wiki-formant 0.100.0 → 0.104.0

package/README.md

@@ -565,7 +565,7 @@ const off = onTweetResize((h, source) => sizeTweetEmbed(el, h, source));
 
 ## Search
 
-`wiki-formant/search`: what a literal tier and a full-text tier must agree on: `proseSql`, `HEADLINE_OPTIONS`, `FTS_RANK_NORMALIZATION`, and `searchTsvDdl(table)`, the statements that drop and re-add the generated `search_tsv` column and its index in one transaction.
+`wiki-formant/search`: what a literal tier and a full-text tier must agree on: `proseSql`, `HEADLINE_OPTIONS`, `FTS_RANK_NORMALIZATION`, and `searchTsvDdl(table)`, the statements that drop and re-add the generated `search_tsv` column and its index in one transaction. `fullTextQuery` is a query as the full-text tier reads it, a hyphen or underscore inside a word a space; `correctQuery(query, vocabulary)` corrects a query's misspelt words against a site's own, which the store's `search` retries with when its first page is empty, returning the query that found the rows as `corrected`.
 
 ```ts
 const { column, index } = searchTsvDdl('"WikiPage"');
@@ -715,7 +715,7 @@ export const POST = viewRoute();
 
 - `createTracker` records only on a Vercel production deployment unless `enabled` says otherwise. `trackMcpCall` fits `McpServerConfig.onCall`; `trackSearch` normalises a query and drops an empty one; `recordEvent` is the undeferred write.
 - A proxy reaches the tracker through a dynamic `import()` inside `event.waitUntil`, since it imports the database client.
-- `digest(sql, { days })` is the site as one JSON object: visitors, agents, views, bounce rate, visit length, sources, pages, countries, devices, by day, agents by name, MCP calls and searches that found nothing. A visit is a run of views with no gap over 30 minutes; a bounce is a one-view visit with no event of the reader's. `ownAgents` leaves the site's own out. A multi-tenant app passes `site`.
+- `digest(sql, { days })` is the site as one JSON object: visitors, agents, views, visits, bounce rate, visit length, sources, pages, countries, devices, by day, agents by name, MCP calls and searches that found nothing, less a search its visitor went on typing within five minutes and any of one character. A visit is a run of views with no gap over 30 minutes; a bounce is a one-view visit with no event of the reader's. `ownAgents` leaves the site's own out. A multi-tenant app passes `site`.
 - A browser under automation (`navigator.webdriver`) sends no beacon.
 
 ## Passkeys and the stats page

package/dist/analytics.d.ts

@@ -110,6 +110,8 @@ export interface DigestDay {
 export interface Digest {
     visitors: number;
     pageviews: number;
+    /** Runs of one visitor's views with no gap over 30 minutes. */
+    visits: number;
     /** Percent of visits that are bounces (`DigestDay.bounces`); null with no visits. */
     bounce_rate: number | null;
     /** Mean seconds from a visit's first view to its last; null with no visits. */
@@ -129,9 +131,10 @@ export interface Digest {
      * the archive cannot stand in for it: event figures always, all of them
      * when filtered.
      */
-    previous: Record<'visitors' | 'pageviews' | 'bounce_rate' | 'visit_duration' | 'agents' | 'visitors_incl_agents' | 'follow_clicks', number | null>;
+    previous: Record<'visitors' | 'pageviews' | 'visits' | 'bounce_rate' | 'visit_duration' | 'agents' | 'visitors_incl_agents' | 'follow_clicks', number | null>;
     /** The first UTC day this site counted itself, before which only `ViewDay` speaks; null with no views. */
     counted_from: string | null;
+    /** `(bot)` stands for each referrer `BOT_VISITORS` takes for a crawler's. */
     top_sources: Ranked<'source', 'visitors'>;
     top_pages: Ranked<'page', 'visitors'>;
     entry_pages: Ranked<'page', 'visitors'>;

package/dist/analytics.js

@@ -158,13 +158,26 @@ export const STATS_FILTERS = ['page', 'entry', 'source', 'country', 'device'];
  * `Stats` calls it on the server, where a client module's export is only a reference. */
 export const count = (n) => n.toLocaleString('en');
 /**
- * Each filter as a test on one visit. A page, country or device is any of the
- * visit's views; an entry page or a source is its first. `(none)` is no source.
