or --pair</text>
</g>
<!-- Stage 2: Static checks -->
<g class="flow-node">
<rect x="230" y="12" width="180" height="64" rx="8"/>
<text x="320" y="36" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">Static HTML/CSS checks</text>
<text x="320" y="54" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">PageStats + contract_findings</text>
<text x="320" y="68" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Component parsers</text>
</g>
<!-- Stage 3: Browser capture -->
<g class="flow-node">
<rect x="460" y="12" width="180" height="64" rx="8"/>
<text x="550" y="36" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">Browser capture</text>
<text x="550" y="54" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Chrome --headless=new</text>
<text x="550" y="68" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">--screenshot + --dump-dom</text>
</g>
<!-- Stage 4: Screenshot analysis -->
<g class="flow-node">
<rect x="260" y="120" width="180" height="64" rx="8"/>
<text x="350" y="144" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">Screenshot analysis</text>
<text x="350" y="162" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Pillow palette/diff metrics</text>
<text x="350" y="176" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Blank render detection</text>
</g>
<!-- Stage 5: Geometry audit -->
<g class="flow-node">
<rect x="460" y="120" width="180" height="64" rx="8"/>
<text x="550" y="144" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">Geometry audit</text>
<text x="550" y="162" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Injected JS layout probes</text>
<text x="550" y="176" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">Overflow/wrap detection</text>
</g>
<!-- Stage 6: Output -->
<g class="flow-node" data-shape="term">
<rect x="360" y="240" width="140" height="48" rx="22"/>
<text x="430" y="262" text-anchor="middle" fill="#141413" font-family="system-ui,sans-serif" font-size="13" font-weight="600">Reports</text>
<text x="430" y="277" text-anchor="middle" fill="#87867F" font-family="ui-monospace,monospace" font-size="10">JSON + Markdown</text>
</g>
<!-- Edges -->
<line x1="170" y1="44" x2="222" y2="44" stroke="#87867F" stroke-width="1.5" marker-end="url(#arrow)"/>
<line x1="410" y1="44" x2="452" y2="44" stroke="#87867F" stroke-width="1.5" marker-end="url(#arrow)"/>
<path d="M550,76 L550,112" stroke="#87867F" stroke-width="1.5" marker-end="url(#arrow)"/>
<line x1="550" y1="76" x2="550" y2="96"/>
<path d="M550,76 L350,76 L350,112" stroke="#87867F" stroke-width="1.5" fill="none" marker-end="url(#arrow)"/>
<line x1="350" y1="184" x2="350" y2="204" stroke="#87867F" stroke-width="1.5"/>
<path d="M350,204 L350,234 L430,234" stroke="#87867F" stroke-width="1.5" fill="none" marker-end="url(#arrow)"/>
<line x1="550" y1="184" x2="550" y2="204" stroke="#87867F" stroke-width="1.5"/>
<path d="M550,204 L550,234 L430,234" stroke="#87867F" stroke-width="1.5" fill="none" marker-end="url(#arrow)"/>
</svg>
<p class="chart-caption">Pipeline flow from <code>main()</code> in <code>scripts/check_birch_renderings.py</code>. In --artifact mode (preferred for generated eval outputs), the screenshot analysis and geometry audit run against a single file; in --pair mode, they compare reference and candidate.</p>
</div>
<div class="card stack" data-variant="outlined">
<h3>What the checker actually does</h3>
<p>At its core, <code>check_birch_renderings.py</code> is a <strong>deterministic, parse-first, browser-augmented</strong> validator. It does not use vision models, ML, or heuristics trained on prior artifacts. It reads the HTML file as text, parses it with specialized <code>HTMLParser</code> subclasses, then optionally launches Chrome to capture a screenshot and run a geometry audit script. Every finding is a structured (level, name, evidence) triple produced by a named check function.</p>
