wiki-formant 0.100.0 → 0.105.0

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"');
@@ -702,7 +702,7 @@ const SCHEMA: BlockSchema<Block> = {
 
 ## Analytics
 
-`wiki-formant/analytics`: page views and events in the site's own Postgres. No cookie and no stored address: a visitor is a hash of address and browser under a salt replaced every UTC day. Rows are kept 400 days. The tables ship in `prisma/analytics.prisma`; `ViewDay` holds days a site counted elsewhere first.
+`wiki-formant/analytics`: page views and events in the site's own Postgres. No cookie and no stored address: a visitor is a hash of address and browser under a salt replaced every UTC day. Rows are kept 400 days. A view is marked engaged once its reader has had the page visible fifteen seconds or given it input, so `engaged` counts people where `agents` counts crawlers. The tables ship in `prisma/analytics.prisma`; `ViewDay` holds days a site counted elsewhere first.
 
 ```ts
 // src/lib/track.ts
@@ -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: people, engaged people, agents, views, visits, bounce rate, 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 its reader never engaged with. Agents are counted beside people, never among them: the ranked lists are people'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
@@ -741,7 +741,7 @@ return (
 
 - There is no sign-up: every passkey is saved from a single-use link `npx passkey-invite` prints. A passkey is bound to its host. Deleting a `Passkey` row signs out every session it opened. Tokens are kept only as hashes (`prisma/passkey.prisma`).
 - `gate.signedIn(cookie)` is a page's check and `gate.allows(request)` a write route's.
-- `statsQuery` reads `?days=` (1, 7, 30, 90, 365 or all), `?agents=apart` and the narrowings `?page=`, `?entry=`, `?source=`, `?country=`, `?device=`, which stack and recount every figure from the visits that pass.
+- `statsQuery` reads `?days=` (1, 7, 30, 90, 365 or all) and the narrowings `?page=`, `?entry=`, `?source=`, `?country=`, `?device=`, which stack and recount every figure from the visits that pass.
 - Each headline figure shows its change from the window before (`.stats-delta[data-better]`) and charts on press. A narrowed page over `ViewDay` totals starts at `counted_from` and says so.
 
 ## ROLA sign-in

package/dist/analytics.d.ts

@@ -25,11 +25,12 @@ export interface EventInput {
 }
 /** Record one named event. Never rejects. */
 export declare function recordEvent(sql: Sql, { name, props, url, headers, site }: EventInput): Promise<void>;
-/** What a page sends: a view, or with `name` an event. */
+/** What a page sends: a view, with `engaged` its reader's engagement, or with `name` an event. */
 export interface Beacon {
     path: string;
     referrer?: string;
     utm?: string;
+    engaged?: boolean;
     name?: string;
     props?: Record<string, string | number>;
 }
@@ -83,55 +84,61 @@ export interface DigestOptions {
      * and has no `previous_visitors`.
      */
     filter?: StatsFilter;
-    /**
-     * People and agents counted apart, as they are by default: the ranked lists
-     * count people. False counts agents in the pages, entry pages, sources,
-     * countries and devices too, as `visitors_incl_agents` counts them, so a
-     * stats page's lists add up to the visitors its chart draws.
-     */
-    apart?: boolean;
 }
 type Ranked<K extends string, M extends string> = Array<Record<K, string> & Record<M, number>>;
 /** One UTC day of the figures a digest totals; a visit falls on the day it began. */
 export interface DigestDay {
     date: string;
     visitors: number;
+    /** Visitors who engaged with a page they viewed (`startBeacon`). */
+    engaged: number;
     pageviews: number;
     visits: number;
-    /** Visits that saw one page and did nothing else: a search, a click out or any reader event makes it a visit. */
+    /** Visits that bounced (`Digest.bounce_rate`); an archived day has none. */
     bounces: number;
-    /** Seconds from each visit's first view to its last, summed. */
-    seconds: number;
     agents: number;
-    visitors_incl_agents: number;
     /** Clicks on links to the site's X handle; zero without a handle. */
     follow_clicks: number;
 }
 export interface Digest {
     visitors: number;
+    /**
+     * Visitors who engaged with a page they viewed: had it in a visible tab for
+     * fifteen seconds, or gave it input (`startBeacon`). A crawler that runs
+     * scripts renders a page and goes, so these are people. `ViewDay` keeps no
+     * engagement, so an archived day adds nothing, and the window before is
+     * compared only where engagement was counted from its first day.
