bc-observatory 0.5.0 → 0.7.1

package/README.md

@@ -8,7 +8,7 @@ Microsoft 365 roadmap features, videos and community posts), every claim with a
 claude mcp add bc-observatory -- npx -y bc-observatory@latest
 ```
 
-Or the Claude Code plugin (MCP server plus skills): `claude plugin marketplace add waldo1001/waldo.BCObservatory`.
+Or the Claude Code plugin (MCP server plus skills): `claude plugin marketplace add waldo1001/waldo.BCObservatory`. The server never fetches Learn pages (D01); the plugin also connects Microsoft's Learn MCP (`https://learn.microsoft.com/api/mcp`), and a Learn hit's url is read there with `microsoft_docs_fetch`.
 
 Tools: `search`, `ls`, `cat`, `get_object`, `diff_object`, `localization`, `whats_new`, `blog_footprint`, `feedback`.
 
@@ -21,7 +21,7 @@ same order: references (`t36`, `cu 80`, `table 11300 BE`) and bare ids (`36`) go
 the title and an AL object's caption weigh 3, tags 2, summary 1, through the exact word, a prefix, the plural, a synonym
 (`client` is Customer, `G/L` is general ledger) or one typo; a name equal to the query adds 10; base app before
 first-party app before country layer, then what references, calls and documents an object; topic hubs by their size and
-review state, so "subscription" returns the Subscription billing hub before the codeunits that share the word. Fields,
+review state, so "subscription" returns the Subscription billing hub before the codeunits that share the word. A field, event or procedure ranks where its owner ranks (base application before first-party app before country layer, then importance), and its tooltip adds nothing when its name equals the query; a `kind:` query that finds nothing says `No event named "..."`, names the nearest symbols and is never padded with pages by meaning. A question (`how to ...`, `difference between A and B`, `A vs B`) drops its question words when a content word is left, a quoted phrase keeps all its words, and the hint names the hub of the best Learn article. Fields,
 events, global procedures and enum values are searchable too (`OnAfterPostSalesDoc`, `field: Posting Date`); their path
 ends in an anchor (`objects/codeunit/80#event-OnAfterPostSalesDoc`). Filters: `type`, `tier`, `system`, `country`,
 `app`, `object_type`, `kind`. Each result says where it sits (`in:` the Learn TOC path, the app, the channel) and, for a
@@ -35,7 +35,7 @@ plain JavaScript; the 30 MB model is downloaded once on the first search, pinned
 `BC_OBSERVATORY_MODEL_DIR=<dir with config.json, tokenizer.json, model.safetensors>` uses a model on disk (offline). Without a
 model, search is keyword-only and says so.
 
-`get_object` takes an id, a reference (`t36`, `36-be`), a name or a caption, and a `country`; `whats_new` lists nothing
+`get_object` takes an id, a reference (`t36`, `36-be`), a name or a caption, and a `country`; when that country's layer replaces a W1 object (`get_object("table", "18", "BE")`) the fields, events and procedures the layer adds print before the W1 page, and a lookup without a country ends with a `Replaced in:` line naming the layers; `search("field: Enterprise No.", country: "BE")` finds a member a layer adds. `whats_new` lists nothing
 dated after today unless `include_future: true`.
 
 `cat`, `get_object` and `localization` return a page as it is when it fits 30,000 characters. A longer page

package/dist/server.js

@@ -37,6 +37,7 @@ var KIND_OF = { fields: "field", events: "event", procs: "proc", values: "value"
 var fold = (s) => s.normalize("NFKD").replace(/[̀-ͯ]/g, "").toLowerCase();
 var TOKEN = /[\p{L}\p{N}]+(?:[.\-/][\p{L}\p{N}]+)*/gu;
 var keep = (t) => t.length > 1 || /\d/.test(t);
+var isCodedId = (w) => /^[A-Z]{2,3}\d{4}$/.test(w);
 function tokenize(s) {
   const out = [];
   for (const m of fold(s).matchAll(TOKEN)) if (keep(m[0])) out.push(m[0]);
@@ -50,14 +51,17 @@ function camelParts(raw) {
   if (!/[a-z][A-Z]|[A-Z]{2}[a-z]/.test(raw)) return [];
   return (raw.match(/[A-Z]+(?![a-z])|[A-Z]?[a-z]+|\d+/g) ?? []).map((w) => fold(w)).filter(keep);
 }
-function indexTokens(raw, camel = false) {
-  const out = /* @__PURE__ */ new Set();
+function seqTokens(raw, camel = false) {
+  const out = [];
   for (const t of tokenize(raw)) {
-    out.add(t);
-    for (const p of parts(t)) out.add(p);
+    out.push(t);
+    out.push(...parts(t));
   }
-  if (camel) for (const w of raw.split(/[^\p{L}\p{N}]+/u)) for (const p of camelParts(w)) out.add(p);
-  return [...out];
+  if (camel) for (const w of raw.split(/[^\p{L}\p{N}]+/u)) out.push(...camelParts(w));
+  return out;
+}
+function indexTokens(raw, camel = false) {
+  return [...new Set(seqTokens(raw, camel))];
 }
 function stem(t) {
   if (t.length < 5 || /\d/.test(t) || /[.\-/]/.test(t)) return t;
@@ -181,7 +185,53 @@ var SYSTEM_ALIASES = {
   development: ["al", "extension", "codeunit", "table", "page", "event", "interface", "namespace", "al language", "vs code", "analyzer", "compiler", "testability", "permission set"],
   platform: ["al-go", "pipeline", "docker", "container", "telemetry", "application insights", "performance", "database", "sql", "upgrade", "update", "release", "wave", "deprecation", "obsolete", "web client", "user experience"]
 };
-var QUESTION_WORDS = /* @__PURE__ */ new Set(["how", "what", "why", "where", "when", "which", "who", "do", "does", "did", "can", "could", "should", "is", "are", "was", "were", "the", "an", "to", "in", "of", "for", "on", "my", "me", "it", "from", "with", "and", "or", "changed", "change"]);
+var QUESTION_WORDS = /* @__PURE__ */ new Set([
+  "how",
+  "what",
+  "why",
+  "where",
+  "when",
+  "which",
+  "who",
+  "do",
+  "does",
+  "did",
+  "can",
+  "could",
+  "should",
+  "is",
+  "are",
+  "was",
+  "were",
+  "the",
+  "an",
+  "to",
+  "in",
+  "of",
+  "for",
+  "on",
+  "my",
+  "me",
+  "it",
+  "from",
+  "with",
+  "and",
+  "or",
+  "changed",
+  "change",
+  "difference",
+  "differences",
+  "between",
+  "vs",
+  "versus",
+  "compare",
+  "comparison",
+  "use",
+  "used",
+  "using",
+  "mean",
+  "means"
+]);
 var norm = (s) => s.toLowerCase().replace(/\.(?=\s|$)/g, "").trim();
 var table = null;
 function synonymsOf(word) {
@@ -223,11 +273,22 @@ function term(text2, phrase = false, raw = text2) {
   const alts = [...synonymsOf(text2)];
   const camel = camelParts(raw);
   if (camel.length > 1) alts.push(camel.join(" "));