<p>The checker was designed to be <strong>simpler and more deterministic</strong> than the neighboring Birchline tooling. Its static checks are pure Python <code>html.parser.HTMLParser</code> subclasses — no DOM, no layout engine, no heuristic thresholds beyond explicit CSS minimums. The browser stage adds visual smoke detection that static parsing cannot see: blank renders, overflowing text, misaligned metric bars, and wrapping pathology.</p>
</div>
</section>
<section class="section stack" data-gap="lg">
<div class="section-head">
<span class="eyebrow">Runtime path</span>
<h2>From input paths to findings</h2>
</div>
<ol class="flow-list">
<li class="flow-step">
<span class="flow-num">1</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Parse CLI args and resolve inputs</h3>
<p class="flow-detail"><code>main()</code> in <code>scripts/check_birch_renderings.py</code> resolves <code>--artifact</code> (single-file) or <code>--pair</code> (reference:candidate) paths relative to the repo root. In artifact mode, each path is checked independently; pair mode compares stats and screenshots between original and candidate. The two modes are mutually exclusive.</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">2</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Run static HTML/CSS contract checks</h3>
<p class="flow-detail"><code>page_stats()</code> feeds the HTML through <code>StatsParser</code> to count tags, classes, CSS vars, words, and style bytes. Then <code>contract_findings()</code> runs a battery of checks: doctype, viewport meta, Birch CSS embedding (≥10 KB with all five signatures), <code>.page</code> shell, layout/semantic primitive counts, unknown CSS variables, patch marker lines, gradient prohibition, palette color purity, and SVG chart mark correctness.</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">3</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Run component-specific contract parsers</h3>
<p class="flow-detail">Seven focused <code>HTMLParser</code> subclasses validate child-count contracts for Birch components: <code>GridListParser</code> (insight/takeaway list children), <code>ChecklistParser</code> (inline markup in checklist rows), <code>MetricRowParser</code> (caption + meter + code), <code>DiffRowParser</code> (ln + mark + code columns), <code>DiffPolarityParser</code> (add/del class or data-kind), <code>FlowStepParser</code> (flow-num + one wrapper), and <code>TimelineParser</code> (timeline-time + timeline-body).</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">4</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Launch headless Chrome for screenshot capture</h3>
<p class="flow-detail"><code>capture()</code> opens the HTML file via <code>file://</code> URI in Chrome <code>--headless=new</code> with <code>--no-sandbox</code>. For deep viewports, <code>capture_height_for_viewport()</code> measures the document height first via an injected script that sets <code>data-birch-doc-height</code>, then captures at the measured height (capped at the viewport max). Screenshots land in <code>reports/birch-screenshots/</code>.</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">5</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Analyze screenshots with Pillow</h3>
<p class="flow-detail">In artifact mode, <code>artifact_screenshot_metrics()</code> sub-samples the PNG, computes background/non-background/blackish fractions, and checks for blank renders (file &lt;2 KB or &lt;2.5% non-background). In pair mode, <code>screenshot_metrics()</code> computes per-pixel RGB difference, generates an amplified diff PNG and a contact sheet.</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">6</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Run browser geometry audit</h3>
<p class="flow-detail"><code>geometry_audit()</code> injects a ~500-line JS script that probes the live layout: overflow detection, container text spill, stat-card squeeze, metric bar alignment, timeline marker position, code-block wrapping pathology, and prose text underfill. Results are serialized into <code>data-birch-audit</code> on the body and extracted from Chrome's <code>--dump-dom</code> output.</p>
</div>
</li>
<li class="flow-step">