+     */
+    engaged: number;
     pageviews: number;
-    /** Percent of visits that are bounces (`DigestDay.bounces`); null with no visits. */
+    /** Runs of one visitor's views with no gap over 30 minutes. */
+    visits: number;
+    /**
+     * Percent of visits that bounced: saw one page and never engaged with it,
+     * the complement of `engaged`. Over the site's own visits, since `ViewDay`
+     * keeps no engagement; null with none. Compared with the window before as
+     * `engaged` is.
+     */
     bounce_rate: number | null;
-    /** Mean seconds from a visit's first view to its last; null with no visits. */
-    visit_duration: number | null;
     /**
      * Agents, counted as visitors are: an AI crawler, an MCP client, or anything
      * else reading the machine surface (`trackAiBot`), one per address and user
      * agent a day, the site's own left out. They run no page script, so they are
-     * never among `visitors`.
+     * never among `visitors`, and the ranked lists count people alone.
      */
     agents: number;
-    /** Visitors and `agents` as one count, which a stats page shows unless asked to keep them apart. */
-    visitors_incl_agents: number;
     /**
      * The headline figures over the same number of days before, narrowed alike.
      * A figure is null where that window reaches back past `counted_from` and
      * 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' | 'engaged' | 'pageviews' | 'visits' | 'bounce_rate' | '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;
+    /** The first UTC day, in this window or the one before, with a view marked engaged, before which `engaged` and `bounce_rate` mean nothing; null with none. */
+    engaged_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'>;
@@ -164,11 +171,11 @@ export interface Digest {
  * Visitor figures count distinct visitors, so a person who comes back on
  * another day counts once for each day. Days in `ViewDay` add to the visitor,
  * page, visit, source, country, device and per-day figures; events have no
- * history, so an archived day's visitors are its visitors with agents too.
+ * history.
  * The window before is the same chain of views, visits and events a second
  * time, `_p` on every name, so its figures are the headline's to the letter.
  */
-export declare function digest(sql: Sql, { site, days, handle, siblings, gaps, ownAgents, filter, apart }: DigestOptions): Promise<Digest>;
+export declare function digest(sql: Sql, { site, days, handle, siblings, gaps, ownAgents, filter }: DigestOptions): Promise<Digest>;
 /**
  * Props for an "MCP Call" event, read out of the JSON-RPC envelope.
  *

package/dist/analytics.js

@@ -103,6 +103,19 @@ export function recordView(sql, { headers, path, referrer, utm, site = '' }) {
         return Promise.resolve();
     return quietly(async () => sql('INSERT INTO "View" ("siteId", path, source, country, device, visitor) VALUES ($1, $2, $3, $4, $5, $6)', site, page, sourceOf(referrer, utm, headers.get('host') ?? ''), headers.get('x-vercel-ip-country')?.slice(0, 2) || null, deviceOf(ua), await visitorOf(sql, site, headers)));
 }
+/**
+ * Mark a visitor's latest unengaged view of `path` in the last half hour as
+ * engaged: the beacon's second word on a view, sent once its reader engaged
+ * (`startBeacon`). Never rejects. @internal
+ */
+export function recordEngaged(sql, { headers, path, site = '' }) {
+    const page = normalisePath(path);
+    if (page === null)
+        return Promise.resolve();
+    return quietly(async () => sql(`UPDATE "View" SET engaged = true WHERE id = (
+        SELECT id FROM "View" WHERE "siteId" = $1 AND visitor = $2 AND path = $3 AND NOT engaged AND "createdAt" > now() - interval '30 minutes'
+        ORDER BY id DESC LIMIT 1)`, site, await visitorOf(sql, site, headers), page));
+}
 /** Record one named event. Never rejects. */
 export function recordEvent(sql, { name, props = {}, url, headers, site = '' }) {
     return quietly(async () => {
@@ -147,7 +160,9 @@ export async function collect(request, { sql, defer, site, rateLimit: limit = BE
     // decoded twice, `/100%25-off` was dropped and `/what%3Fnow` cut to `/what`.
     defer(() => typeof body.name === 'string' && body.name
         ? recordEvent(sql, { name: body.name, props: beaconProps(body.props), url: raw, headers, site: key })
-        : recordView(sql, { headers, path: raw, referrer: body.referrer, utm: body.utm, site: key }));
+        : body.engaged === true
+            ? recordEngaged(sql, { headers, path: raw, site: key })
+            : recordView(sql, { headers, path: raw, referrer: body.referrer, utm: body.utm, site: key }));
     return done;
 }
 /** Searches in a wiki's own search box that found nothing (`searchQueryProps` events). */
@@ -158,30 +173,40 @@ 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})`,
 };
-/** An agent's source and device in the lists, which no request of its carries. */
-const AGENT_SOURCE = '(agent)';
-const AGENT_DEVICE = 'agent';
 /**
  * Each filter as a test on one agent: its reads in the window are its visit,
- * the first its entry page. Its source and device are the two above, and its
- * country the one its requests came from, where they were counted with one.