-  return { text: text2, stem: phrase ? text2.split(" ").map(stem).join(" ") : stem(text2), alts, phrase, numeric: /^\d+$/.test(text2) };
-}
+  return { text: text2, stem: phrase ? text2.split(" ").map(stem).join(" ") : stem(text2), alts, phrase, numeric: /^\d+$/.test(text2), raw: raw || text2 };
+}
+var WH_START = /^(how|what|why|where|when|which|who)\b/i;
+var AUX_PRONOUN = /^(is|are|can|could|should|do|does)\s+(i|you|we|it|there|this|that|my|they|he|she)\b/i;
+var AUX_ARTICLE = /^(is|are|can|could|should|do|does)\s+(a|an|the)\b/i;
+var COMPARE_START = /^compare\s+\S+(?:\s+\S+)*?\s+(?:and|with|to|vs|versus)\s+\S/i;
+var VERSUS = /[\p{L}\p{N}]\S*\s+(?:vs|versus)(?![.\p{L}\p{N}])\s+(?!code\b)[\p{L}\p{N}]/iu;
+var USE_AFTER = /* @__PURE__ */ new Set(["to", "i", "we", "you", "it", "can", "should", "between", "and", "or"]);
+var isQuestion = (s) => {
+  if (WH_START.test(s)) return true;
+  if (!/^IS\b/.test(s) && (AUX_PRONOUN.test(s) || AUX_ARTICLE.test(s) && (s.split(/\s+/).length >= 5 || s.endsWith("?")))) return true;
+  return COMPARE_START.test(s) || /\bchanged in\b/i.test(s) || /\bdifferences? between\b/i.test(s) || VERSUS.test(s);
+};
 function parseQuery(raw, opts) {
   const countries = new Set(opts.countries.map((c) => c.toLowerCase()));
-  const q = { raw, ref: null, number: null, types: [], kind: null, major: null, countries: [], question: false, terms: [], warnings: [] };
+  const q = { raw, ref: null, number: null, types: [], kind: null, major: null, countries: [], question: false, dropped: [], shape: null, terms: [], warnings: [] };
   let s = raw.trim();
   const phrases = [];
   s = s.replace(/"([^"]*)"/g, (_, p) => {
@@ -242,7 +303,7 @@ function parseQuery(raw, opts) {
   }
   if (!q.kind && !phrases.length) {
     const m = /^([a-z]+)\s*(\d+)(?:(?:\s+|\s*-\s*)([a-z]{2}))?$/i.exec(s);
-    if (m) {
+    if (m && !(isCodedId(s.toUpperCase()) && !typeOfWord(m[1]))) {
       const type = typeOfWord(m[1]);
       const cc = m[3]?.toLowerCase() ?? null;
       if (type && (!cc || countries.has(cc))) {
@@ -262,7 +323,8 @@ function parseQuery(raw, opts) {
     q.major = v;
     return " ";
   }).trim();
-  q.question = /^(how|what|why|where|when|which)\b/i.test(s) || /\bchanged in\b/i.test(s);
+  q.question = !q.kind && isQuestion(s);
+  if (q.question) q.shape = /\bchanged in\b/i.test(s) ? "what-changed" : /^(how (do|to)|when to)\b/i.test(s) ? "how-to" : "concept";
   const words = s.split(/\s+/).filter(Boolean);
   const typeAt = (i) => i >= 0 && i < words.length && !!TYPE_WORDS[words[i].toLowerCase()];
   const typeWords = [];
@@ -291,12 +353,24 @@ function parseQuery(raw, opts) {
     const t = tokenize(w);
     for (const x of t) toks.push({ text: x, raw: t.length === 1 ? w.replace(/[^\p{L}\p{N}.\-/]/gu, "") : x });
   }
-  if (q.question) toks = toks.filter((t) => !QUESTION_WORDS.has(t.text) && !(t.text.length === 1));
+  if (q.question) {
+    const dropWord = (i) => {
+      const w = toks[i].text;
+      if (w === "vs" || w === "versus") return i > 0 && i < toks.length - 1 && !toks[i].raw.endsWith(".") && toks[i + 1].text !== "code";
+      if (w === "use" || w === "used" || w === "using") return USE_AFTER.has(toks[i - 1]?.text);
+      return QUESTION_WORDS.has(w);
+    };
+    const gone = new Set(toks.map((_, i) => i).filter((i) => dropWord(i) || toks[i].text.length === 1));
+    if (phrases.length || toks.some((_, i) => !gone.has(i))) {
+      for (const i of gone) if (QUESTION_WORDS.has(toks[i].text) && !q.dropped.includes(toks[i].text)) q.dropped.push(toks[i].text);
+      toks = toks.filter((_, i) => !gone.has(i));
+    }
+  }
   const multi = synonymPhrases().map((p) => p.split(" ")).sort((a, b) => b.length - a.length);
   for (let i = 0; i < toks.length; i++) {
     const hit = multi.find((p) => p.every((w, j) => toks[i + j]?.text === w));
     if (hit) {
-      q.terms.push(term(hit.join(" "), true));
+      q.terms.push({ ...term(hit.join(" "), true), raw: toks.slice(i, i + hit.length).map((t) => t.raw || t.text).join(" ") });
       i += hit.length - 1;
       continue;
     }
@@ -349,6 +423,8 @@ function pageToRecord(p) {
 function symbolToRecord(kind, row, owner) {
   const name = String(row[1]);
   const key = kind === "field" || kind === "value" ? Number(row[2]) : name;
+  const last = row[kind === "field" ? 5 : kind === "value" ? 4 : 6];
+  const cc = typeof last === "string" && /^[a-z]{2}$/i.test(last) ? last.toLowerCase() : null;
   const base = {
     id: `${owner.id}${anchorOf(kind, key)}`,
     kind,
@@ -357,9 +433,9 @@ function symbolToRecord(kind, row, owner) {
     title: name,
     objectType: owner.objectType,
     objectId: owner.objectId ?? null,
-    country: owner.country ?? null,
-    app: owner.app ?? null,
-    layer: owner.layer,
+    country: cc ?? owner.country ?? null,
+    app: cc ? `${cc.toUpperCase()} layer` : owner.app ?? null,
+    layer: cc ? 2 : owner.layer,
     importance: owner.importance,
     inbound: owner.inbound ?? 0,
     system: owner.system ?? null,
@@ -387,6 +463,7 @@ var push = (m, k, v) => {
   if (l) l.push(v);
   else m.set(k, [v]);
 };
+var INDEX_PAGE_IDS = 16;
 function normOf(r) {
   const cached = NORM.get(r);
   if (cached) return cached;
@@ -395,12 +472,21 @@ function normOf(r) {
   const own = r.type === "topic" ? new Set(tokenize(r.title)) : null;
   const learn = r.type === "learn";
   const tags = (r.tags ?? []).flatMap((t) => indexTokens(t, learn)).filter((t) => !own?.has(t));
-  const eqs = r.type === "object" || symbol ? [tokenize(r.name), ...r.caption ? [tokenize(r.caption)] : []] : [tokenize(r.title)];
+  const coded = learn ? [...new Set((r.tags ?? []).filter(isCodedId))] : [];
+  const eqs = r.type === "object" || symbol ? [tokenize(r.name), ...r.caption ? [tokenize(r.caption)] : []] : [tokenize(r.title), ...learn ? [...(r.title.match(/\b[A-Z]{2,3}\d{4}\b/g) ?? []).filter(isCodedId), ...coded.length <= INDEX_PAGE_IDS ? coded : []].map((t) => tokenize(t)).filter((t) => t.length === 1) : []];
   const n = { main: [...main], tags: learn ? tags : [...new Set(tags)], text: symbol ? null : indexTokens(r.text ?? ""), eqs };
   NORM.set(r, n);
   return n;
 }
 var textOf = (r, n) => n.text ??= indexTokens(r.text ?? "");
+function seqs(r) {
+  let main, text2;
+  const symbol = r.kind !== "page";
+  return {