+ * A referrer that sent this many visitors in the window, not one of whom read a
+ * second page, is a crawler naming a site it never came from: aocr.org sent
+ * radix.wiki 319 one-page visitors in a month while Google's bounced four in five.
+ */
+const BOT_VISITORS = 20;
+const BOT_SOURCE = '(bot)';
+/** A visit's source: its first view's, or `(bot)` where that source is among `bots`. */
+const visitSource = (s) => {
+    const first = '(array_agg(source ORDER BY at, id))[1]';
+    return `CASE WHEN ${first} IN (SELECT source FROM bots${s}) THEN '${BOT_SOURCE}' ELSE ${first} END`;
+};
+/**
+ * Each filter as a test on one visit in the window suffixed `s`. A page,
+ * country or device is any of the visit's views; an entry page or a source is
+ * its first. `(none)` is no source.
  */
 const VISIT_TEST = {
     page: v => `bool_or(path = ${v})`,
     entry: v => `(array_agg(path ORDER BY at, id))[1] = ${v}`,
-    source: v => `coalesce((array_agg(source ORDER BY at, id))[1], '(none)') = ${v}`,
+    source: (v, s) => `coalesce(${visitSource(s)}, '(none)') = ${v}`,
     country: v => `bool_or(country = ${v})`,
     device: v => `bool_or(device = ${v})`,
 };
@@ -207,17 +220,34 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
     const merged = (label, column, live) => `SELECT k AS ${label}, sum(n) AS visitors FROM (${live} UNION ALL SELECT key, value::int FROM a, jsonb_each_text(a."${column}")) t(k, n) GROUP BY 1`;
     /** Clicks on links to the site's own X account, among `e`. */
     const follows = handle && `name = 'Outbound Link: Click' AND props->>'url' ILIKE ANY(${$([`%x.com/${handle}%`, `%twitter.com/${handle}%`])}::text[])`;
-    const tests = STATS_FILTERS.flatMap(k => (filter[k] ? [VISIT_TEST[k]($(filter[k]))] : []));
+    const tests = STATS_FILTERS.flatMap(k => {
+        if (!filter[k])
+            return [];
+        const v = $(filter[k]);
+        return [(s) => VISIT_TEST[k](v, s)];
+    });
     const agentTests = STATS_FILTERS.flatMap(k => (filter[k] ? [AGENT_TEST[k]($(filter[k]))] : []));
     /** Agents among the rows of a ranked list, unless they are counted apart. */
     const agentsToo = (select) => (apart ? '' : ` UNION ALL ${select}`);
     const day = (at) => `to_char(${at} AT TIME ZONE 'UTC', 'YYYY-MM-DD')`;
     /** The site's own agents, as a test on an event's user agent. */
     const ours = `coalesce(props->>'ua', '') ILIKE ANY(${$(ownAgents.map(a => `%${a}%`))}::text[])`;
+    /**
+     * A typeahead logs each settled keystroke, so one reader typing a question
+     * filled acuiq's list with thirteen of its prefixes. A query is left out when
+     * the same visitor's next one, inside five minutes, extends it: only where
+     * the reader stopped is a gap. So is a single character.