+ * the first its entry page, and its country the one its requests came from.
+ * No request of an agent's carries a source or a device, so neither passes one.
  */
 const AGENT_TEST = {
     page: v => `bool_or(path = ${v})`,
     entry: v => `(array_agg(path ORDER BY at))[1] = ${v}`,
-    source: v => `${v} = '${AGENT_SOURCE}'`,
+    source: () => 'false',
     country: v => `bool_or(country = ${v})`,
-    device: v => `${v} = '${AGENT_DEVICE}'`,
+    device: () => 'false',
 };
 /**
  * An agent's name from its event: the roster's label, else what its user agent
@@ -193,11 +218,11 @@ const AGENT_NAME = `coalesce(props->>'bot', substring(props->>'ua' from 'compati
  * Visitor figures count distinct visitors, so a person who comes back on
  * another day counts once for each day. Days in `ViewDay` add to the visitor,
  * page, visit, source, country, device and per-day figures; events have no
- * history, so an archived day's visitors are its visitors with agents too.
+ * history.
  * The window before is the same chain of views, visits and events a second
  * time, `_p` on every name, so its figures are the headline's to the letter.
  */
-export async function digest(sql, { site = '', days, handle, siblings = [], gaps = [SEARCH_GAPS], ownAgents = ['-mcp-test', 'radix-studio'], filter = {}, apart = true }) {
+export async function digest(sql, { site = '', days, handle, siblings = [], gaps = [SEARCH_GAPS], ownAgents = ['-mcp-test', 'radix-studio'], filter = {} }) {
     // The window is the database's clock, which stamped the rows; the app's can run behind it.
     // All time is a century: further back than any row, and still an int to Postgres.
     const values = [site, Number.isFinite(days) ? Math.max(1, Math.round(days)) : 36_500];
@@ -207,17 +232,29 @@ 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 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}`);
+    /** Each filter given, its value a parameter that the test on visits and the one on agents share. */
+    const given = STATS_FILTERS.flatMap(k => (filter[k] ? [[k, $(filter[k])]] : []));
+    const tests = given.map(([k, v]) => (s) => VISIT_TEST[k](v, s));
+    const agentTests = given.map(([k, v]) => AGENT_TEST[k](v));
     const day = (at) => `to_char(${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)`;
     });
     /**
@@ -229,17 +266,20 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
         const dayAgo = (k) => `(now() AT TIME ZONE 'UTC' - make_interval(days => ${k} * $2::int))::date`;
         const at = `"createdAt" >= ${ago(back + 1)}${back ? ` AND "createdAt" < ${ago(back)}` : ''}`;
         return `viewed${s} AS (
-      SELECT id, path, source, country, device, visitor, "createdAt" AS at FROM "View" WHERE "siteId" = $1 AND ${at}
+      SELECT id, path, source, country, device, visitor, engaged, "createdAt" AS at FROM "View" WHERE "siteId" = $1 AND ${at}
     ), marked${s} AS (
       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 ')}` : ''}
+      SELECT visitor, visit, min(at) AS at, count(*) = 1 AND NOT bool_or(engaged) AS bounced,
+        (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}`}
+      ${tests.length ? `SELECT n.id, n.path, n.source, n.country, n.device, n.visitor, n.engaged, n.at FROM numbered${s} n JOIN visits${s} USING (visitor, visit)` : `SELECT * FROM viewed${s}`}
     ), read${s} AS (
       SELECT visitor, "createdAt" AS at, path, props->>'country' AS country, ${AGENT_NAME} AS agent FROM "Event"
       WHERE "siteId" = $1 AND ${at} AND name IN ('AI Bot Visit', 'MCP Call') AND visitor IS NOT NULL AND NOT ${ours}
@@ -248,55 +288,53 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
     ), e${s} AS (
       SELECT name, props, visitor, "createdAt" AS at FROM "Event" WHERE "siteId" = $1 AND ${at}
       ${tests.length ? `AND (visitor IN (SELECT visitor FROM v${s}) OR visitor IN (SELECT visitor FROM ag${s}))` : ''}
-    ), acted${s} AS MATERIALIZED (
-      SELECT visitor, at FROM e${s} WHERE name NOT IN ('AI Bot Visit', 'MCP Call') AND visitor IS NOT NULL
-    ), bounced${s} AS (
-      -- Against the few reader events alone: probing every agent event once per visit took seconds.