<span class="flow-num">7</span>
<div class="stack" data-gap="xs">
<h3 class="flow-title">Emit JSON and Markdown reports</h3>
<p class="flow-detail">All findings are collected into a payload with a summary hash (mode, artifact/pair count, failures, warnings, notes). <code>render_markdown()</code> produces a human-readable Markdown file with per-artifact tables; the raw JSON is written to the <code>--out</code> path. The script exits with code 1 on failures (or on warnings when <code>--fail-on-warn</code>).</p>
</div>
</li>
</ol>
</section>
<section class="section stack" data-gap="lg">
<div class="section-head">
<div>
<span class="eyebrow">File tour</span>
<h2>Reading order for new contributors</h2>
</div>
</div>
<p class="muted">Each file is described with its role in the checker pipeline, the key functions/classes to read first, and how it connects to the others. Start at the top.</p>
<div class="stack" data-gap="md">
<article class="card stack" data-variant="outlined" data-accent="clay">
<div class="card-head">
<div class="avatar">1</div>
<div class="card-titles">
<h3 class="card-title"><code>scripts/check_birch_renderings.py</code></h3>
<p class="card-sub">~1,400 lines · The main checker</p>
</div>
</div>
<p>This is the heart of the pipeline. Start with <code>main()</code> to see the top-level flow, then read <code>check_artifact()</code> (the artifact-mode entry point) and <code>contract_findings()</code> (the static check battery). The <code>HTMLParser</code> subclasses — <code>StatsParser</code>, <code>GridListParser</code>, <code>ChecklistParser</code>, <code>MetricRowParser</code>, <code>DiffRowParser</code>, <code>DiffPolarityParser</code>, <code>FlowStepParser</code>, <code>TimelineParser</code> — are each ~40–80 lines and validate a single component contract. The browser integration lives in <code>find_chrome()</code>, <code>capture()</code>, <code>geometry_audit()</code>, and <code>geometry_findings()</code>.</p>
<p class="muted">Key functions: <code>main</code>, <code>check_artifact</code>, <code>contract_findings</code>, <code>page_stats</code>, <code>geometry_audit</code>, <code>geometry_findings</code>, <code>svg_visual_findings</code>.</p>
</article>
<article class="card stack" data-variant="outlined">
<div class="card-head">
<div class="avatar">2</div>
<div class="card-titles">
<h3 class="card-title"><code>scripts/birch_mpl.py</code></h3>
<p class="card-sub">~120 lines · Matplotlib helpers</p>
</div>
</div>
<p>Provides the Birch Matplotlib theme used by artifact generators like <code>build_chart_brief.py</code>. Read <code>rc_params()</code> for the token-to-Matplotlib color mapping, <code>birch_theme()</code> (a context manager that applies rcParams), <code>polish_axes()</code> for grid/spine cleanup, and <code>svg_string()</code> for serializing a figure as inlineable SVG. This file is not called by the checker; it is used during artifact generation.</p>
<p class="muted">Key functions: <code>rc_params</code>, <code>birch_theme</code>, <code>polish_axes</code>, <code>svg_string</code>, <code>close</code>.</p>
</article>
<article class="card stack" data-variant="outlined">
<div class="card-head">
<div class="avatar">3</div>
<div class="card-titles">
<h3 class="card-title"><code>evals/charts/build_chart_brief.py</code></h3>
<p class="card-sub">~200 lines · Example artifact generator</p>
</div>
</div>
<p>Generates <code>24-birch-chart-brief.html</code> from <code>evals/charts/sample-data.csv</code>. This is the best example of what a checked artifact looks like: it loads CSV rows, builds a Matplotlib SVG via <code>birch_mpl</code>, assembles KPI cards, metric rows, and a numeric table, then writes the complete Birch HTML. Read <code>main()</code> and <code>build_svg()</code> first; the HTML template at the end shows the canonical Birch structure the checker expects.</p>
<p class="muted">Key functions: <code>main</code>, <code>build_svg</code>, <code>metric_rows</code>, <code>table_rows</code>.</p>
</article>
<article class="card stack" data-variant="outlined">
