A question's answer reaches the candidates (engine lifts articles half a tier) (#192)

ddiff --git a/server/api/query.js b/server/api/query.js
iindex cbadc2e..abf2d26 100644
---- a/server/api/query.js
++++ b/server/api/query.js
@@@ -29,7 +29,7 @@ const { AuthError } = require('../auth');
  const { OWNER_RE, TYPE_RE, ID_RE } = require('./documents');
  const { fuseRrf } = require('../fabric/rrf');
  const { createQueryEmbeddingCache } = require('../fabric/embed');
--const { docTier, isThin, dedupeKey, siteHosts } = require('../ranking');
++const { docTier, isThin, dedupeKey, siteHosts, isQuestion } = require('../ranking');
  
  const FACET_KEY_RE = /^[a-z][a-z0-9_]{0,39}$/;
  const LANG_RE = /^[a-zA-Z]{2,3}(-[A-Za-z0-9]{2,8})*$/;
@@@ -184,11 +184,7 @@ async function dealsOrder(fused, { match, store }) {
  //   - a title match (3) outranks a summary-only match (2), which outranks a body-only one (1);
  //   - a thin username-titled VOD ranks below a real title+summary inside the same tier.
  // The fused order is the final tie-break, so relevance still decides among equals.
--// A question ("how do I stream", "can I go live without OBS") wants an answer: inside its tier, an article (Help's
--// answers, a site's docs) comes before a recording or a post that only shares a word (2026-11-02: "how do I stream"
--// ranked four VODs titled "… Stream" above Help's "Can I stream without OBS?").
--const QUESTION_WORDS = new Set(['how', 'what', 'why', 'when', 'where', 'which', 'who', 'can', 'do', 'does', 'is', 'are', 'should', 'will']);
--const isQuestion = (match) => Boolean(match && match.words.length >= 2 && QUESTION_WORDS.has(match.words[0]));
++// A question wants an answer first inside its tier (ranking.js isQuestion; the engine lifts articles the same way).
  const isAnswer = (doc) => Boolean(doc && doc.type === 'article');
  
  async function rankOrder(fused, { match, store }) {
ddiff --git a/server/engine/pg.js b/server/engine/pg.js
iindex b5c3bc8..37fa878 100644
---- a/server/engine/pg.js
++++ b/server/engine/pg.js
@@@ -22,7 +22,7 @@
   * Nothing outside this file writes SQL against search_fts.
   */
  const { EXPOSURE } = require('../document');
--const { fold, words, siteOnlyQuery, siteUrlPattern } = require('../ranking');
++const { fold, words, siteOnlyQuery, siteUrlPattern, isQuestion } = require('../ranking');
  
  const CONFIG = 'simple';
  const WEIGHTS = '{0.1, 0.1, 0.4, 1.0}';   // D, C body, B summary, A title: ts_rank weights 1 / 4 / 10
@@@ -163,7 +163,10 @@ function createEngine(db, { freshness = { weight: 1, halfLifeDays: 30 } } = {})
          const relevance = `CASE WHEN ${boilerplate} THEN 0 ELSE ${base} END`;
          // Thin VODs sink below documents with a real title and summary inside the same tier; -2 is
          // larger than any base score (ts_rank × freshness < 2), so it decides before relevance does.
--        const rank = `-(${tier} * 100 + (${relevance}) + CASE WHEN ${thin} THEN -2 ELSE 0 END)`;
++        // A question lifts an article half a tier: above the recordings that share its tier, below the next tier up,
++        // so it is among the candidates the page ranks (ranking.js isQuestion).
++        const answer = isQuestion(match) ? `CASE WHEN d.type = 'article' THEN 50 ELSE 0 END` : '0';
++        const rank = `-(${tier} * 100 + ${answer} + (${relevance}) + CASE WHEN ${thin} THEN -2 ELSE 0 END)`;
          const facetNumber = (key) => `(SELECT x.value::float8 FROM doc_facets x
              WHERE x.rid = d.rid AND x.key = '${key}' AND x.value ~ '^[0-9]+([.][0-9]+)?$'
              ORDER BY x.value LIMIT 1)`;
ddiff --git a/server/ranking.js b/server/ranking.js
iindex c01fd6d..f261035 100644
---- a/server/ranking.js
++++ b/server/ranking.js
@@@ -158,7 +158,16 @@ function dedupeKey(doc) {
      return `${String((doc && doc.owner) || '').toLowerCase()}\u0000${title}`;
  }
  
