doxa-maps/library/scripts/posters/packet.mjs

// packet.mjs — the UPG-100 PRAYER PACKET generator.
//
// One printable packet per zone. Structure is the Driver's, from the ideation
// (_qa-cycle/01-ideation-research/prayer-points.md):
//
//   page 1      the MAP — a lightweight vertical one-pager, NOT the poster format
//   pages 2..n  the PEOPLE GROUPS — 20 per page, photo + name + brief detail
//   page n+1    the 12 PRAYER POINTS
//   page n+2    the PRAYER WHEEL — the rhythm these twelve points follow
//
// Printed front-and-back: Laos (95 groups) is 8 pages = 4 sheets.
//
// ── WHY A SEPARATE SCRIPT, not a FORMAT inside generate.mjs ──────────────────
// generate.mjs is a 2,600-line POSTER renderer. The packet is a different artefact
// with different pages; forking it would duplicate the design. Instead this reads
// the SAME zone resolution (clusters.json + the selector evaluator) so a packet can
// never disagree with its poster about which groups belong to a zone.
//
// ── EVERY STRING IS IN packet-strings.json ──────────────────────────────────
// Driver: "it all needs to be generative so that I can generate it in multiple
// languages." So no human-readable text is written in this file. LANG=xx picks a
// block; missing keys fall back to English individually, so a partial translation
// renders what it has rather than blanking the page.
//
// ── PHOTOS ──────────────────────────────────────────────────────────────────
// MEASURED: 1,072 of 2,107 groups have one (51%); they average ~244KB on Joshua
// Project's CDN, ~0.8s each. Fetched in parallel at packet time and inlined as
// data URIs so the finished HTML is ONE self-contained file that prints anywhere.
// Groups without a photo get a typographic card, never a broken image box.
//
// Usage:
//   node packet.mjs                      # every zone, English
//   ONLY=laos node packet.mjs            # one zone
//   LANG=es ONLY=laos node packet.mjs    # one zone, Spanish
//   NOPHOTOS=1 ONLY=laos node packet.mjs # skip the fetch (fast layout iteration)
//
// Output: out/packets/<zoneKey>[.<lang>].html

import { readFileSync, writeFileSync, mkdirSync, existsSync } from 'node:fs';
import { fileURLToPath } from 'node:url';
import { dirname, join } from 'node:path';

const __dirname = dirname(fileURLToPath(import.meta.url));
const DATA = join(__dirname, '../../../plugin-designs/my-upg-100-list/data');
const OUT = join(__dirname, 'out', 'packets');
mkdirSync(OUT, { recursive: true });

// ⚠ NOT process.env.LANG — that is the SHELL's POSIX locale ("en_US.UTF-8"), which
// is not a valid BCP-47 tag and throws in localeCompare/toLocaleString. Measured:
// it crashed the first run. PACKET_LANG is the only source.
const LANG = process.env.PACKET_LANG || 'en';
const ONLY = process.env.ONLY || '';
const NOPHOTOS = process.env.NOPHOTOS === '1';
const API = process.env.VITE_API_BASE_URL || process.env.MAP_APP_API_URL;
// FAIL CLOSED without an API — EXCEPT when records come from a committed fixture, which
// is the one case where no API is needed at all. Measured 2026-09-24: this guard runs
// BEFORE loadRecords() ever looks at PACKET_RECORDS, so `bun run packet:ci` exited 1 on a
// tree that had everything it needed. It only passed in hand-testing because I had been
// supplying a dummy host — the real invocation path is what exposed it.
if (!API && !process.env.PACKET_RECORDS) {
  console.error('[packet] No API base — set VITE_API_BASE_URL, or PACKET_RECORDS for a fixture build.');
  console.error('         Failing closed.');
  process.exit(1);
}

// ── STRINGS: pick the language, fall back per-key to English ────────────────
const STRINGS = JSON.parse(readFileSync(join(__dirname, 'packet-strings.json'), 'utf8'));
const EN = STRINGS.en;
const L = STRINGS[LANG] || {};
// Per-KEY fallback (not whole-block): a half-finished translation should render
// its finished keys, not fall back wholesale to English.
const t = (k) => (L[k] !== undefined ? L[k] : EN[k]);
const tr = (k) => ((L.rhythms || {})[k] !== undefined ? L.rhythms[k] : EN.rhythms[k]);
// NO `POINTS` HERE. The renderer owns point resolution (packet-render.js), and it must
// merge per-point/per-field — a partial `points` array is truncating, not falling back.
// A second copy of that logic in this file was DEAD (assigned, never read) and would have
// re-introduced the bug the moment someone wired it up.
const fill = (s, vars) => String(s).replace(/\{(\w+)\}/g, (_, k) => (vars[k] ?? ''));