+    main: () => main ??= [seqTokens(r.name, symbol), seqTokens(r.title, symbol), ...r.caption ? [seqTokens(r.caption)] : []].flatMap((f, k) => k ? ["", ...f] : f),
+    text: () => text2 ??= seqTokens(r.text ?? "")
+  };
+}
 function prepare(records, prev) {
   const index2 = prev ?? { records: [], postings: /* @__PURE__ */ new Map(), vocab: /* @__PURE__ */ new Map(), byType: /* @__PURE__ */ new Map(), maxInbound: /* @__PURE__ */ new Map() };
   let st = STATE.get(index2);
@@ -522,7 +608,7 @@ function matchersOf(index2, q) {
   }
   return m;
 }
-function best(m, toks) {
+function best(m, toks, seq = () => toks) {
   let b = null;
   for (const t of toks) {
     const f = m.words.get(t);
@@ -530,7 +616,8 @@ function best(m, toks) {
   }
   for (const p of m.phrases) {
     if (b && b.f >= p.f) continue;
-    const f = phraseForm(p.words, toks);
+    if (!p.words.every((w) => holds(toks, w))) continue;
+    const f = phraseForm(p.words, seq());
     if (f) {
       const v = p.f * f;
       if (!b || v > b.f) b = { f: v, how: p.how, ...p.via ? { via: p.via } : {} };
@@ -538,6 +625,21 @@ function best(m, toks) {
   }
   return b;
 }
+var STEMS = /* @__PURE__ */ new Map();
+var stemOf = (t) => {
+  let s = STEMS.get(t);
+  if (s === void 0) {
+    s = stem(t);
+    STEMS.set(t, s);
+  }
+  return s;
+};
+function holds(toks, w) {
+  if (toks.includes(w)) return true;
+  const sw = stemOf(w);
+  for (const t of toks) if (t[0] === w[0] && stemOf(t) === sw) return true;
+  return false;
+}
 function phraseForm(words, toks) {
   let bestF = 0;
   for (let i = 0; i + words.length <= toks.length; i++) {
@@ -551,7 +653,7 @@ function phraseForm(words, toks) {
   }
   return bestF;
 }
-function nameMatch(eq, ms, plural) {
+function nameMatch(eq, ms, plural2) {
   let pos = 0, f = 1, lastPrefix = false;
   for (const m of ms) {
     const ws = m.term.phrase ? m.term.text.split(" ") : null;
@@ -559,7 +661,7 @@ function nameMatch(eq, ms, plural) {
       for (const w of ws) {
         const t2 = eq[pos++];
         if (t2 === void 0) return { eq: 0, starts: 0 };
-        f = Math.min(f, t2 === w ? 1 : stem(t2) === stem(w) ? plural ? 1 : 0.75 : 0);
+        f = Math.min(f, t2 === w ? 1 : stem(t2) === stem(w) ? plural2 ? 1 : 0.75 : 0);
       }
       if (!f) return { eq: 0, starts: 0 };
       lastPrefix = false;
@@ -576,10 +678,10 @@ function nameMatch(eq, ms, plural) {
       lastPrefix = p.how === "prefix";
       continue;
     }
-    f = Math.min(f, form.how === "stem" && plural ? 1 : form.f);
+    f = Math.min(f, form.how === "stem" && plural2 ? 1 : form.f);
     lastPrefix = form.how === "prefix";
   }
-  if (pos === eq.length && !lastPrefix && (plural || f === 1)) return { eq: f, starts: 0 };
+  if (pos === eq.length && !lastPrefix && (plural2 || f === 1)) return { eq: f, starts: 0 };
   return { eq: 0, starts: pos <= eq.length ? f : 0 };
 }
 var NARRATIVE = { reviewed: 2, unreviewed: 0, none: -2, derived: -2 };
@@ -600,7 +702,7 @@ function scoreRecord(index2, r, q) {
     if (q.ref.country) return { s: country === q.ref.country ? 1e3 : country === null ? 900 : 850, why };
     return { s: country === null ? 1e3 : 900, why };
   }
-  const prior = object ? TYPE_PRIOR[r.objectType ?? ""] ?? 0 : 0;
+  const prior = objectish ? TYPE_PRIOR[r.objectType ?? ""] ?? 0 : 0;
   const lift = (base) => base + (object ? LAYER[r.layer] : symbol ? 0 : 1) + (objectish ? 2 * r.importance : 0) + prior;
   if (q.number !== null) {
     if (!object || r.objectId !== q.number) return none;
@@ -616,10 +718,10 @@ function scoreRecord(index2, r, q) {
     if (object && country && q.countries.includes(country)) return { s: 3 + 2 * r.importance + prior, why };
     return none;
   }
-  const n = normOf(r), ms = matchersOf(index2, q);
-  let s = 0, named = false, gate = !symbol || !!q.kind;
+  const n = normOf(r), ms = matchersOf(index2, q), sq = seqs(r);
+  let s = 0, textPts = 0, named = false, gate = !symbol || !!q.kind;
   for (const m of ms) {
-    const main = best(m, n.main), tags = best(m, n.tags), text2 = best(m, textOf(r, n));
+    const main = best(m, n.main, sq.main), tags = best(m, n.tags), text2 = best(m, textOf(r, n), sq.text);
     if (!main && !tags && !text2) return none;
     if (main || tags) named = true;
     if (symbol && !gate && n.eqs[0].some((t) => {
@@ -628,21 +730,25 @@ function scoreRecord(index2, r, q) {
     })) gate = true;
     if (symbol && /^on[a-z]/.test(m.term.text)) gate = true;
     s += 3 * (main?.f ?? 0) + 2 * (tags?.f ?? 0) + (text2?.f ?? 0);
+    textPts += text2?.f ?? 0;
     const via = [main, tags, text2].find((x) => x?.via && (x.how === "syn" || x.how === "fuzzy"))?.via;
     if (via && !why.includes(`matched "${via}"`)) why.push(`matched "${via}"`);
   }
   if (!gate) return none;
+  const learn = r.type === "learn";
   let eq = 0, starts = 0, byName = false;
   n.eqs.forEach((e, k) => {
     const x = nameMatch(e, ms, objectish);
     if (x.eq > eq) {
       eq = x.eq;
-      byName = k === 0;
+      byName = k === 0 || learn;
     }
-    starts = Math.max(starts, x.starts);
+    if (!learn || k === 0) starts = Math.max(starts, x.starts);
   });
+  const lifted = q.kind || q.types.length || /\d/.test(q.raw) || OBJECT_WORDS.test(q.raw) || object && eq && byName || symbol && ms.some((m) => /^on[a-z]/.test(m.term.text) && m.term.alts.some((a) => a.startsWith("on ")));
+  const owned = symbol && !!lifted;
   if (eq) {
-    s += 10 * eq;
+    s += 10 * eq - (owned ? textPts : 0);
     why.push(!byName ? "caption" : eq >= 0.75 ? "name" : "name~");
   } else if (starts) {
     s += 3 * starts;
@@ -652,8 +758,7 @@ function scoreRecord(index2, r, q) {
   if ((r.type === "topic" || r.type === "app") && !symbol) s += NARRATIVE[r.narrative ?? (r.type === "app" ? "none" : "unreviewed")] ?? 0;
   s += countryPoints(q, r, country);
   if (q.types.length && objectish) s += 3;
-  s = lift(s);
-  const lifted = q.kind || q.types.length || /\d/.test(q.raw) || OBJECT_WORDS.test(q.raw) || object && eq && byName || symbol && ms.some((m) => /^on[a-z]/.test(m.term.text) && m.term.alts.some((a) => a.startsWith("on ")));
+  s = symbol ? s + 2 * r.importance + (owned ? LAYER[r.layer] + prior : 0) : lift(s);
   if (objectish && !lifted) {
     s *= 0.6;
     why.push("demoted");
@@ -695,11 +800,11 @@ function search(index2, q, opts = {}) {
   return opts.limit ? hits.slice(0, opts.limit) : hits;
