PIG/PIG-Demo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Capture harness, fixture verification, CI, and the public README

Browse Source 
The site does no live inference. Rollouts are captured once against spark-1 and
replayed at their recorded wall-clock — a public demo with no auth cannot hold
an API key, and a recorded run can be scrubbed, permalinked, blind-compared and
verified in ways a live one cannot. What stops it being a video is that the
browser re-derives every number from the recorded moves.

verify_fixtures.py is the Python half of that: it replays every committed
fixture through the engine and reproduces its own rewards. All 16 land at
delta 0.0. A fixture that cannot be regenerated is a claim with no receipt.

First real measurement, thinking off, 8 seeds: solved 0/8. The model repeats
guesses it has already played, invents words (trape, slith, postt, boomy),
and contradicts its own feedback — consistency 0.09 to 0.17. That is the
published failure taxonomy showing up in our own data on the first run, and it
is why `consistency` is a reward component rather than a footnote.

A capture failure is recorded as a turn with a null reply, never dropped. A
capture that silently discarded failed turns would be reporting a better model
than the one that ran.

CI gates both halves and four things that fail silently in production: the word
lists must rebuild byte-identically, the prerendered routes must carry their own
baked og tags (crawlers do not run JS, so without them every shared link
previews as the homepage), no blob: URL may reach the bundle (the site's CSP has
no worker-src, so it falls back to default-src 'self' and a blob worker is
blocked with no error), and the conformance digest must match across languages.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019mt6sHQHEnEYrJZvoMCJSB
This commit is contained in:
karti-ai
2026-08-28 15:47:31 -07:00
parent 69607fbfe9
commit 408ce4a525
43 changed files with 5279 additions and 136 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/_lib.mjs
+58 -18
View File
@@ -294,7 +294,7 @@ function codeSpans(src) {
* @param {'{' | '['} open
* @returns {string | null}
*/
export function literalAfter(src, anchor, open = '{') {
export function literalsAfter(src, anchor, open = '{') {
const flags = anchor.flags.includes('g') ? anchor.flags : `${anchor.flags}g`;
const re = new RegExp(anchor.source, flags);
const spans = codeSpans(src);
@@ -302,18 +302,29 @@ export function literalAfter(src, anchor, open = '{') {
const hit = spans.find((s) => offset >= s.start && offset < s.end);
return hit ? hit.code : true;
};
const found = [];
let match;
while ((match = re.exec(src)) !== null) {
// The anchor may legally match inside a comment or a string — a docblock
// The anchor may legally match inside a comment or a string \u2014 a docblock
// that quotes `const meta = {`. Only a match in real code counts.
if (!isCode(match.index)) continue;
const from = src.indexOf(open, match.index + Math.max(match[0].length - 1, 0));
if (from === -1) continue;
const between = src.slice(match.index + match[0].length, from);
if (/[;=}]/.test(between)) continue;
return src.slice(from, scanBalanced(src, from));
const end = scanBalanced(src, from);
found.push({ text: src.slice(from, end), start: from, end });
// Skip past this literal so a nested anchor of the same name is not
// reported as a second, overlapping hit.
re.lastIndex = end;
}
return null;
return found;
}
/** First literal following `anchor`, or null. See {@link literalsAfter}. */
export function literalAfter(src, anchor, open = '{') {
const [first] = literalsAfter(src, anchor, open);
return first ? first.text : null;
}
/** Marker key on the stand-in an unresolvable identifier evaluates to. */
@@ -501,7 +512,10 @@ const VERTICAL_ANCHORS = [
const asVertical = (id, v) => ({
id,
title: v?.title ?? v?.label ?? v?.name ?? null,
description: v?.description ?? v?.blurb ?? v?.tagline ?? v?.summary ?? null,
// `anxiety` is this repo's field for the one line a vertical page leads