<div class="card-head">
<div class="avatar">4</div>
<div class="card-titles">
<h3 class="card-title"><code>evals/charts/run_eval.py</code></h3>
<p class="card-sub">~50 lines · Eval runner</p>
</div>
</div>
<p>Ties generation and checking together. Runs <code>build_chart_brief.py</code> first, then calls <code>check_birch_renderings.py --artifact</code> at four viewports (desktop, mobile, deep, mobile-deep). Each viewport produces its own JSON and Markdown report. This is the pattern to follow when adding new eval artifacts.</p>
<p class="muted">Key: the <code>VIEWPORTS</code> dict and the <code>run()</code> loop.</p>
</article>
<article class="card stack" data-variant="outlined">
<div class="card-head">
<div class="avatar">5</div>
<div class="card-titles">
<h3 class="card-title"><code>docs/birch-llm-style-guide.md</code></h3>
<p class="card-sub">~460 lines · The contract being checked</p>
</div>
</div>
<p>This is the authoritative LLM contract that the checker enforces. It defines required shell structure, layout primitives, semantic components, CSS variable tokens, and strict child-count contracts for diff rows, flow steps, metric rows, and timeline items. When you add a new check to the checker, it should correspond to a rule in this guide.</p>
<p class="muted">Key sections: Required page shell, quick primitive index, diff, timeline, flow steps, metric rows, charts and numeric data, rules for LLM generation.</p>
</article>
<article class="card stack" data-variant="outlined">
<div class="card-head">
<div class="avatar">6</div>
<div class="card-titles">
<h3 class="card-title"><code>styles/birch-system.css</code></h3>
<p class="card-sub">~1,100 lines · The CSS that implements the contracts</p>
</div>
</div>
<p>The canonical stylesheet that the checker verifies is embedded (or linked). The checker asserts the embedded Birch CSS block is ≥10 KB and contains the five signature strings: <code>--color-ivory</code>, <code>--font-serif</code>, <code>.page</code>, <code>.section</code>, <code>.card</code>. When geometry audit probes measure overflow or wrapping, they are testing against the layout defined here.</p>
<p class="muted">Key: the CSS variable tokens section, the layout primitives responsive breakpoints (860px, 620px), and the component styles for diff, timeline, flow-step, metric-row.</p>
</article>
</div>
</section>
<section class="section stack" data-gap="lg">
<div class="section-head">
<span class="eyebrow">Gotchas</span>
<h2>Things that will surprise you</h2>
</div>
<div class="grid" data-cols="2">
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>HTMLParser is not a DOM</h3>
<p>Every parser subclass implements its own depth-tracking stack. There is no shared tree, no CSS cascade, and no computed style. The <code>handle_starttag</code> / <code>handle_endtag</code> / <code>handle_data</code> callbacks are streaming callbacks — you must track nesting state manually. This is why every parser has its own <code>self.depth</code>, <code>self.in_*</code> flag, and <code>self.*_stack</code>.</p>
</article>
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>Chrome needs the real file path</h3>
<p>The checker opens the HTML via <code>file://</code> URI pointing at the actual repo path so that relative <code>styles/birch-system.css</code> links resolve. The geometry audit injects a <code>&lt;base href="..."&gt;</code> element into a temp copy to preserve these relative links. If you move the repo or rename directories, screenshots will capture unstyled HTML.</p>
</article>
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>Diff polarity requires both the mark and the class</h3>
<p>A <code>+</code> in <code>.mark</code> is visible but does not trigger green row styling. <code>DiffPolarityParser</code> checks that rows with <code>+</code> or <code>-</code> in the mark column also have <code>.add</code> / <code>.del</code> class or <code>data-kind="add/del"</code>. The checker fails rows that rely on mark text alone.</p>
</article>