// ── THE RENDERER IS SHARED ──────────────────────────────────────────────────
// The pages themselves live in library/packet/packet-render.js so that the browser
// (the UPG-100 legend's "Prayer packet" export) and this build-time script render
// the SAME packet from the SAME code. Changing a page here changes it in both.
const { createPacketRenderer } = await import(
  join(__dirname, '..', '..', 'packet', 'packet-render.js'));
// SCRIPT FONT, inlined. Some languages need a face the system stack does not
// carry — Hindi renders as tofu boxes in Georgia/Palatino/serif. The packet is a
// standalone file meant to print offline, so the font is base64'd INTO it rather
// than linked, exactly as the photos are. Absent, the renderer omits the rule and
// falls back to serif rather than failing.
const SCRIPT_FONT_FILE = {
  hi: 'noto-serif-devanagari/noto-serif-devanagari-400-devanagari.woff2',
  mr: 'noto-serif-devanagari/noto-serif-devanagari-400-devanagari.woff2',
  ne: 'noto-serif-devanagari/noto-serif-devanagari-400-devanagari.woff2',
};
let fontData = null;
const fontRel = SCRIPT_FONT_FILE[LANG];
if (fontRel) {
  const fp = join(__dirname, '..', '..', '..', 'data', 'fonts', fontRel);   // fonts are sources in data/ (BUILD-OUTPUT-IS-ONE-FOLDER)
  if (existsSync(fp)) {
    fontData = readFileSync(fp).toString('base64');
  } else {
    // Loud, not silent: a missing font here means the packet prints as empty
    // rectangles, which looks like a content bug rather than a missing asset.
    console.warn(`[packet] ⚠ ${LANG} needs ${fontRel} and it is MISSING — `
      + 'the packet will render as tofu boxes. Vendor the font first.');
  }
}
const R = createPacketRenderer({ strings: STRINGS, lang: LANG, fontData });
const { mapPage, listPages, pointsPage, wheelPage, css, mapCss, PER_PAGE, W, H } = R;

// ── DATA ────────────────────────────────────────────────────────────────────
const clusters = JSON.parse(readFileSync(join(DATA, 'clusters.json'), 'utf8'));
const { evaluate } = await import(join(DATA, 'selector-evaluator.js'));
const { normalizeRecords, extractRecords } = await import(join(DATA, 'normalize.js'));

async function loadRecords() {
  // ★ A COMMITTED FIXTURE, so CI can build a packet at all. out/ is gitignored — both the
  // records cache AND every generated packet — so on a clean checkout check:packets-all
  // finds nothing and SKIPs. MEASURED 2026-09-24: CI runs the sweep on EVERY build via
  // gates:post and it has SKIPped every single time, which means every packet assertion in
  // this tree (the 12 points, the wheel labels, page 1's geography, the print layer, the
  // build stamp) has never once been evaluated there. A SKIP that is always taken is
  // indistinguishable from an unwired gate, and the wave-membership census cannot see the
  // difference because membership measures INVOCATION, not EVALUATION.
  // PACKET_RECORDS points at a small committed subset instead of the live API or the
  // gitignored cache. It is deliberately NOT the default: real packets must use real data.
  const fixture = process.env.PACKET_RECORDS;
  if (fixture) {
    if (!existsSync(fixture)) throw new Error(`PACKET_RECORDS set but missing: ${fixture}`);
    return JSON.parse(readFileSync(fixture, 'utf8'));
  }
  const cache = join(__dirname, 'out', '_records.json');
  if (process.env.OFFLINE && existsSync(cache)) return JSON.parse(readFileSync(cache, 'utf8'));
  const fields = 'slug,name,country_code,population,religion,image_url,has_photo,picture_credit,latitude,longitude';
  const r = await fetch(`${API}/api/people-groups/list?fields=${fields}&lang=${LANG}`,
    { headers: { accept: 'application/json' } });
  if (!r.ok) throw new Error(`PG API ${r.status}`);
  const recs = normalizeRecords(extractRecords(await r.json()));
  try { writeFileSync(cache, JSON.stringify(recs)); } catch { /* best effort */ }
  return recs;
}

function allZones() {
  const out = [];
  for (const k of ['lists', 'asia_lists', 'uncovered_lists']) {
    for (const z of clusters[k] || []) out.push({ ...z, zoneKey: z.id });
  }
  return out;
}