-      SELECT * FROM visits${s} b WHERE views = 1 AND NOT EXISTS (
-        SELECT 1 FROM acted${s} x WHERE x.visitor = b.visitor AND x.at >= b.at AND x.at < b.at + interval '30 minutes')
     ), a${s} AS (
       SELECT * FROM "ViewDay" WHERE "siteId" = $1 AND day >= ${dayAgo(back + 1)}${back ? ` AND day < ${dayAgo(back)}` : ''}${tests.length ? ' AND false' : ''}
-    ), sessions${s} AS (
-      SELECT sum(bounces) AS bounces, sum(visits) AS visits, sum(seconds) AS seconds FROM (
-        SELECT (SELECT count(*) FROM bounced${s}), count(*), coalesce(sum(seconds), 0) FROM visits${s}
-        UNION ALL SELECT coalesce(sum(bounces), 0), coalesce(sum(visits), 0), coalesce(sum(seconds), 0) FROM a${s}
-      ) t(bounces, visits, seconds)
     ), archived${s} AS (
-      SELECT coalesce(sum(visitors), 0) AS visitors, coalesce(sum(pageviews), 0) AS pageviews FROM a${s}
+      SELECT coalesce(sum(visitors), 0) AS visitors, coalesce(sum(pageviews), 0) AS pageviews, coalesce(sum(visits), 0) AS visits FROM a${s}
     )`;
     };
+    /**
+     * A figure engagement decides. The window before has it only where
+     * engagement was counted from its first day: its first engaged view is on the day of its first view.
+     */
+    const engagement = (s, figure) => {
+        if (!s)
+            return figure;
+        const first = (where) => `(SELECT ${day('min(at)')} FROM viewed${s}${where})`;
+        return `CASE WHEN ${first(' WHERE engaged')} = ${first('')} THEN ${figure} END`;
+    };
     /** One window's headline figures, as `json_build_object` arguments. */
     const headline = (s) => `'visitors', (SELECT count(DISTINCT visitor) FROM v${s}) + (SELECT visitors FROM archived${s}),
+      'engaged', ${engagement(s, `(SELECT count(DISTINCT visitor) FROM v${s} WHERE engaged)`)},
       'pageviews', (SELECT count(*) FROM v${s}) + (SELECT pageviews FROM archived${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${s}),
-      'visitors_incl_agents', (SELECT count(DISTINCT visitor) FROM (SELECT visitor FROM v${s} UNION SELECT visitor FROM ag${s}) u) + (SELECT visitors FROM archived${s})`;
+      'visits', (SELECT count(*) FROM visits${s}) + (SELECT visits FROM archived${s}),
+      'bounce_rate', ${engagement(s, `(SELECT round(100.0 * count(*) FILTER (WHERE bounced) / nullif(count(*), 0)) FROM visits${s})`)},
+      'agents', (SELECT count(DISTINCT visitor) FROM ag${s})`;
     const [row] = (await sql(`WITH ${chain('', 0)}, ${chain('_p', 1)}, src AS (
       SELECT coalesce(source, '(none)') AS source, count(DISTINCT visitor) AS visitors FROM visits GROUP BY 1
-      UNION ALL SELECT key, value::int FROM a, jsonb_each_text(a.sources)${agentsToo(`SELECT '${AGENT_SOURCE}', count(DISTINCT visitor) FROM ag HAVING count(*) > 0`)}
+      UNION ALL SELECT key, value::int FROM a, jsonb_each_text(a.sources)
     )
     SELECT json_build_object(
       ${headline('')},
       'previous', json_build_object(${headline('_p')}, 'follow_clicks', ${follows ? `(SELECT count(*) FROM e_p WHERE ${follows})` : 'NULL'}),
       'counted_from', (SELECT ${day('min("createdAt")')} FROM "View" WHERE "siteId" = $1),
+      'engaged_from', (SELECT ${day('min(at)')} FROM (SELECT at FROM viewed_p WHERE engaged UNION ALL SELECT at FROM viewed WHERE engaged) t),
       'top_sources', ${list('SELECT source, sum(visitors) AS visitors FROM src GROUP BY 1', 6)},
-      'top_pages', ${list(merged('page', 'pages', `SELECT path, count(DISTINCT visitor) FROM v GROUP BY 1${agentsToo('SELECT path, count(DISTINCT visitor) FROM ag WHERE path IS NOT NULL GROUP BY 1')}`), 6)},