+     */
     const gapSelects = gaps.map(g => {
         const prop = $(g.prop);
-        return `(SELECT props->>${prop} AS query, count(*) AS events FROM e
-      WHERE name = ${$(g.event)} AND props @> ${$(JSON.stringify(g.where ?? {}))}::jsonb AND props->>${prop} IS NOT NULL
+        return `(SELECT query, count(*) AS events FROM (
+        SELECT props, visitor, props->>${prop} AS query,
+          lead(props->>${prop}) OVER typed AS next, lead(at) OVER typed - at AS after
+        FROM e WHERE name = ${$(g.event)} AND props->>${prop} IS NOT NULL
+        WINDOW typed AS (PARTITION BY visitor ORDER BY at)
+      ) s
+      WHERE props @> ${$(JSON.stringify(g.where ?? {}))}::jsonb AND char_length(query) > 1
+        AND NOT coalesce(visitor IS NOT NULL AND after <= interval '5 minutes' AND char_length(next) > char_length(query) AND starts_with(next, query), false)
       GROUP BY 1 ORDER BY 2 DESC, 1 LIMIT 15)`;
     });
     /**
@@ -234,10 +264,13 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
       SELECT *, CASE WHEN at - lag(at) OVER (PARTITION BY visitor ORDER BY at, id) <= interval '30 minutes' THEN 0 ELSE 1 END AS starts FROM viewed${s}
     ), numbered${s} AS (
       SELECT *, sum(starts) OVER (PARTITION BY visitor ORDER BY at, id) AS visit FROM marked${s}
+    ), bots${s} AS (
+      SELECT source FROM (SELECT visitor, (array_agg(source ORDER BY at, id))[1] AS source, count(*) AS views FROM numbered${s} GROUP BY visitor, visit) f
+      WHERE source IS NOT NULL GROUP BY 1 HAVING count(DISTINCT visitor) >= ${BOT_VISITORS} AND max(views) = 1
     ), visits${s} AS (
       SELECT visitor, visit, min(at) AS at, count(*) AS views, extract(epoch FROM max(at) - min(at)) AS seconds,
-        (array_agg(path ORDER BY at, id))[1] AS entry, (array_agg(source ORDER BY at, id))[1] AS source
-      FROM numbered${s} GROUP BY visitor, visit${tests.length ? ` HAVING ${tests.join(' AND ')}` : ''}
+        (array_agg(path ORDER BY at, id))[1] AS entry, ${visitSource(s)} AS source
+      FROM numbered${s} GROUP BY visitor, visit${tests.length ? ` HAVING ${tests.map(t => t(s)).join(' AND ')}` : ''}
     ), v${s} AS (
       ${tests.length ? `SELECT n.id, n.path, n.source, n.country, n.device, n.visitor, n.at FROM numbered${s} n JOIN visits${s} USING (visitor, visit)` : `SELECT * FROM viewed${s}`}
     ), read${s} AS (
@@ -268,6 +301,7 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
     /** One window's headline figures, as `json_build_object` arguments. */
     const headline = (s) => `'visitors', (SELECT count(DISTINCT visitor) FROM v${s}) + (SELECT visitors FROM archived${s}),
       'pageviews', (SELECT count(*) FROM v${s}) + (SELECT pageviews FROM archived${s}),
+      'visits', (SELECT visits FROM sessions${s}),
       'bounce_rate', (SELECT round(100.0 * bounces / nullif(visits, 0)) FROM sessions${s}),
       'visit_duration', (SELECT round(seconds / nullif(visits, 0)) FROM sessions${s}),
       'agents', (SELECT count(DISTINCT visitor) FROM ag${s}),

package/dist/search.d.ts

@@ -72,3 +72,22 @@ export declare const FTS_RANK_NORMALIZATION = 32;
  * rather than as a pattern matching everything.
  */
 export declare function escapeLikeTerm(query: string): string;
+/**
+ * A query as the full-text tier reads it: a hyphen or underscore inside a word
+ * is a space. `websearch_to_tsquery` keeps a hyphenated word as a phrase that
+ * needs the compound itself, so the slug `radix-pipeline` found nothing where
+ * `radix pipeline` found sixteen pages. A leading `-` is websearch's NOT, and
+ * stays.
+ */
+export declare function fullTextQuery(query: string): string;
+/**
+ * A query with each misspelled word replaced by the nearest word in `vocabulary`
+ * (word → how often it occurs), or `null` when no word changes. Called only for
+ * a query that found nothing, so a word the wiki already uses is never touched.
+ *
+ * Tight on purpose: a word of four to seven letters may be one edit out, a
+ * longer one two, and a shorter one none. Looser, a name the wiki lacks turns
+ * into a word it has – `bondix` into `bonds` – and the reader gets an answer to
+ * a question they did not ask. Ties go to the commoner word.
+ */
+export declare function correctQuery(query: string, vocabulary: ReadonlyMap<string, number>): string | null;

package/dist/search.js

@@ -120,3 +120,62 @@ export const FTS_RANK_NORMALIZATION = 32;
 export function escapeLikeTerm(query) {
     return (query ?? '').trim().replace(/[\\%_]/g, char => `\\${char}`);
 }
+/**
+ * A query as the full-text tier reads it: a hyphen or underscore inside a word
+ * is a space. `websearch_to_tsquery` keeps a hyphenated word as a phrase that
+ * needs the compound itself, so the slug `radix-pipeline` found nothing where
+ * `radix pipeline` found sixteen pages. A leading `-` is websearch's NOT, and
+ * stays.