// ── PHOTOS: parallel fetch, inlined as data URIs ────────────────────────────
// Bounded concurrency: 16 was measured at 20 photos in 5.9s. Unbounded would open
// ~118 sockets against someone else's CDN, which is rude and slower in practice.
async function fetchPhotos(groups, limit = 16) {
  if (NOPHOTOS) return new Map();
  const out = new Map();
  // ★ PHOTOS_EVERY=N — keep only every Nth photo, so a MIXED packet can be rendered on purpose.
  // Every packet ever produced has been all-photo (laos: 95 photos, 0 typographic) or
  // all-typographic (every fixture: 0 photos, N typographic). The MIXED case — some groups with a
  // photo, some without — has NEVER been rendered, and it is the realistic one: this loop already
  // degrades per-photo on a failed fetch (the `continue` below), so any zone where a few images
  // 404 produces exactly that layout in front of a reader.
  // It is also an I18N SURFACE nobody has seen: the fallback card renders t('no_photo'), which is
  // translated in all 11 catalogs (sin fotografía disponible / 沒有照片 / لا توجد صورة متاحة).
  // Deterministic by INDEX rather than random, so the same command always yields the same packet —
  // a fixture that differs run to run cannot be a gate.
  const every = Number(process.env.PHOTOS_EVERY || 0);
  let i = 0;
  const worker = async () => {
    while (i < groups.length) {
      const idx = i
      const g = groups[i++];
      if (!g.image_url) continue;
      // Skip every photo whose index is not a multiple of N — see PHOTOS_EVERY above.
      if (every > 1 && idx % every !== 0) continue;
      try {
        const r = await fetch(g.image_url, { signal: AbortSignal.timeout(20000) });
        if (!r.ok) continue;
        const buf = Buffer.from(await r.arrayBuffer());
        const mime = r.headers.get('content-type') || 'image/jpeg';
        out.set(g.slug, `data:${mime};base64,${buf.toString('base64')}`);
      } catch { /* a missing photo degrades to the typographic card */ }
    }
  };
  await Promise.all(Array.from({ length: Math.min(limit, groups.length) }, worker));
  return out;
}

// ── PAGE SHELL ──────────────────────────────────────────────────────────────
// Photo fetching is the only part that differs between the two callers (Node
// inlines base64 so the file is self-contained and printable offline; the browser
// hands the renderer plain URLs the page already has). Everything downstream is
// the shared renderer, so the two packets cannot drift.
async function buildPacket(zone, groups, mapSvg) {
  const photos = await fetchPhotos(groups);
  const html = R.buildPacket(zone, groups, photos, mapSvg);
  const withPhoto = groups.filter((g) => photos.has(g.slug)).length;
  // 1 map + N list pages + points + wheel — asked of the renderer, not recomputed
  // here, so this count cannot disagree with what was actually emitted.
  const pageCount = 1 + Math.ceil(groups.length / R.PER_PAGE) + 2;
  return { html, pageCount, withPhoto };
}

// ── MAIN ────────────────────────────────────────────────────────────────────
const records = await loadRecords();
const bySlug = new Map(records.map((r) => [r.slug, r]));
const zones = allZones().filter((z) => !ONLY || z.id === ONLY);
if (!zones.length) { console.error(`[packet] no zone matches ONLY=${ONLY}`); process.exit(1); }

let made = 0;
const mapless = [];       // zones whose page 1 rendered without its map
for (const z of zones) {
  let slugs = [];
  try {
    // evaluate() returns a SET, not an array — spread it rather than assuming.
    slugs = z.selector?.type === 'pg_slugs'
      ? (z.selector.values || [])
      : (z.selector ? [...evaluate(z.selector, records)] : []);
  } catch (e) { console.warn(`[packet] selector failed for ${z.id}: ${e.message}`); continue; }
  const groups = slugs.map((s) => bySlug.get(s)).filter(Boolean)
    .sort((a, b) => String(a.name).localeCompare(String(b.name), LANG));
  if (!groups.length) { console.warn(`[packet] ${z.id}: no groups resolved — skipped`); continue; }

  // The map one-pager reuses the poster's SVG when it has been generated; without
  // it the packet still builds (cover page without the map) rather than failing.
  const posterPath = join(__dirname, 'out', 'posters', `${z.id}.html`);
  let mapSvg = '';
  let mapMiss = null;                       // why page 1 has no map, if it has none
  if (!existsSync(posterPath)) {
    mapMiss = 'no out/posters/' + z.id + '.html';
  }
  if (existsSync(posterPath)) {
    const poster = readFileSync(posterPath, 'utf8');
    const m = poster.match(/<svg[\s\S]*?<\/svg>/);
    // ⚠ The map SVG carries almost NO presentation attributes — generate.mjs styles
    // it entirely from the poster document's <style> block (.ctry, .lbl, .lead, the
    // :root palette...). Lifting the <svg> alone leaves those rules behind, and SVG
    // `fill` then falls back to its initial value, which is BLACK. Measured: 349
    // <text>, 203 <path class="ctry"> and 95 <circle> painted solid black, with no
    // error and no broken image — the map was a black rectangle.
    // So carry the poster's stylesheet across too, MINUS its document-level rules
    // (html/body/.stage/@media print) which would fight the packet's own shell.
    // Taking the stylesheet rather than duplicating hex codes means a palette or
    // map-style change in generate.mjs follows the packet automatically.
    if (m) mapSvg = `<style>${mapCss(poster)}</style>
      <div style="position:absolute;left:0;top:250px;width:${W}px;
      transform:scale(0.62);transform-origin:top center">${m[0]}</div>`;
    // The poster EXISTS but carries no <svg> — a truncated or failed generate run. Same
    // outcome as a missing file (page 1 is blank) and it was equally silent, so it is
    // reported the same way rather than left as the one unexplained branch.
    else mapMiss = `out/posters/${z.id}.html has no <svg>`;
  }