++// A question ("how do I stream", "can I go live without OBS") wants an answer: an article (Help's answers, a site's
++// docs) comes before a recording or a post that only shares a word, inside its tier (2026-11-02: "how do I stream"
++// ranked hundreds of VODs titled "… Stream" above Help's "Can I stream without OBS?", which never reached the top 100).
++const QUESTION_WORDS = new Set(['how', 'what', 'why', 'when', 'where', 'which', 'who', 'can', 'do', 'does', 'is', 'are', 'should', 'will']);
++function isQuestion(match) {
++    return Boolean(match && !match.column && Array.isArray(match.words) && match.words.length >= 2 && QUESTION_WORDS.has(match.words[0]));
++}
++
  module.exports = {
++    isQuestion,
      SITE_ALIASES, fold, words, tokens, siteHosts, siteOnlyQuery, siteUrlPattern, urlBelongsToSites,
      textMatches, fieldTier, docTier, isThin, dedupeKey,
  };
ddiff --git a/test/question-candidates.test.js b/test/question-candidates.test.js
nnew file mode 100644
iindex 0000000..2fdff3e
---- /dev/null
++++ b/test/question-candidates.test.js
@@@ -0,0 +1,41 @@
++'use strict';
++/**
++ * A question's answer must reach the candidates: on production "how do I stream" matched hundreds of recent VODs titled
++ * "… Stream" in the title tier, and Help's older "Can I stream without OBS?" never made the top 100 the page ranks.
++ * The engine lifts an article half a tier on a question: above the recordings that share its tier, below the next
++ * tier up. A plain query is unchanged.
++ */
++const assert = require('assert');
++const { boot, request, doc, suite } = require('./helpers');
++
++const t = suite('question candidates');
++let svc;
++const put = (overrides) => svc.store.apply(doc({ summary: '', body: '', ...overrides }));
++const ids = (r) => r.body.results.map((x) => x.id);
++const q = (text) => request(svc.base, 'GET', `/api/v1/search?q=${encodeURIComponent(text)}&limit=10`);
++
++t('boot and seed', async () => {
++    svc = await boot();
++    const recent = new Date(Date.now() - 3600e3).toISOString();
++    const old = '2026-06-01T00:00:00Z';
++    for (let i = 1; i <= 8; i++) {
++        await put({ owner: 'live', type: 'vod', id: `vod-${i}`, title: `Late night stream number ${i}`, summary: 'A recorded stream with chat, music and talk about the week and the plans ahead.', canonical_url: `https://openvibe.video/v/${8100 + i}`, published_at: recent, updated_at: recent });
++    }
++    await put({ owner: 'help', type: 'article', id: 'help-obs', title: 'Can I stream without OBS?', summary: 'Yes. Go live from your browser on OpenVibe.Live: press Go Live and you are live in seconds, no software.', canonical_url: 'https://openvibe.help/a/openvibe.live/can-i-stream-without-obs', published_at: old, updated_at: old });
++    await put({ owner: 'wiki', type: 'article', id: 'wiki-body', title: 'Camera notes', summary: 'Lenses and light.', body: 'how to stream with a phone camera', canonical_url: 'https://openvibe.wiki/p/camera-notes', published_at: old, updated_at: old });
++});
++
++t('a question puts the answer first, even older than the recordings in its tier', async () => {
++    const got = ids(await q('how do I stream'));
++    assert.strictEqual(got[0], 'help-obs', `the answer first (${got})`);
++    assert.ok(got.indexOf('wiki-body') > got.indexOf('vod-1'), 'an article that only mentions it in its body stays below the title tier');
++});
++
++t('a plain query is unchanged: the recordings first, by freshness', async () => {
++    const got = ids(await q('stream'));
++    assert.ok(got.indexOf('help-obs') > 0, `not lifted without a question (${got})`);
++});
++
++t('close', async () => { await svc.stop(); });
++
++t.run();