-      'entry_pages', ${list(merged('page', 'entryPages', `SELECT entry, count(DISTINCT visitor) FROM visits GROUP BY 1${agentsToo('SELECT entry, count(*) FROM (SELECT (array_agg(path ORDER BY at))[1] AS entry FROM ag GROUP BY visitor) f WHERE entry IS NOT NULL GROUP BY 1')}`), 6)},
+      'top_pages', ${list(merged('page', 'pages', 'SELECT path, count(DISTINCT visitor) FROM v GROUP BY 1'), 6)},
+      'entry_pages', ${list(merged('page', 'entryPages', 'SELECT entry, count(DISTINCT visitor) FROM visits GROUP BY 1'), 6)},
       'from_posts', ${list(`SELECT entry AS page, count(DISTINCT visitor) AS visitors FROM visits WHERE source = 'x' GROUP BY 1`, 15)},
       'sibling_referrals', ${list(`SELECT source, sum(visitors) AS visitors FROM src WHERE source = ANY(${$(siblings)}::text[]) GROUP BY 1`, 6)},
-      'countries', ${list(merged('country', 'countries', `SELECT country, count(DISTINCT visitor) FROM v WHERE country IS NOT NULL GROUP BY 1${agentsToo('SELECT country, count(DISTINCT visitor) FROM ag WHERE country IS NOT NULL GROUP BY 1')}`), 10)},
-      'devices', ${list(merged('device', 'devices', `SELECT device, count(DISTINCT visitor) FROM v GROUP BY 1${agentsToo(`SELECT '${AGENT_DEVICE}', count(DISTINCT visitor) FROM ag HAVING count(*) > 0`)}`))},
-      'by_day', ${list(`SELECT date, sum(visitors) AS visitors, sum(pageviews) AS pageviews, sum(visits) AS visits, sum(bounces) AS bounces,
-          sum(seconds) AS seconds, sum(agents) AS agents, sum(everyone) AS visitors_incl_agents, sum(follows) AS follow_clicks FROM (
-          SELECT ${day('at')}, count(DISTINCT visitor), count(*), 0, 0, 0, 0, 0, 0 FROM v GROUP BY 1
-          UNION ALL SELECT ${day('at')}, 0, 0, count(*), 0, coalesce(sum(seconds), 0), 0, 0, 0 FROM visits GROUP BY 1
-          UNION ALL SELECT ${day('at')}, 0, 0, 0, count(*), 0, 0, 0, 0 FROM bounced GROUP BY 1
-          UNION ALL SELECT ${day('at')}, 0, 0, 0, 0, 0, count(DISTINCT visitor), 0, 0 FROM ag GROUP BY 1
-          UNION ALL SELECT ${day('at')}, 0, 0, 0, 0, 0, 0, count(DISTINCT visitor), 0 FROM (SELECT at, visitor FROM v UNION ALL SELECT at, visitor FROM ag) u GROUP BY 1
-          ${follows ? `UNION ALL SELECT ${day('at')}, 0, 0, 0, 0, 0, 0, 0, count(*) FROM e WHERE ${follows} GROUP BY 1` : ''}
-          UNION ALL SELECT to_char(day, 'YYYY-MM-DD'), visitors, pageviews, visits, bounces, seconds, 0, visitors, 0 FROM a
-        ) t(date, visitors, pageviews, visits, bounces, seconds, agents, everyone, follows) GROUP BY 1`, undefined, '1')},
+      'countries', ${list(merged('country', 'countries', 'SELECT country, count(DISTINCT visitor) FROM v WHERE country IS NOT NULL GROUP BY 1'), 10)},
+      'devices', ${list(merged('device', 'devices', 'SELECT device, count(DISTINCT visitor) FROM v GROUP BY 1'))},
+      'by_day', ${list(`SELECT date, sum(visitors) AS visitors, sum(engaged) AS engaged, sum(pageviews) AS pageviews, sum(visits) AS visits, sum(bounces) AS bounces,
+          sum(agents) AS agents, sum(follows) AS follow_clicks FROM (
+          SELECT ${day('at')}, count(DISTINCT visitor), count(DISTINCT visitor) FILTER (WHERE engaged), count(*), 0, 0, 0, 0 FROM v GROUP BY 1
+          UNION ALL SELECT ${day('at')}, 0, 0, 0, count(*), count(*) FILTER (WHERE bounced), 0, 0 FROM visits GROUP BY 1
+          UNION ALL SELECT ${day('at')}, 0, 0, 0, 0, 0, count(DISTINCT visitor), 0 FROM ag GROUP BY 1
+          ${follows ? `UNION ALL SELECT ${day('at')}, 0, 0, 0, 0, 0, 0, count(*) FROM e WHERE ${follows} GROUP BY 1` : ''}
+          UNION ALL SELECT to_char(day, 'YYYY-MM-DD'), visitors, 0, pageviews, visits, 0, 0, 0 FROM a