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>SVG <code>polyline</code> defaults to black fill</h3>
<p>Browsers fill un-styled <code>&lt;polyline&gt;</code> and <code>&lt;path&gt;</code> elements with black. <code>svg_visual_findings()</code> scans for chart-line elements that lack <code>fill="none"</code> or an explicit stroke attribute. This is a common LLM generation failure that produces large black wedges instead of line charts.</p>
</article>
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>Metric alignment is measured in the browser</h3>
<p>The <code>geometry_audit()</code> JS measures the left offset of every <code>.meter</code> within a <code>.metric-list</code>. A spread &gt;2 px across meters in the same list produces a warning. This cannot be detected statically because it depends on the rendered width of label text.</p>
</article>
<article class="card stack" data-variant="outlined" data-tone="warning">
<h3>Code wrap detection uses tree-walker rects</h3>
<p>The geometry audit's <code>rectStatsForText()</code> walks text nodes with <code>document.createTreeWalker</code> and groups <code>getClientRects()</code> into visual lines. It compares visual line count to logical (newline-split) line count. If a code block that should wrap produces 3× more visual lines than logical lines with &lt;18 chars per line, it is flagged as underfilled.</p>
</article>
</div>
</section>
<section class="section" data-gap="xl">
<div class="section-rail">
<div class="section-head">
<span class="eyebrow">Extension points</span>
<h2>Where to add or change behavior</h2>
</div>
<div class="stack" data-gap="md">
<article class="card stack" data-variant="outlined">
<h3>Add a new static contract check</h3>
<p>Find the relevant class in <code>BIRCH_CLASSES</code>, <code>LAYOUT_CLASSES</code>, or <code>SEMANTIC_CLASSES</code> in <code>check_birch_renderings.py</code> (lines ~60–100) and add it if missing. Then add a new <code>check(...)</code> call inside <code>contract_findings()</code> (line ~400). If the check requires custom HTML parsing, write a new <code>HTMLParser</code> subclass (~40–80 lines) following the pattern of <code>MetricRowParser</code> or <code>FlowStepParser</code>:</p>
<ul class="plain-list">
<li>Track entry/exit with a boolean flag and a depth counter.</li>
<li>Collect direct children (depth=1 elements).</li>
<li>Validate the child contract in <code>handle_endtag</code> when depth returns to 0.</li>
<li>Append to a <code>self.bad_*</code> list with a preview snippet.</li>
</ul>
<p>Wire the new parser into <code>contract_findings()</code> and emit a <code>Finding</code> when the bad-items list is non-empty.</p>
</article>
<article class="card stack" data-variant="outlined">
<h3>Add a new geometry audit probe</h3>
<p>Inside the <code>audit_js</code> string in <code>geometry_audit()</code>, add a new <code>querySelectorAll</code> loop that measures element positions, sizes, or overflow. Serialize results into the <code>data-birch-audit</code> JSON object. Then add a corresponding block in <code>geometry_findings()</code> that reads the key from the parsed JSON and emits <code>Finding</code> objects. The pattern is: <code>if problems: emit fail; elif key exists: emit pass; else: skip (pre-audit-checker baseline)</code>.</p>
</article>
<article class="card stack" data-variant="outlined">
<h3>Add a new supported viewport</h3>
<p>Add an entry to the <code>VIEWPORTS</code> dict in <code>evals/charts/run_eval.py</code> (or your own eval runner). The format is <code>"name": "label:WxH"</code>. The checker's <code>parse_viewport()</code> splits on <code>:</code> and <code>x</code>. For "deep" viewports (name contains "deep"), the checker measures document height before capturing; for standard viewports it uses the exact height.</p>
</article>
<article class="card stack" data-variant="outlined">
<h3>Change palette/token validation</h3>