 }
 function symbolLine(r) {
-  const of = r.owner ? ` of ${r.owner.title}` : "", x = r.extra ?? {};
-  if (r.kind === "field") return `field ${x.id}${of} \xB7 ${x.type}`;
-  if (r.kind === "event") return `event${of} \xB7 ${x.kind} \xB7 ${x.subscribers} ${x.subscribers === 1 ? "subscriber" : "subscribers"}${x.obsolete ? ` \xB7 obsolete ${x.obsolete}` : ""}`;
-  if (r.kind === "proc") return `procedure${of} \xB7 ${x.params} ${x.params === 1 ? "parameter" : "parameters"}${x.returns ? ` \xB7 returns ${x.returns}` : ""}${x.obsolete ? ` \xB7 obsolete ${x.obsolete}` : ""}`;
-  if (r.kind === "value") return `value ${x.ordinal}${of}${x.caption && x.caption !== r.name ? ` \xB7 "${x.caption}"` : ""}`;
+  const of2 = `${r.owner ? ` of ${r.owner.title}` : ""}${r.country && r.app === `${r.country.toUpperCase()} layer` ? ` \xB7 ${r.app}` : ""}`, x = r.extra ?? {};
+  if (r.kind === "field") return `field ${x.id}${of2} \xB7 ${x.type}`;
+  if (r.kind === "event") return `event${of2} \xB7 ${x.kind} \xB7 ${x.subscribers} ${x.subscribers === 1 ? "subscriber" : "subscribers"}${x.obsolete ? ` \xB7 obsolete ${x.obsolete}` : ""}`;
+  if (r.kind === "proc") return `procedure${of2} \xB7 ${x.params} ${x.params === 1 ? "parameter" : "parameters"}${x.returns ? ` \xB7 returns ${x.returns}` : ""}${x.obsolete ? ` \xB7 obsolete ${x.obsolete}` : ""}`;
+  if (r.kind === "value") return `value ${x.ordinal}${of2}${x.caption && x.caption !== r.name ? ` \xB7 "${x.caption}"` : ""}`;
   return r.title;
 }
 function objectsAt(index2, type, id) {
@@ -720,7 +825,8 @@ function typoWords(index2, q) {
 function namedBy(index2, r, q) {
   if (!STATE.get(index2)) return 0;
   const n = normOf(r);
-  return matchersOf(index2, q).reduce((a, m) => a + (best(m, n.main)?.f ?? best(m, n.tags)?.f ?? 0), 0);
+  const sq = seqs(r);
+  return matchersOf(index2, q).reduce((a, m) => a + (best(m, n.main, sq.main)?.f ?? best(m, n.tags)?.f ?? 0), 0);
 }
 function symbolRecords(kind, rows, ownerOf) {
   const out = [];
@@ -736,7 +842,7 @@ function countriesOf(index2) {
 }
 
 // ../search/src/hints.ts
-function hints(index2, q, hits) {
+function hints(index2, q, hits, opts = {}) {
   const out = [];
   const unknown = q.warnings.map((w) => /^unknown abbreviation (\w+)$/.exec(w)?.[1]).find(Boolean);
   const digits = /(\d+)\s*$/.exec(q.raw)?.[1];
@@ -768,21 +874,89 @@ function hints(index2, q, hits) {
     out.push({ text: `${t.term}: no page; ${words.map((x) => `${x.w} (${x.n.toLocaleString("en")} ${x.n === 1 ? "page" : "pages"})`).join(", ")}`, query: swap(words[0].w) });
   }
   if (q.question) {
-    let hub = null, best2 = 0;
-    for (const i of index2.byType.get("topic") ?? []) {
-      const r = index2.records[i], named = namedBy(index2, r, q);
-      if (!named) continue;
-      const s = 3 * named + Math.log2((r.members ?? 0) + 1) + (r.narrative === "reviewed" ? 2 : r.narrative === "none" ? -2 : 0);
-      if (s > best2) {
-        best2 = s;
-        hub = r;
+    const hubOf = () => {
+      let hub = null, best2 = 0;
+      for (const i of index2.byType.get("topic") ?? []) {
+        const r = index2.records[i], named = namedBy(index2, r, q);
+        if (!named) continue;
+        const s = 3 * named + Math.log2((r.members ?? 0) + 1) + (r.narrative === "reviewed" ? 2 : r.narrative === "none" ? -2 : 0);
+        if (s > best2) {
+          best2 = s;
+          hub = r;
+        }
       }
+      return hub;
+    };
+    if (q.shape === "what-changed" || !q.shape) {
+      const hub = hubOf();
+      out.push({ text: `For what changed, open the object's Versions section or Changes; for how-to, start with the hub${hub ? `: ${hub.title}` : ""}.`, ...hub ? { path: hub.id } : {} });
+    } else {
+      const hub = learnHub(index2, hits) ?? hubOf();
+      if (hub) out.push({ text: `For ${q.shape === "how-to" ? "how-to" : "a concept"}, start with the hub: ${hub.title}.`, path: hub.id });
+    }
+  }
+  const kind = q.kind ?? opts.kind ?? null;
+  if (kind && !hits.length) {
+    const name = nameOf2(q);
+    if (opts.unfiltered?.length) {
+      const n = opts.unfiltered.length;
+      out.unshift({ text: `No ${KIND_WORD[kind]} named "${name}" matches the filters${opts.filters ? ` (${opts.filters})` : ""}; without them: ${n} ${n === 1 ? "hit" : "hits"}.` });
+    } else {
+      const h = nearestSymbols(index2, kind, name, opts.keep);
+      if (h) out.unshift(h);
     }
-    out.push({ text: `For what changed, open the object's Versions section or Changes; for how-to, start with the hub${hub ? `: ${hub.title}` : ""}.`, ...hub ? { path: hub.id } : {} });
   }
-  void hits;
   return out;
 }
+function learnHub(index2, hits) {
+  for (const h of hits.slice(0, 10)) {
+    const path = h.r.type === "learn" ? h.r.page?.hub : void 0;
+    if (!path) continue;
+    const hub = (index2.byType.get("topic") ?? []).map((i) => index2.records[i]).find((r) => r.id === path);
+    if (hub) return hub;
+  }
+  return null;
+}
+var KIND_WORD = { field: "field", event: "event", proc: "procedure", value: "value" };
+function nameOf2(q) {
+  const typed = q.terms.map((t) => t.raw || t.text).join(" ").replace(/"/g, "").trim();
+  return typed || (q.kind ? q.raw.replace(/^\s*\w+\s*:\s*/, "").replace(/"/g, "").trim() : "");
+}
+function nearestSymbols(index2, kind, name, keep2) {
+  const all = index2.byType.get(kind) ?? [];
+  const want = name.toLowerCase();
+  if (!all.length || !want) return null;
+  const word = KIND_WORD[kind], head1 = `No ${word} named "${name}".`;
+  const near = [];
+  if (want.length >= 4) {
+    const head = want.slice(0, 5), shared = Math.ceil(0.75 * want.length), reach = Math.min(2, Math.max(1, Math.floor(want.length / 4)));
+    for (const i of all) {
+      const r = index2.records[i], n = r.name.toLowerCase();
+      if (n === want) return null;
+      if (keep2 && !keep2(r)) continue;
+      let ok = editDistance(want, n, reach) <= reach;
+      if (!ok && n.startsWith(head)) {
+        let k = 0;
+        while (k < n.length && k < want.length && n[k] === want[k]) k++;
+        ok = k >= shared;
+      }
+      if (ok) near.push({ r, d: editDistance(want, n, 60) });
+    }
+  } else if (all.some((i) => index2.records[i].name.toLowerCase() === want)) return null;
+  near.sort((a, b) => a.d - b.d || b.r.importance - a.r.importance || a.r.name.localeCompare(b.r.name) || a.r.id.localeCompare(b.r.id));