+ */
+export function fullTextQuery(query) {
+    return (query ?? '').trim().replace(/([\p{L}\p{N}])[-_]+(?=[\p{L}\p{N}])/gu, '$1 ');
+}
+/** Edits between two words, a swapped adjacent pair counting as one. */
+function editDistance(a, b) {
+    const d = Array.from({ length: a.length + 1 }, (_, i) => [i, ...Array(b.length).fill(0)]);
+    for (let j = 1; j <= b.length; j++)
+        d[0][j] = j;
+    for (let i = 1; i <= a.length; i++)
+        for (let j = 1; j <= b.length; j++) {
+            const cost = a[i - 1] === b[j - 1] ? 0 : 1;
+            d[i][j] = Math.min(d[i - 1][j] + 1, d[i][j - 1] + 1, d[i - 1][j - 1] + cost);
+            if (i > 1 && j > 1 && a[i - 1] === b[j - 2] && a[i - 2] === b[j - 1])
+                d[i][j] = Math.min(d[i][j], d[i - 2][j - 2] + 1);
+        }
+    return d[a.length][b.length];
+}
+/**
+ * A query with each misspelled word replaced by the nearest word in `vocabulary`
+ * (word → how often it occurs), or `null` when no word changes. Called only for
+ * a query that found nothing, so a word the wiki already uses is never touched.
+ *
+ * Tight on purpose: a word of four to seven letters may be one edit out, a
+ * longer one two, and a shorter one none. Looser, a name the wiki lacks turns
+ * into a word it has – `bondix` into `bonds` – and the reader gets an answer to
+ * a question they did not ask. Ties go to the commoner word.
+ */
+export function correctQuery(query, vocabulary) {
+    let changed = false;
+    const fixed = query.trim().toLowerCase().split(/\s+/).map(word => {
+        const budget = word.length >= 8 ? 2 : word.length >= 4 ? 1 : 0;
+        if (!budget || vocabulary.has(word) || !/^[\p{L}\p{N}]+$/u.test(word))
+            return word;
+        let best = null;
+        let bestDistance = Infinity;
+        let bestCount = 0;
+        for (const [candidate, count] of vocabulary) {
+            if (Math.abs(candidate.length - word.length) > budget)
+                continue;
+            const distance = editDistance(word, candidate);
+            if (distance > budget)
+                continue;
+            if (distance < bestDistance || (distance === bestDistance && count > bestCount))
+                [best, bestDistance, bestCount] = [candidate, distance, count];
+        }
+        if (!best)
+            return word;
+        changed = true;
+        return best;
+    });
+    return changed ? fixed.join(' ') : null;
+}

package/dist/stats-figures.js

@@ -13,6 +13,8 @@ import { TimeChart, stepDecimals } from './chart.js';
 const places = (step) => (step ? stepDecimals(step) : 0);
 const duration = (s, step) => (s < 60 ? `${s.toFixed(places(step))}s` : `${Math.floor(Math.round(s) / 60)}m ${Math.round(s) % 60}s`);
 const percent = (n, step) => `${n.toFixed(places(step))}%`;
+// Views per visit sit between one and a few, so a tenth tells them apart.
+const perVisit = (n, step) => n.toFixed(step ? stepDecimals(step) : 1);
 /**
  * The change from the window before: a share of the figure before, or for a
  * rate its difference in points, since a bounce rate up 10% of itself reads
@@ -42,6 +44,14 @@ const figuresOf = ({ previous: p, ...d }, apart) => [
             },
         ]),
     { label: 'Page views', value: count(d.pageviews), delta: change(d.pageviews, p.pageviews), day: r => r.pageviews },
+    {
+        label: 'Views per visit',
+        value: d.visits ? perVisit(d.pageviews / d.visits) : '–',
+        delta: change(d.visits ? d.pageviews / d.visits : null, p.pageviews != null && p.visits ? p.pageviews / p.visits : null),
+        day: r => (r.visits ? r.pageviews / r.visits : null),
+        format: perVisit,
+        aggregate: 'mean',
+    },
     {
         label: 'Bounce rate',
         value: d.bounce_rate == null ? '–' : percent(d.bounce_rate),

package/dist/stats.d.ts

@@ -30,7 +30,8 @@ export declare function statsQuery(params: Record<string, string | string[] | un
  * A site's own analytics on one page: the headline figures against the window
  * before, visitors counting agents unless `apart`, whichever of them is chosen per day, and every ranked list the
  * digest carries, each row a link that narrows the page to it. Lists with
- * nothing in them are left out. The digest is made with the same `apart`, so
+ * nothing in them are left out, but for the searches that found nothing,
+ * which says it found none, so a quiet window never reads as one not counted. The digest is made with the same `apart`, so
  * the lists count the visitors the chart draws. The chart and the bars are `currentColor`
  * until the importing stylesheet gives them a colour.