// with, and it is the only prose in a VerticalEntry short enough to be a
// meta description.
description: v?.description ?? v?.blurb ?? v?.tagline ?? v?.summary ?? v?.anxiety ?? null,
});
/**
@@ -559,8 +573,14 @@ export function loadManifest() {
}
/** @type {Map<string, {runs: any[], extras: Record<string, any>}>} */
const byDemo = new Map();
const container =
raw && typeof raw === 'object' && !Array.isArray(raw) && raw.demos && typeof raw.demos === 'object' ? raw.demos : raw;
// Unwrap the two envelopes the shell's own reader accepts. `{runs: [...]}`
// has to be unwrapped BEFORE the record branch below, or "runs" is read as a
// demo slug and every real demo reports zero runs.
let container = raw;
if (raw && typeof raw === 'object' && !Array.isArray(raw)) {
if (raw.demos && typeof raw.demos === 'object') container = raw.demos;
else if (Array.isArray(raw.runs)) container = raw.runs;
}
if (Array.isArray(container)) {
// A flat array of runs, each carrying its own `demo`/`slug`.
@@ -706,12 +726,32 @@ export function discoverRoutePatterns() {
return [...patterns];
}
const DEMO_PARAMS = /^(slug|demo|demoSlug)$/;
const VERTICAL_PARAMS = /^(vertical|verticalId|sector|category)$/;
/**
* Which set fills a pattern's one parameter.
*
* Decided by the STATIC segment in front of it, not by the parameter's name.
* This router calls both of them `:slug` — `demos/:slug` and `verticals/:slug`
* — so a name-based reader confidently prerenders `/verticals/wordle`, twelve
* 404s, and no vertical pages at all.
*/
function fillFor(pattern) {
const head = pattern.replace(/^\/+/, '').split('/')[0]?.toLowerCase() ?? '';
if (/^(demos?|d)$/.test(head)) return 'demo';
if (/^(verticals?|v|sectors?)$/.test(head)) return 'vertical';
const param = pattern.match(/:([A-Za-z0-9_]+)/)?.[1] ?? '';
if (/^(vertical|verticalId|sector|category)$/.test(param)) return 'vertical';
if (/^(slug|demo|demoSlug|id)$/.test(param)) return 'demo';
return null;
}
/**
* Expands the router's patterns against the real data into concrete paths.
*
* Child paths are taken as written. This router nests everything exactly one
* level under `/`, so a child's `path` is already the full path; a second level
* of nesting would need the parent prefix joined on, and this would silently
* emit the wrong routes rather than fail. If you nest deeper, fix this.
*
* @param {{metas: Map<string, any>, verticals: {id: string}[]}} data
* @returns {{routes: {path: string, kind: string, slug?: string, id?: string}[], errors: string[]}}
*/
@@ -749,18 +789,18 @@ export function expandRoutes(data) {
errors.push(`route pattern "${pattern}" has more than one parameter; prerender cannot expand it.`);
continue;
}
const param = params[0];
const fill = (value, record) => add(pattern.replace(/:[A-Za-z0-9_]+\??/, value), record);
if (VERTICAL_PARAMS.test(param)) {
if (!verticalIds.length) errors.push(`route "${pattern}" needs verticals, but none were readable from src/content/verticals.ts.`);
for (const id of verticalIds) fill(id, { kind: 'vertical', id });
} else if (DEMO_PARAMS.test(param) || param === 'id') {
const fill = fillFor(pattern);
const substitute = (value) => pattern.replace(/:[A-Za-z0-9_]+\??/, value);
if (fill === 'vertical') {
if (!verticalIds.length) errors.push(`route "${pattern}" needs verticals, but none were readable from ${rel(VERTICALS_FILE)}.`);
for (const id of verticalIds) add(substitute(id), { kind: 'vertical', id });
} else if (fill === 'demo') {
if (!demoSlugList.length) errors.push(`route "${pattern}" needs demos, but no demo meta was readable under src/demos/.`);
for (const slug of demoSlugList) fill(slug, { kind: 'demo', slug });
for (const slug of demoSlugList) add(substitute(slug), { kind: 'demo', slug });
} else {
errors.push(
`route pattern "${pattern}" uses parameter ":${param}", which prerender cannot fill. ` +
'Name it :slug (a demo) or :vertical (a vertical), or teach scripts/_lib.mjs about it.',
`route pattern "${pattern}" has a parameter prerender cannot fill. Put it under /demos/ or ` +
'/verticals/, or teach fillFor() in scripts/_lib.mjs about it.',
);
}
}