+  const seen = /* @__PURE__ */ new Set(), pick = [];
+  for (const { r } of near) {
+    if (pick.length >= 3) break;
+    if (seen.has(r.id)) continue;
+    seen.add(r.id);
+    pick.push(r);
+  }
+  if (!pick.length) return { text: head1 };
+  const plural2 = word === "procedure" ? "procedures" : `${word}s`;
+  const label = (r) => r.owner ? `${r.name} (${r.owner.title})` : r.name;
+  const qk = kind === "proc" ? "proc" : kind;
+  return { text: `${head1} Nearest ${plural2}: ${pick.map(label).join(", ")}.`, query: `${qk}: ${pick[0].name}` };
+}
 
 // ../search/src/learn.ts
 function learnRecords(doc) {
@@ -912,6 +1086,139 @@ function dot(a, b, bOffset = 0) {
   return s;
 }
 
+// src/country.ts
+var BLOCK_MEMBERS = 40;
+var DELTA_CLIP = 140;
+var KINDS2 = ["fields", "values", "events", "procedures", "keys"];
+var NOUN = { fields: ["field", "fields"], values: ["value", "values"], events: ["event", "events"], procedures: ["procedure", "procedures"], keys: ["key", "keys"] };
+var plural = (n, kind) => `${n} ${NOUN[kind][n === 1 ? 0 : 1]}`;
+var of = (o, kind, change) => (o[kind] ?? []).filter((m) => m.change === change);
+var clip = (s, n = DELTA_CLIP) => s.length > n ? `${s.slice(0, n - 1)}\u2026` : s;
+function titleOfKey(key, name) {
+  const [type, id] = key.split("/");
+  return `${type.charAt(0).toUpperCase()}${type.slice(1)} ${id} "${name}"`;
+}
+function addedSummary(o) {
+  return KINDS2.filter((k) => k !== "keys").map((k) => [k, of(o, k, "added").length]).filter(([, n]) => n > 0).map(([k, n]) => `+${plural(n, k)}`).join(", ");
+}
+function replacedInLine(layers) {
+  if (!layers.length) return null;
+  const parts2 = [...layers].sort((a, b) => a.cc.localeCompare(b.cc)).map(({ cc, object }) => {
+    const s = addedSummary(object);
+    return `${cc.toUpperCase()}${s ? ` (${s})` : ""}`;
+  });
+  return `Replaced in: ${parts2.join(", ")}.`;
+}
+var paramsOf = (ps) => (ps ?? []).map((p) => `${p.var ? "var " : ""}${p.name}: ${p.type}`).join("; ");
+var signature = (name, to) => `${name}(${paramsOf(to.params)})${to.returns ? `: ${to.returns}` : ""}`;
+var procLine = (m) => {
+  const to = m.to ?? {};
+  return `${signature(m.name, to)} (${to.event && to.event !== "subscriber" ? to.event : to.scope ?? "global"})${to.obsolete?.state ? ` [obsolete ${to.obsolete.state}]` : ""}`;
+};
+var fieldLine = (m) => `${m.id} "${m.name}" ${(m.to ?? {}).type ?? ""}`.trimEnd();
+function propertiesText(p) {
+  return Object.entries(p ?? {}).map(([k, v]) => v === "true" ? k : /^(Caption|ToolTip)/.test(k) ? `${k} "${v}"` : `${k} ${v}`).join("; ");
+}
+function deltaText(d) {
+  if (!d) return "";
+  const show = (v) => typeof v === "string" ? v : JSON.stringify(v);
+  const parts2 = [];
+  for (const [k, v] of Object.entries(d)) {
+    if (k === "properties") {
+      for (const [pk, pv] of Object.entries(v)) parts2.push(`${pk}: ${pv[0] ?? "-"} -> ${pv[1] ?? "-"}`);
+      continue;
+    }
+    const [a, b] = v;
+    if (Array.isArray(a) && Array.isArray(b)) {
+      const add = b.filter((x) => !a.includes(x)), rem = a.filter((x) => !b.includes(x));
+      parts2.push(`${k} ${[...add.map((x) => `+${x}`), ...rem.map((x) => `-${x}`)].join(", ") || "reordered"}`);
+    } else parts2.push(`${k}: ${show(a ?? "-")} -> ${show(b ?? "-")}`);
+  }
+  return clip(parts2.join("; "));
+}
+function keyLine(m) {
+  if (m.change === "added") return `${m.name} (new: ${((m.to ?? {}).fields ?? []).join(", ")})`;
+  if (m.change === "removed") return `${m.name} (removed)`;
+  return `${m.name} (${deltaText(m.delta) || "changed"})`.replace(/\(fields /, "(");
+}
+function headline(o, cc, major, title) {
+  const added = addedSummary(o);
+  const other = [];
+  for (const k of KINDS2) {
+    const changed = of(o, k, "changed").length + (k === "keys" ? of(o, k, "added").length : 0), removed = of(o, k, "removed").length;
+    if (changed) other.push(`${plural(changed, k)} changed`);
+    if (removed) other.push(`${plural(removed, k)} removed`);
+  }
+  if (o.properties?.length) other.push(`${o.properties.length} ${o.properties.length === 1 ? "property" : "properties"} changed`);
+  const tail = [added || "no member added", ...other.length ? [other.join(", ")] : []].join("; ");
+  return `${cc.toUpperCase()} replaces ${title} (BC${major}): ${tail}.`;
+}
+function countryBlock(o, cc, major, opts = {}) {
+  const C = cc.toUpperCase(), title = opts.title ?? titleOfKey(o.key, o.name);
+  const groups = [
+    ["Fields added", of(o, "fields", "added").map(fieldLine)],
+    ["Values added", of(o, "values", "added").map((m) => `${m.id} ${m.name}`)],
+    ["Events added", of(o, "events", "added").map(procLine)],
+    ["Procedures added", of(o, "procedures", "added").map(procLine)],
+    ["Keys changed", [...of(o, "keys", "changed"), ...of(o, "keys", "added"), ...of(o, "keys", "removed")].map(keyLine)],
+    ["Fields changed", of(o, "fields", "changed").map((m) => `${m.id} "${m.name}"${m.delta ? ` (${deltaText(m.delta)})` : ""}`)],
+    ["Events changed", of(o, "events", "changed").map((m) => `${m.name}${m.delta ? ` (${deltaText(m.delta)})` : ""}`)],
+    ["Procedures changed", of(o, "procedures", "changed").map((m) => `${m.name}${m.delta ? ` (${deltaText(m.delta)})` : ""}`)]
+  ];
+  const out = [headline(o, cc, major, title)];
+  let room = BLOCK_MEMBERS, more = 0;
+  for (const [label, items] of groups) {
+    if (!items.length) continue;
+    const shown = items.slice(0, Math.max(0, room));
+    room -= shown.length;
+    more += items.length - shown.length;
+    if (shown.length) out.push(`${label}: ${shown.join("; ")}`);
+  }
+  if (more) out.push(`... and ${more} more: localization("${C}", section: "w1_objects_this_country_changes")`);
+  out.push(`Source: data/code/diffs/country/${major}-${cc.toLowerCase()}.json; the ${C} page: localizations/${cc.toLowerCase()}.`);
+  return out.join("\n");
+}
+function parseAnchor(anchor) {
+  const m = /^#?(field|value|event|proc)-(.+)$/.exec(anchor);
+  return m ? { kind: m[1], key: m[2] } : null;
+}
+function memberAnswer(o, a, cc, opts) {
+  const C = cc.toUpperCase(), title = opts.title ?? titleOfKey(o.key, o.name);
+  const head = (what) => `${title}, ${what} (added by the ${C} layer, not in W1):`;
+  const foot = `W1 page: ${opts.site}${opts.path}/ \xB7 ${C} layer: localizations/${cc.toLowerCase()}`;
+  if (a.kind === "field") {