  */

package/dist/stats.js

@@ -46,9 +46,9 @@ const country = (code) => {
 };
 const NAMES = { page: 'Page', entry: 'Entry page', source: 'Source', country: 'Country', device: 'Device' };
 const DAY_MS = 86_400_000;
-function StatsList({ title, rows, label = v => v, to, link: Link, }) {
+function StatsList({ title, rows, label = v => v, to, empty, link: Link, }) {
     if (!rows.length)
-        return null;
+        return empty ? (_jsxs("section", { className: "stats-list", children: [_jsx("h2", { children: title }), _jsx("p", { children: empty })] })) : null;
     const max = Math.max(...rows.map(([, n]) => n));
     return (_jsxs("section", { className: "stats-list", children: [_jsx("h2", { children: title }), _jsx("ol", { children: rows.map(([v, n]) => (_jsxs("li", { style: { '--share': n / max }, children: [_jsx("span", { children: to ? (_jsx(Link, { href: to(v), prefetch: false, children: label(v) })) : (label(v)) }), _jsx("span", { className: "stats-count", children: count(n) })] }, v))) })] }));
 }
@@ -56,7 +56,8 @@ function StatsList({ title, rows, label = v => v, to, link: Link, }) {
  * A site's own analytics on one page: the headline figures against the window
  * before, visitors counting agents unless `apart`, whichever of them is chosen per day, and every ranked list the
  * digest carries, each row a link that narrows the page to it. Lists with
- * nothing in them are left out. The digest is made with the same `apart`, so
+ * nothing in them are left out, but for the searches that found nothing,
+ * which says it found none, so a quiet window never reads as one not counted. The digest is made with the same `apart`, so
  * the lists count the visitors the chart draws. The chart and the bars are `currentColor`
  * until the importing stylesheet gives them a colour.
  */
@@ -77,7 +78,7 @@ export function Stats({ digest: d, days, filter = {}, apart = false, ranges = ST
     // Each window is another digest, so a link rather than a button, pinned
     // with the chart's own controls, and so is the switch that counts agents apart.
     const controls = (_jsxs(_Fragment, { children: [_jsx("div", { role: "group", "aria-label": "Window", children: ranges.map(n => (_jsx(Link, { href: at(filter, n), "aria-current": n === days ? 'page' : undefined, prefetch: false, children: windowLabel(n) }, n))) }), _jsx("div", { children: _jsx(Link, { href: at(filter, days, !apart), role: "switch", "aria-checked": !apart, "aria-label": "Agents counted among visitors", title: "Agents counted among visitors", prefetch: false, children: _jsxs("svg", { className: "chart-icon", viewBox: "0 0 12 12", "aria-hidden": true, children: [_jsx("path", { d: "M6 1.5v2M1 6.5v2M11 6.5v2", stroke: "currentColor" }), _jsx("rect", { x: "2.5", y: "3.5", width: "7", height: "7", rx: "1.5", fill: "none", stroke: "currentColor" }), _jsx("circle", { cx: "4.75", cy: "7", r: "0.9", fill: "currentColor" }), _jsx("circle", { cx: "7.25", cy: "7", r: "0.9", fill: "currentColor" })] }) }) })] }));
-    return (_jsxs("div", { className: "stats", children: [narrowed.length > 0 && (_jsxs("div", { className: "stats-filters", children: [_jsx("ul", { "aria-label": "Narrowed to", children: narrowed.map(k => (_jsx("li", { children: _jsxs(Link, { href: at({ ...filter, [k]: undefined }), title: "Remove", prefetch: false, children: [NAMES[k], ": ", k === 'country' ? country(filter[k]) : filter[k], " \u00D7"] }) }, k))) }), start < counted && d.counted_from && (_jsxs("p", { children: ["Narrowed figures start on ", d.counted_from, ", the first day this site counted its own visits."] }))] })), _jsx(StatsFigures, { digest: d, first: first, apart: apart, controls: controls }), _jsxs("div", { className: "stats-lists", children: [_jsx(StatsList, { link: Link, title: "Pages", rows: rows(d.top_pages, 'page', 'visitors'), to: to('page') }), _jsx(StatsList, { link: Link, title: "Entry pages", rows: rows(d.entry_pages, 'page', 'visitors'), to: to('entry') }), _jsx(StatsList, { link: Link, title: "Sources", rows: rows(d.top_sources, 'source', 'visitors'), to: to('source') }), _jsx(StatsList, { link: Link, title: "From X posts", rows: rows(d.from_posts, 'page', 'visitors'), to: v => at({ ...filter, source: 'x', entry: v }) }), _jsx(StatsList, { link: Link, title: "Sibling sites", rows: rows(d.sibling_referrals, 'source', 'visitors'), to: to('source') }), _jsx(StatsList, { link: Link, title: "Countries", rows: rows(d.countries, 'country', 'visitors'), label: country, to: to('country') }), _jsx(StatsList, { link: Link, title: "Devices", rows: rows(d.devices, 'device', 'visitors'), to: to('device') }), _jsx(StatsList, { link: Link, title: "Agents", rows: rows(d.top_agents, 'agent', 'visitors') }), _jsx(StatsList, { link: Link, title: "Agent tool calls", rows: rows(d.agent_tool_calls, 'tool', 'events') }), _jsx(StatsList, { link: Link, title: "Searches that found nothing", rows: rows(d.gaps, 'query', 'events') })] })] }));
+    return (_jsxs("div", { className: "stats", children: [narrowed.length > 0 && (_jsxs("div", { className: "stats-filters", children: [_jsx("ul", { "aria-label": "Narrowed to", children: narrowed.map(k => (_jsx("li", { children: _jsxs(Link, { href: at({ ...filter, [k]: undefined }), title: "Remove", prefetch: false, children: [NAMES[k], ": ", k === 'country' ? country(filter[k]) : filter[k], " \u00D7"] }) }, k))) }), start < counted && d.counted_from && (_jsxs("p", { children: ["Narrowed figures start on ", d.counted_from, ", the first day this site counted its own visits."] }))] })), _jsx(StatsFigures, { digest: d, first: first, apart: apart, controls: controls }), _jsxs("div", { className: "stats-lists", children: [_jsx(StatsList, { link: Link, title: "Pages", rows: rows(d.top_pages, 'page', 'visitors'), to: to('page') }), _jsx(StatsList, { link: Link, title: "Entry pages", rows: rows(d.entry_pages, 'page', 'visitors'), to: to('entry') }), _jsx(StatsList, { link: Link, title: "Sources", rows: rows(d.top_sources, 'source', 'visitors'), to: to('source') }), _jsx(StatsList, { link: Link, title: "From X posts", rows: rows(d.from_posts, 'page', 'visitors'), to: v => at({ ...filter, source: 'x', entry: v }) }), _jsx(StatsList, { link: Link, title: "Sibling sites", rows: rows(d.sibling_referrals, 'source', 'visitors'), to: to('source') }), _jsx(StatsList, { link: Link, title: "Countries", rows: rows(d.countries, 'country', 'visitors'), label: country, to: to('country') }), _jsx(StatsList, { link: Link, title: "Devices", rows: rows(d.devices, 'device', 'visitors'), to: to('device') }), _jsx(StatsList, { link: Link, title: "Agents", rows: rows(d.top_agents, 'agent', 'visitors') }), _jsx(StatsList, { link: Link, title: "Agent tool calls", rows: rows(d.agent_tool_calls, 'tool', 'events') }), _jsx(StatsList, { link: Link, title: "Searches that found nothing", rows: rows(d.gaps, 'query', 'events'), empty: "None in this window." })] })] }));
 }
 /**
  * A stats page's door: `children` and a sign-out button for a signed-in

package/dist/store.d.ts

@@ -47,6 +47,13 @@ export type WikiRevisionEntry = Omit<WikiRevisionRow, 'content'> & {
     changes: HistoryChange[];
 };
 export type ListSort = 'title' | 'recent' | 'date';
+/** A search's scope and page. `content` adds each row's blocks. */
+export interface SearchOptions {
+    tagPath?: string | null;
+    skip?: number;
+    take?: number;
+    content?: boolean;
+}
 export interface WikiStoreOptions<B extends DiffBlock> {