<p>The canonical palette lives in <code>PALETTE</code> (line ~100 of <code>check_birch_renderings.py</code>) and is duplicated in <code>COLORS</code> in <code>birch_mpl.py</code>. Update both if you change the Birch palette. The <code>is_palette_hex()</code> function uses a tolerance of 3 per channel. <code>palette_metrics()</code> uses tolerance 34 for the "palette-close" fraction. Adjust these tolerances in the checker only; the CSS token values in <code>birch-system.css</code> are the source of truth.</p>
</article>
<article class="card stack" data-variant="outlined">
<h3>Change the Birch CSS contract signature</h3>
<p>The five strings in <code>BIRCH_CSS_SIGNATURES</code> (line ~40) are what the checker scans for to validate the embedded Birch CSS block. If you add or rename a critical CSS variable or class, update this tuple. The <code>BIRCH_CSS_MIN_BYTES</code> threshold (10,000) is also defined there. These signatures are matched against the raw CSS text, not parsed selectors.</p>
</article>
</div>
<aside class="reference-panel stack" data-gap="md">
<h3>Key constants</h3>
<div class="stack" data-gap="sm">
<div class="stat-card">
<span class="caption">BIRCH_CSS_MIN_BYTES</span>
<strong class="stat-value" style="font-size:24px">10,000</strong>
<p class="muted">Minimum embedded Birch CSS block size. Defined at line ~40 of the checker.</p>
</div>
<div class="stat-card">
<span class="caption">SEMANTIC_CLASS_MIN</span>
<strong class="stat-value" style="font-size:24px">6</strong>
<p class="muted">Minimum distinct semantic component classes. Below this, the artifact likely uses page-specific CSS.</p>
</div>
<div class="stat-card">
<span class="caption">LAYOUT_CLASS_MIN</span>
<strong class="stat-value" style="font-size:24px">8</strong>
<p class="muted">Minimum layout primitive uses. Below this, the layout may be brittle.</p>
</div>
<div class="stat-card">
<span class="caption">PALETTE_TOLERANCE</span>
<strong class="stat-value" style="font-size:24px">3/ch</strong>
<p class="muted">Per-channel RGB tolerance for <code>is_palette_hex()</code>. Palette-metrics tolerance is 34.</p>
</div>
</div>
<h3>Report paths</h3>
<div class="stack" data-gap="xs">
<pre class="code-block" data-kind="command">reports/birch-rendering-check.json</pre>
<pre class="code-block" data-kind="command">reports/birch-rendering-check.md</pre>
<pre class="code-block" data-kind="command">reports/birch-screenshots/*.png</pre>
</div>
<h3>Key commands</h3>
<div class="stack" data-gap="xs">
<pre class="code-block" data-kind="command">uv run --with pillow python scripts/check_birch_renderings.py --artifact 24-birch-chart-brief.html --viewport desktop:1365x900</pre>
<pre class="code-block" data-kind="command">uv run --with pillow python scripts/check_birch_renderings.py --artifact 24-birch-chart-brief.html --viewport mobile:390x900 --fail-on-warn</pre>
<pre class="code-block" data-kind="command">uv run python evals/charts/run_eval.py</pre>
</div>
<h3>Finding levels</h3>
<ul class="plain-list">
<li><strong>fail</strong> — hard contract violation; blocks adoption</li>
<li><strong>warn</strong> — likely problem, non-blocking unless <code>--fail-on-warn</code></li>
<li><strong>note</strong> — informational delta or diagnostic</li>
<li><strong>pass</strong> — check passed (emitted only for geometry/audit checks)</li>
</ul>
</aside>
</div>
</section>
<section class="section stack" data-gap="md">
<div class="card stack" data-variant="flat">
<h2>Source and caveat</h2>
<p>All claims are grounded in the six files listed in <code>evals/module-explainer/source-files.txt</code>, inspected at the time of writing. The checker is a living tool; function line numbers may drift. The geometry audit JS is the most complex and fragile part — it probes live browser layout and depends on the Birch CSS being loaded correctly. If Chrome or the Birch CSS changes, audit probes may need adjustment. The <code>--pair</code> mode is maintained for backward compatibility but <code>--artifact</code> mode is preferred for generated eval outputs.</p>
</div>
</section>
</main>
</body>
</html>

Xet Storage Details

Size:
61.4 kB
·
Xet hash:
fd5691872cbe8b1a95e206d4e92f21e665e43a50dbd8ed7fe9cdef3356917b9b

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.