+    const m = of(o, "fields", "added").find((x) => x.id === a.key);
+    if (!m) return null;
+    const to = m.to ?? {};
+    return `${head(`field ${m.id}`)}
+| No. | Name | Type | Properties | Layer |
+|---|---|---|---|---|
+| ${m.id} | ${m.name} | ${to.type ?? ""} | ${propertiesText(to.properties)} | ${C} |
+
+${foot}
+`;
+  }
+  if (a.kind === "value") {
+    const m = of(o, "values", "added").find((x) => x.id === a.key);
+    if (!m) return null;
+    return `${head(`value ${m.id}`)}
+| Ordinal | Name | Layer |
+|---|---|---|
+| ${m.id} | ${m.name} | ${C} |
+
+${foot}
+`;
+  }
+  const list = a.kind === "event" ? of(o, "events", "added") : of(o, "procedures", "added");
+  const hits = list.filter((x) => x.name === a.key);
+  if (!hits.length) return null;
+  return `${head(`${a.kind === "event" ? "event" : "procedure"} ${a.key}`)}
+${hits.map((m) => `- \`${signature(m.name, m.to ?? {})}\` (${(m.to ?? {}).event ?? (m.to ?? {}).scope ?? "global"})`).join("\n")}
+
+${foot}
+`;
+}
+
 // src/sections.ts
 var DEFAULT_BUDGET = 3e4;
 var MIN_BUDGET = 5e3;
@@ -1089,7 +1396,7 @@ function findSection(p, want) {
 }
 var keyList = (p, counts = false) => [...p.sections.map((s) => counts && s.rows.length ? `${s.key} (${s.rows.length})` : s.key), "frontmatter"].join(", ");
 var titleOf = (p, path) => p.preamble.split("\n").find((l) => l.startsWith("# "))?.slice(2).trim() || path;
-function clip(s, n, mark = "\u2026") {
+function clip2(s, n, mark = "\u2026") {
   if (s.length <= n) return s;
   const end = /[\uD800-\uDBFF]/.test(s[n - 2]) ? n - 2 : n - 1;
   return `${s.slice(0, end)}${mark}`;
@@ -1152,7 +1459,7 @@ function compactView(md, path, budget) {
     const more = `\u2026 (section: "${s.key}")`;
     if (!s.rows.length) {
       const text2 = s.head.filter((l) => !blank(l)).join("\n");
-      if (text2) out.push(clip(text2, PROSE_CLIP, more));
+      if (text2) out.push(clip2(text2, PROSE_CLIP, more));
     } else {
       const lead = s.head.find((l) => !blank(l) && !l.startsWith("|") && !l.startsWith("### "));
       if (lead && lead.length <= HEAD_LINE_MAX) out.push(lead);
@@ -1165,12 +1472,12 @@ function compactView(md, path, budget) {
         const t = r.table ? r.table.join("\n") : null;
         if (r.table && t !== table2) out.push(...r.table);
         table2 = t;
-        out.push(clip(r.lines[0], ROW_CLIP));
+        out.push(clip2(r.lines[0], ROW_CLIP));
       }
       if (s.rows.length > PREVIEW_ROWS) out.push(`\u2192 ${s.rows.length - PREVIEW_ROWS} more of ${s.rows.length}: cat("${path}", section: "${s.key}", offset: ${PREVIEW_ROWS})`);
       else {
         const text2 = s.tail.filter((l) => !blank(l)).join("\n");
-        if (text2) out.push(clip(text2, PROSE_CLIP, more));
+        if (text2) out.push(clip2(text2, PROSE_CLIP, more));
       }
     }
     out.push("");
@@ -1264,7 +1571,7 @@ ${pageView(md, clean, {}, budget, site)}`;
 }
 
 // src/server.ts
-var VERSION = "0.5.0";
+var VERSION = "0.7.1";
 var SITE = (process.env.BC_OBSERVATORY_SITE ?? "https://waldo1001.github.io/waldo.BCObservatory/").replace(/\/?$/, "/");
 var LOCAL = process.env.BC_OBSERVATORY_LOCAL ? resolve(process.env.BC_OBSERVATORY_LOCAL) : null;
 var BUDGET = budgetOf(process.env.BC_OBSERVATORY_MAX_CHARS);
@@ -1293,6 +1600,7 @@ async function pageMarkdown(path) {
 var pages = [];
 var index = null;
 var loadedAt = 0;
+var symbolsMajor = null;
 async function shard(file, sha) {
   const cached = join(cacheDir, `${sha}.json`);
   if (!LOCAL && existsSync(cached)) return readFileSync(cached, "utf8");
@@ -1320,6 +1628,7 @@ async function loadIndex() {
   } catch {
     sm = null;
   }
+  symbolsMajor = sm?.major ?? null;
   if (sm) {
     const byPath = new Map(idx.records.map((r) => [r.id, r]));
     for (const [k, e] of Object.entries(sm.kinds)) {
@@ -1332,7 +1641,37 @@ async function loadIndex() {
   }
   index = idx;
   loadedAt = Date.now();
+  countryDiffs.clear();
+}
+var countryDiffs = /* @__PURE__ */ new Map();
+function diffMajor() {
+  if (symbolsMajor) return symbolsMajor;
+  let best2 = 0;
+  for (const r of index?.records ?? []) for (const part of (r.major ?? "").split(/[\s,]+/)) {
+    const n = Number(part.split("-").pop());
+    if (n > best2) best2 = n;
+  }
+  return best2 ? String(best2) : null;
 }
+function replacedObjects(major, cc) {
+  const key = `${major}-${cc}`, hit = countryDiffs.get(key);
+  if (hit && Date.now() - hit.at < DAY) return Promise.resolve(hit.objects);
+  const loading = (async () => {
+    try {
+      const diff = JSON.parse(await dataFile(`code/diffs/country/${key}.json`));
+      return new Map(diff.objects.filter((o) => o.change === "replaced").map((o) => [o.key, o]));
+    } catch {
+      return null;
+    }
+  })();
+  countryDiffs.set(key, { at: Date.now(), objects: loading });
+  void loading.then((o) => {
+    const e = countryDiffs.get(key);
+    if (e) e.objects = o;
+  });
+  return loading;
+}
+var replacedBy = async (major, cc, key) => (await replacedObjects(major, cc))?.get(key) ?? null;
 var url = (path) => `${SITE}${path}/`;
 var line = (r, tag = "") => `- ${r.title}${r.caption ? ` (captioned "${r.caption}")` : ""} [${r.type}${r.tier ? `, ${r.tier}` : ""}${r.date ? `, ${r.date}` : ""}] path=${r.path}${tag}${r.path_label ? `
   in: ${r.path_label}` : ""}${r.stats ? `
@@ -1396,7 +1735,7 @@ function fuse(lists, k = 60) {
 }
 function keepOf(a) {
   const lc = (x) => x?.trim().toLowerCase() || void 0;
-  const type = lc(a.type), tier = lc(a.tier), country = lc(a.country), app = lc(a.app), kind = lc(a.kind);
+  const type = lc(a.type), tier = lc(a.tier), country = lc(a.country), app = lc(a.app), kind = KINDS[lc(a.kind) ?? ""] ?? lc(a.kind);
   const system = a.system ? systemOf(a.system) ?? lc(a.system) : void 0;
   const objectType = a.object_type ? typeOfWord(a.object_type.replace(/\s+/g, "")) ?? lc(a.object_type) : void 0;
   return (r) => (!type || r.type === type) && (!tier || r.tier === tier) && (!system || r.system === system) && (!country || (r.country ?? "") === country) && (!app || (r.app ?? "").toLowerCase() === app) && (!objectType || r.objectType === objectType) && (!kind || r.kind === kind);
@@ -1407,9 +1746,11 @@ async function toolSearch(a) {
   const limit = a.limit ?? 10, mode = a.mode ?? "hybrid", POOL = 50;
   const q = parseQuery(a.query, { countries: countriesOf(index) });