     sql: Sql;
     /** The site every read and write is scoped to. "" in an app that serves one wiki. */
@@ -169,17 +176,14 @@ export declare function wikiStore<B extends DiffBlock = DiffBlock>(o: WikiStoreO
         pages: number;
         updated: Date | null;
     }>;
-    search: (query: string, { tagPath, skip, take, content: withContent }?: {
-        tagPath?: string | null;
-        skip?: number;
-        take?: number;
-        content?: boolean;
-    }) => Promise<{
+    search: (query: string, options?: SearchOptions) => Promise<Awaited<ReturnType<(query: string, { tagPath, skip, take, content: withContent }?: SearchOptions) => Promise<{
         items: (WikiPageListing & {
             score: number;
             content?: B[];
         })[];
         total: number;
+    }>>> & {
+        corrected?: string;
     }>;
     history: (pageId: string, { limit, offset }?: {
         limit?: number;

package/dist/store.js

@@ -18,7 +18,7 @@
 import { computeRevisionDiff } from './revisions.js';
 import { bump } from './versioning.js';
 import { withChanges } from './history.js';
-import { proseSql, escapeLikeTerm, HEADLINE_OPTIONS, FTS_RANK_NORMALIZATION } from './search.js';
+import { proseSql, escapeLikeTerm, fullTextQuery, correctQuery, HEADLINE_OPTIONS, FTS_RANK_NORMALIZATION } from './search.js';
 import { openingText, matchSnippet } from './text.js';
 import { describe } from './metadata.js';
 import { decodeEntities } from './markdown.js';
@@ -132,7 +132,7 @@ export function wikiStore(o) {
      * page is never a result. A tier-3 row's snippet is Postgres' own headline,
      * because a stem match has no literal substring to window on.
      */
-    const search = async (query, { tagPath = null, skip = 0, take = 25, content: withContent = false } = {}) => {
+    const ranked = async (query, { tagPath = null, skip = 0, take = 25, content: withContent = false } = {}) => {
         const term = escapeLikeTerm(query);
         if (!term)
             return { items: [], total: 0 };
@@ -163,7 +163,7 @@ export function wikiStore(o) {
        SELECT ${COLUMNS}, tier, total,
               CASE WHEN tier = 3 THEN ts_headline('english', ${proseSql()}, tsq, '${HEADLINE_OPTIONS}') END AS headline
          FROM (SELECT count(*)::int AS total FROM matched) t LEFT JOIN paged ON true
-        ORDER BY tier, CASE WHEN tier = 0 THEN title END, CASE WHEN tier = 3 THEN fts END DESC, "updatedAt" DESC`, query.trim(), `${term}%`, `%${term}%`, hidden, site, tagPath, take, skip);
+        ORDER BY tier, CASE WHEN tier = 0 THEN title END, CASE WHEN tier = 3 THEN fts END DESC, "updatedAt" DESC`, fullTextQuery(query), `${term}%`, `%${term}%`, hidden, site, tagPath, take, skip);
         const items = found.filter(hit => hit.id !== null).map(({ tier, total: _, headline, content, ...row }) => ({
             ...row,
             ...(withContent ? { content: content } : {}),
@@ -173,6 +173,26 @@ export function wikiStore(o) {
         }));
         return { items, total: found[0]?.total ?? 0 };
     };
+    /** Every word of the searchable titles, with how many titles use it: what a misspelling is corrected against. */
+    const titleWords = async (tagPath) => new Map((await rows(`SELECT word, count(*)::int AS n FROM ${PAGE} p, regexp_split_to_table(lower(p.title), '[^[:alnum:]]+') word
+        WHERE p."siteId" = $1 AND word <> '' AND ($2::text IS NULL OR p."tagPath" = $2) AND ($2::text IS NOT NULL OR NOT ${hiddenSql('$3')})
+        GROUP BY 1`, site, tagPath, hidden)).map(r => [r.word, r.n]));
+    /**
+     * {@link ranked}, and when its first page is empty, the same search with each
+     * misspelled word corrected against the titles' own words (`correctQuery`).