  const { html, pageCount, withPhoto } = await buildPacket(z, groups, mapSvg);
  const name = LANG === 'en' ? `${z.id}.html` : `${z.id}.${LANG}.html`;
  // ★ STAMP HOW THIS PACKET WAS BUILT, INTO THE PACKET. out/packets/ is gitignored, so the
  // sweep gate (check:packets-all) grades whatever .html files happen to be sitting there —
  // an ACCIDENTAL set, not a current one. MEASURED 2026-09-24: doxa-maps-plugin-studio, which
  // shares main with this worktree, had two packets from a NOPHOTOS=1 probe days old and the
  // sweep was RED there, on artifacts nobody intended to ship. A fast layout-iteration build
  // and a shippable one were byte-indistinguishable: the packet carried lang and dir and
  // nothing else about its own provenance.
  // The GENERATOR stamps this rather than the renderer, because packet-render.js is shared
  // with the browser export path and must not know about CLI flags.
  // ⚠ NOT A TIMESTAMP. An mtime rule would fail a correct tree — measured in this very
  // session, packet-strings.json was 303s newer than the packets while being byte-identical
  // to HEAD, because a probe restored it from backup. Record the INTENT, which is a fact
  // about the build, not the clock.
  // `fixture` is a build fact like the others: a fixture packet is built from a committed
  // records subset with image_url stripped, so it legitimately has no photos and its
  // records are not today's live data. Without this the gate reports "every source record
  // carries an image_url, so this is a skipped or failed photo fetch" — a true sentence
  // about real records and a false one about this packet.
  // `mixedphotos` is the same kind of fact: PHOTOS_EVERY=N deliberately keeps only every Nth
  // photo, so SOME cards are typographic on purpose. Without the stamp the gate cannot tell a
  // deliberately mixed packet from a partly-failed photo fetch — and those deserve opposite
  // verdicts.
  const provenance = [process.env.PACKET_RECORDS ? 'fixture' : null,
    NOPHOTOS ? 'nophotos' : null,
    Number(process.env.PHOTOS_EVERY || 0) > 1 ? 'mixedphotos' : null,
    mapMiss ? 'nomap' : null]
    .filter(Boolean).join(' ') || 'full';
  const stamped = html.replace('<html ', `<html data-packet-build="${provenance}" `);
  writeFileSync(join(OUT, name), stamped);
  made++;
  if (mapMiss) mapless.push(z.id);
  // ⚠ DO NOT PRINT A BARE ✓ FOR A PACKET MISSING ITS FIRST PAGE. The map one-pager is the
  // first line of the packet spec, and the fallback above is deliberate — but it was also
  // SILENT: this printed "✓ laos  95 groups  8 pages" for a packet whose page 1 is empty,
  // with the page count still 8 because the blank sheet is still a sheet. Measured
  // 2026-09-24 in doxa-maps-plugin-studio, which has 0 generated posters: every packet
  // produced there was degraded this way and the only clue was the packet itself.
  // The empty-groups branch a few lines above already warns and skips; this branch has the
  // same shape and had no warning, which is the whole defect.
  console.log(`  ${mapMiss ? '⚠' : '✓'} ${z.id.padEnd(34)} ${String(groups.length).padStart(4)} groups  `
    + `${pageCount} pages  ${withPhoto} photos  ${(html.length / 1024 / 1024).toFixed(1)} MB`
    + (mapMiss ? `   NO MAP on page 1 (${mapMiss})` : ''));
}
console.log(`\n${made} packet(s) → ${OUT}${ONLY ? ` (ONLY=${ONLY})` : ''}${LANG !== 'en' ? ` [${LANG}]` : ''}`);
// Repeat it at the END too. A per-line ⚠ scrolls off in a 27-zone run; the summary is the
// part an operator actually reads, and this is the one fault that cannot be seen by
// counting pages.
if (mapless.length) {
  console.warn(`\n⚠ ${mapless.length} of ${made} packet(s) have NO MAP on page 1: ${mapless.join(', ')}`);
  console.warn('  The map one-pager is lifted from out/posters/<zone>.html, which `bun run posters`');
  console.warn('  generates. Without it the packet still builds — page 1 is simply blank.');
  console.warn('  Fix: bun run posters   (then re-run this script)');
}