   const hits = search(index, q, { filter: keep2 });
-  const byId = new Map(hits.map((h) => [h.r.id, h.r]));
+  const kind = q.kind ?? KINDS[a.kind?.trim().toLowerCase() ?? ""] ?? null;
+  const uid = (r) => r.kind !== "page" && r.country ? `${r.id} ${r.country}` : r.id;
+  const byId = new Map(hits.map((h) => [uid(h.r), h.r]));
   let semantic = [], note = "";
-  if (mode !== "keyword") {
+  if (mode !== "keyword" && !kind) {
     const m = await loadModel();
     if (m) {
       const v = embed(m, a.query), data = pageVectors(m);
@@ -1426,27 +1767,36 @@ async function toolSearch(a) {
       }
     } else if (mode === "semantic") note = ` (semantic search unavailable: ${modelError}; keyword results instead)`;
   }
-  const keyword = hits.map((h) => h.r.id);
+  const keyword = hits.map((h) => uid(h.r));
   let order, exactBand = 0;
   if (mode === "semantic") order = semantic.length ? semantic : keyword.slice(0, POOL);
   else if (mode === "keyword" || !semantic.length) order = keyword;
   else {
-    const exact = hits.filter((h) => h.band === "exact").map((h) => h.r.id), pinned = new Set(exact);
+    const exact = hits.filter((h) => h.band === "exact").map((h) => uid(h.r)), pinned = new Set(exact);
     exactBand = exact.length;
-    order = [...exact, ...fuse([hits.filter((h) => h.band !== "exact").slice(0, POOL).map((h) => h.r.id), semantic.filter((p) => !pinned.has(p))])];
+    order = [...exact, ...fuse([hits.filter((h) => h.band !== "exact").slice(0, POOL).map((h) => uid(h.r)), semantic.filter((p) => !pinned.has(p))])];
   }
   if (mode !== "semantic" && mode !== "keyword" && !semantic.length) exactBand = hits.filter((h) => h.band === "exact").length;
   const inKeyword = new Set(keyword.slice(0, Math.max(POOL, limit))), inMeaning = new Set(semantic);
   const tag = (id) => inKeyword.has(id) && inMeaning.has(id) ? " [both]" : inMeaning.has(id) ? " [meaning]" : " [keyword]";
   const shown = order.slice(0, limit).map((id) => byId.get(id)).filter(Boolean);
-  const hs = hints(index, q, hits).filter((h) => !hits.length || hits[0].s < 6 || q.question || !!q.ref || /: no page; /.test(h.text));
+  const filters = ["type", "tier", "system", "country", "app", "object_type"].filter((k) => a[k]?.trim()).map((k) => `${k}=${a[k].trim()}`).join(", ");
+  const unfiltered = kind && !hits.length && filters ? search(index, q, { filter: (r) => r.kind === kind }) : void 0;
+  const hs = hints(index, q, hits, { kind, unfiltered, filters, keep: keep2 }).filter((h) => !hits.length || hits[0].s < 6 || q.question || !!q.ref || /: no page; /.test(h.text));
   const hintText = hs.length ? `
 ${hs.map((h) => `Hint: ${h.text}${h.query ? ` (search "${h.query}")` : h.path ? ` (cat("${h.path}"))` : ""}`).join("\n")}` : "";
+  if (!shown.length && kind && /^No \w+ named /.test(hs[0]?.text ?? "")) {
+    const [first, ...more] = hs, via = (h) => h.query ? ` (search "${h.query}")` : h.path ? ` (cat("${h.path}"))` : "";
+    return `${first.text}${via(first)}${note}${more.map((h) => `
+Hint: ${h.text}${via(h)}`).join("")}`;
+  }
   if (!shown.length) return `No pages match "${a.query}". Try fewer words, or ls("") to browse sections.${note}${hintText}`;
   const how = semantic.length && keyword.length && mode === "hybrid" ? "keyword + meaning" : mode === "semantic" && semantic.length ? "meaning" : "keyword";
-  const learnNote = shown.some((r) => r.type === "learn") ? "; a Learn article's url opens on learn.microsoft.com, cat(hub) reads its hub" : "";
-  return `${shown.length} results, ${how}${note}${how === "keyword + meaning" || mode === "hybrid" ? ` (exact band: ${exactBand})` : ""} (read one with cat(path)${learnNote}):
-${shown.map((r) => hitLine(r, tag(r.id))).join("\n")}${hintText}`;
+  const learnNote = shown.some((r) => r.type === "learn") ? "; a Learn article's url reads with the Learn MCP's microsoft_docs_fetch (the plugin connects it), else open it on learn.microsoft.com; cat(hub) reads its hub" : "";
+  if (mode === "hybrid" && !keyword.length && semantic.length) return `No keyword match for "${a.query}"; ${shown.length} pages by meaning only (read one with cat(path)${learnNote}):
+${shown.map((r) => hitLine(r, tag(uid(r)))).join("\n")}${hintText}`;
+  return `${shown.length} results, ${how}${note}${how === "keyword + meaning" ? ` (exact band: ${exactBand})` : ""} (read one with cat(path)${learnNote}):
+${shown.map((r) => hitLine(r, tag(uid(r)))).join("\n")}${hintText}`;
 }
 async function toolLs(a) {
   await loadIndex();
@@ -1468,7 +1818,27 @@ async function toolLs(a) {
 async function toolCat(a) {
   const [base, anchor] = a.path.split(/(?=#)/);
   const clean = `${base.replace(/^\/+|\/+$/g, "").replace(/\.md$/, "")}${anchor ?? ""}`;
-  return pageView(await pageMarkdown(clean.replace(/#.*$/, "")), clean, a, BUDGET, SITE);
+  const path = clean.replace(/#.*$/, ""), md = await pageMarkdown(path);
+  if (anchor && anchorView(md, path, anchor, SITE) === null) {
+    const added = await layerMember(path, anchor);
+    if (added) return added;
+  }
+  return pageView(md, clean, a, BUDGET, SITE);
+}
+async function layerMember(path, anchor) {
+  const want = parseAnchor(anchor), major = diffMajor();
+  if (!want || !major) return null;
+  await loadIndex();
+  const id = `${path}${anchor}`;
+  const layers = [...new Set(index.records.filter((r) => r.kind !== "page" && r.id === id && r.country && r.app === `${r.country.toUpperCase()} layer`).map((r) => r.country))].sort();
+  const owner = index.records.find((r) => r.kind === "page" && r.id === path);
+  const out = [];
+  for (const cc of layers) {
+    const o = owner?.objectId != null ? await replacedBy(major, cc, `${owner.objectType}/${owner.objectId}`) : null;
+    const text2 = o && memberAnswer(o, want, cc, { title: owner?.title, site: SITE, path });
+    if (text2) out.push(text2);
+  }
+  return out.length ? out.join("\n") : null;
 }
 async function toolGetObject(a) {
   await loadIndex();
@@ -1503,12 +1873,24 @@ async function toolGetObject(a) {
     const pick = country ? inCountry.length ? inCountry : w1 : w1.length ? w1 : found;
     if (pick.length === 1) {
       const twins = found.filter((r) => r !== pick[0] && r.country).map((r) => r.country.toUpperCase());
-      const note = country && !inCountry.length ? `(${country.toUpperCase()} has no ${type} of its own here; the W1 ${type} applies.)