+     * A result found that way carries `corrected`, the query that found it, so a
+     * page can say what it searched for instead.
+     */
+    const search = async (query, options = {}) => {
+        const found = await ranked(query, options);
+        if (found.total || options.skip)
+            return found;
+        const corrected = correctQuery(query, await titleWords(options.tagPath ?? null));
+        if (!corrected)
+            return found;
+        const again = await ranked(corrected, options);
+        return again.total ? { ...again, corrected } : found;
+    };
     /** A revision query's guard: `$1`, its page, is this site's. Another site's page id finds nothing. */
     const OWN_PAGE = `"pageId" IN (SELECT id FROM ${PAGE} WHERE id = $1 AND "siteId" = $2)`;
     /**

package/dist/wiki-tools.d.ts

@@ -75,5 +75,6 @@ export declare function searchWiki<B extends DiffBlock>(o: Pick<WikiToolsOptions
 }): Promise<{
     items: WikiRow[];
     total: number;
+    corrected?: string;
 }>;
 export declare function wikiReadTools<B extends DiffBlock>(o: WikiToolsOptions<B>): WikiTool[];

package/dist/wiki-tools.js

@@ -54,8 +54,8 @@ const rowFor = (o) => (p) => {
 export async function searchWiki(o, query, { tagPath, page = 1, size = 20 } = {}) {
     if (offFor(o)(tagPath))
         return { items: [], total: 0 };
-    const { items, total } = await o.store.search(query, { tagPath, skip: (page - 1) * size, take: size, content: true });
-    return { items: items.map(rowFor(o)), total };
+    const { items, total, corrected } = await o.store.search(query, { tagPath, skip: (page - 1) * size, take: size, content: true });
+    return { items: items.map(rowFor(o)), total, ...(corrected ? { corrected } : {}) };
 }
 export function wikiReadTools(o) {
     const { store, tree, noun } = o;
@@ -116,7 +116,7 @@ export function wikiReadTools(o) {
             skill: { id: 'search', tags: ['search', 'wiki'] },
             description: `Keyword search across ${noun}: the first call when you know what you are looking for but not where it lives. ` +
                 'Matches titles and body prose, ranked in four tiers: titles starting with the term, titles containing it, literal body hits, then English full-text hits by relevance, which stem words and drop stopwords. ' +
-                'Not semantic: one distinctive word beats a sentence. ' +
+                'Not semantic: one distinctive word beats a sentence. A query that finds nothing is retried with its misspelled words corrected, and `corrected` then names what was searched. ' +
                 'Each row carries a `path`, its size in `chars` and a snippet windowed on the match. Feed the paths to get_pages in one call rather than looping get_page. ' +
                 'Narrow with tagPath when a term is common. Defaults: page 1, pageSize 20 (max 50); keep paging while `hasMore` is true.',
             inputSchema: {
@@ -137,9 +137,10 @@ export function wikiReadTools(o) {
                 const size = r.num('pageSize', 20, 1, 50);
                 if (!query)
                     throw new McpToolError('`query` is empty. Pass a keyword to search for, or call list_pages to enumerate a branch instead.');
-                const { items, total } = await searchWiki(o, query, { tagPath, page, size });
+                const { items, total, corrected } = await searchWiki(o, query, { tagPath, page, size });
                 return withAdjustments(r, {
                     query,
+                    ...(corrected ? { corrected } : {}),
                     ...listEnvelope(items, total, page, size, o.noMatch?.(query, tagPath)
                         ?? `Nothing matches "${query}"${tagPath ? ` under "${tagPath}"` : ''}. Search is keyword, not semantic: try one distinctive word${tagPath ? ', or drop the tagPath filter' : ''}, or call get_categories to see what ${noun} covers.`),
                 });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wiki-formant",
-  "version": "0.100.0",
+  "version": "0.104.0",
   "description": "The portable half of a wiki: derived taxonomy and facet controls, a version-negotiating MCP transport, markdown twins, block rendering, a rich-text editor engine, conditional-GET plumbing, and the headers and CSP every origin sends.",
   "type": "module",
   "license": "MIT",