+      const major = diffMajor(), key = `${type}/${pick[0].objectId ?? pick[0].name.toLowerCase()}`;
+      let note = "";
+      if (country && !inCountry.length) {
+        const replaced = major ? await replacedBy(major, country, key) : null;
+        note = replaced && major ? `${countryBlock(replaced, country, major, { title: pick[0].title })}
 
-` : "";
-      const also = !country && twins.length ? `
+` : `(${country.toUpperCase()} has no ${type} of its own here; the W1 ${type} applies.)
+
+`;
+      }
+      let also = !country && twins.length ? `
 
 (Also in ${twins.join(", ")}: get_object("${type}", "${pick[0].objectId ?? pick[0].name}", country).)` : "";
+      if (!country && !pick[0].country && major && !a.section) {
+        const layers = (await Promise.all(countriesOf(index).map(async (cc2) => ({ cc: cc2, object: await replacedBy(major, cc2, key) })))).filter((l) => !!l.object);
+        const line2 = replacedInLine(layers);
+        if (line2) also += `${also ? "\n" : "\n\n"}${line2}`;
+      }
       return `${note}${pageView(await pageMarkdown(pick[0].id), pick[0].id, a, BUDGET, SITE)}${a.section ? "" : also}`;
     }
     if (pick.length > 1) return `${pick.length} ${type} objects match "${a.idOrName}"; read one with cat(path):
@@ -1566,13 +1948,13 @@ ${a.message}
 var VIEW = { section: z.string().optional(), offset: z.number().int().min(0).max(1e5).optional(), limit: z.number().int().min(1).max(5e3).optional() };
 function createServer() {
   const server = new McpServer({ name: "bc-observatory", version: VERSION }, {
-    instructions: "BC Observatory: an agent-first knowledge base of Microsoft Dynamics 365 Business Central (Learn hubs, AL objects from the code for BC28-30, localizations, roadmap features, videos, community posts, code changes: merged pull requests of Microsoft's Business Central repositories, BCApps joined to the AL objects they changed, under changes/, and AL Language extension releases (the marketplace changelog, under releases/)). Every page carries a trust tier (official = Microsoft, community = everyone else) and a review state: say which tier a claim comes from. Never invent AL object ids or version numbers: look them up with get_object. Start with search(), read pages with cat(path), browse with ls(path). search() also finds Microsoft Learn articles by title, summary and the identifiers they document (type learn): those are not pages here, open their url on learn.microsoft.com."
+    instructions: "BC Observatory: an agent-first knowledge base of Microsoft Dynamics 365 Business Central (Learn hubs, AL objects from the code for BC28-30, localizations, roadmap features, videos, community posts, code changes: merged pull requests of Microsoft's Business Central repositories, BCApps joined to the AL objects they changed, under changes/, and AL Language extension releases (the marketplace changelog, under releases/)). Every page carries a trust tier (official = Microsoft, community = everyone else) and a review state: say which tier a claim comes from. Never invent AL object ids or version numbers: look them up with get_object. Start with search(), read pages with cat(path), browse with ls(path). search() also finds Microsoft Learn articles by title, summary and the identifiers they document (type learn): those are not pages here: read their url with the Learn MCP's microsoft_docs_fetch (the plugin connects it), else open it on learn.microsoft.com."
   });
   server.registerTool(
     "search",
     {
       title: "Search the knowledge base",
-      description: "Search every page, and the fields, published events, global procedures and enum values of AL objects, ranked like the site's search. Takes names ('Sales Header'), references ('t36', 'cu 80', 'table 11300 BE'), bare ids ('36') and symbol prefixes ('event: OnAfterPost', 'field: Posting Date', 'proc: CopyToTempLines'); a symbol hit's path ends in its anchor (#event-<Name>, #field-<id>, #proc-<Name>, #value-<ordinal>); cat(path) reads that member alone. Learn articles too: type learn, a url, not a path. Hybrid by default: exact matches stay on top and the rest is fused with a meaning ranking; each hit says [keyword], [meaning] or [both]; mode 'keyword' or 'semantic' for one list. Filters: type (topic, learn, app, feature, object, localization, video, post, change, source, digest, release), tier (official, community), system (finance, sales, ... or an alias such as 'g/l'), country ('BE'), app ('Subscription Billing'), object_type ('table'), kind (page, field, event, proc, value).",
+      description: "Search every page and the fields, events, global procedures and enum values of AL objects. Takes names ('Sales Header'), references ('t36', 'cu 80'), bare ids ('36') and symbol prefixes ('event: OnAfterPost', 'proc: CopyToTempLines'); symbols rank by their owner's layer and importance; a kind: query with no hit names the nearest symbols; a symbol hit's path ends in its anchor (#event-<Name>, #field-<id>, #proc-<Name>, #value-<ordinal>); cat(path) reads that member. Learn articles: type learn, a url not a path. Hybrid by default: exact matches on top, the rest fused with a meaning ranking; each hit says [keyword], [meaning] or [both]; mode 'keyword' or 'semantic' for one list. Filters: type (topic, learn, app, feature, object, localization, video, post, change, source, digest, release), tier (official, community), system (finance, sales, ... or alias 'g/l'), country ('BE'), app ('Subscription Billing'), object_type ('table'), kind (page, field, event, proc, value).",
       inputSchema: { query: z.string(), type: z.string().optional(), tier: z.string().optional(), system: z.string().optional(), country: z.string().optional(), app: z.string().optional(), object_type: z.string().optional(), kind: z.enum(["page", "field", "event", "proc", "value"]).optional(), limit: z.number().int().min(1).max(50).optional(), mode: z.enum(["hybrid", "keyword", "semantic"]).optional() }
     },
     async (a) => text(await toolSearch(a))
@@ -1589,7 +1971,7 @@ function createServer() {
   );
   server.registerTool("get_object", {
     title: "Look up an AL object",
-    description: `An AL object of W1, a Microsoft first-party app or a country layer by type and id, reference, name or caption: ('table', '18'), ('table', 't36'), ('table', '36-be'), ('codeunit', 'Sales-Post'), ('page', 'Customers'), or with country 'BE'. Returns its page: fields, keys, events, procedures, calls, obsolete state, versions, countries, Learn pages; a W1 object wins over its country twins unless a country is given, and two candidates are listed with their paths. A page over ${B} characters comes back compact (frontmatter, the first 5 rows of each section, the section keys): pass section ('fields', 'events', 'procedures', 'calls', ...) with offset and limit to read one. To check that a field, event or procedure exists, search('event: <Name>') answers without the page.`,
+    description: `An AL object of W1, a Microsoft first-party app or a country layer by type and id, reference, name or caption: ('table', '18'), ('table', 't36'), ('table', '36-be'), ('codeunit', 'Sales-Post'), ('page', 'Customers'), or with country 'BE' (what that layer adds prints first). Returns its page: fields, keys, events, procedures, calls, obsolete state, versions, countries, Learn pages; a W1 object wins over its country twins unless a country is given, and two candidates are listed with their paths. A page over ${B} characters comes back compact (frontmatter, the first 5 rows of each section, the section keys): pass section ('fields', 'events', 'procedures', 'calls', ...) with offset and limit to read one. To check that a field, event or procedure exists, search('event: <Name>') answers without the page. counts.procedures counts every procedure (public, local, events, subscribers); the summary's 'public procedures' excludes locals and events.`,
     inputSchema: { type: z.string(), idOrName: z.string(), country: z.string().length(2).optional(), ...VIEW }
   }, async (a) => text(await toolGetObject(a)));
   server.registerTool("diff_object", {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bc-observatory",
-  "version": "0.5.0",
+  "version": "0.7.1",
   "description": "MCP server for BC Observatory: an agent-first, cross-referenced knowledge base of Microsoft Dynamics 365 Business Central (Learn, code, roadmap, videos, community blogs).",
   "type": "module",
   "bin": {