+        ) t(date, visitors, engaged, pageviews, visits, bounces, agents, follows) GROUP BY 1`, undefined, '1')},
       'follow_clicks', ${follows ? `(SELECT json_build_object('events', count(*), 'visitors', count(DISTINCT visitor)) FROM e WHERE ${follows})` : 'NULL'},
       'top_agents', ${list('SELECT agent, count(DISTINCT visitor) AS visitors FROM ag GROUP BY 1', 10)},
       'agent_tool_calls', ${list(`SELECT coalesce(props->>'tool', '(none)') AS tool, count(*) AS events FROM e
@@ -316,7 +354,7 @@ export async function digest(sql, { site = '', days, handle, siblings = [], gaps
             for (const k of Object.keys(p))
                 p[k] = null;
         else
-            p.agents = p.visitors_incl_agents = p.follow_clicks = null;
+            p.agents = p.follow_clicks = p.engaged = p.bounce_rate = null;
     }
     return d;
 }

package/dist/dom.d.ts

@@ -111,6 +111,11 @@ export declare function scrollTables(html: string): string;
  * A click on a link to another host is sent as an "Outbound Link: Click" event
  * carrying its `url`. Starts once per page load however often it is called, so
  * React's doubled effects in development send one view, not two.
+ *
+ * Each view is then marked engaged, `{ path, engaged: true }`, once its reader
+ * has had the page in a visible tab for `ENGAGED_MS`, or touches, clicks, types
+ * or wheels on it first. A crawler that runs scripts renders a page and leaves
+ * within seconds and gives it no input, so engaged views are the people's.
  */
 export declare function startBeacon(endpoint?: string): void;
 /** Record a named event from the browser, on the endpoint `startBeacon` was given. */

package/dist/dom.js

@@ -330,6 +330,8 @@ export function scrollTables(html) {
 // ---- analytics beacon -------------------------------------------------------
 let beaconEndpoint = '/api/view';
 let beaconStarted = false;
+/** How long a view's page must be visible, in all, before its reader counts as engaged. */
+const ENGAGED_MS = 15_000;
 // A browser under automation (Puppeteer, Playwright, Selenium) sets
 // `navigator.webdriver`: a crawler that runs scripts, whose user agent the bot
 // pattern cannot tell from a reader's, so it sends nothing at all.
@@ -345,6 +347,11 @@ const sendBeacon = (body) => navigator.webdriver || navigator.sendBeacon?.(beaco
  * A click on a link to another host is sent as an "Outbound Link: Click" event
  * carrying its `url`. Starts once per page load however often it is called, so
  * React's doubled effects in development send one view, not two.
+ *
+ * Each view is then marked engaged, `{ path, engaged: true }`, once its reader
+ * has had the page in a visible tab for `ENGAGED_MS`, or touches, clicks, types
+ * or wheels on it first. A crawler that runs scripts renders a page and leaves
+ * within seconds and gives it no input, so engaged views are the people's.
  */
 export function startBeacon(endpoint = '/api/view') {
     if (beaconStarted || typeof window === 'undefined')
@@ -353,6 +360,30 @@ export function startBeacon(endpoint = '/api/view') {
     beaconEndpoint = endpoint;
     let last = '';
     let first = true;
+    /** The path whose view is not yet engaged, the visible time it has had, and since when it has been visible. */
+    let pending = null;
+    let seen = 0;
+    let since = 0;
+    let timer;
+    const engage = () => {
+        clearTimeout(timer);
+        if (pending !== null)
+            sendBeacon({ path: pending, engaged: true });
+        pending = null;
+    };
+    const watch = () => {
+        clearTimeout(timer);
+        if (pending === null)
+            return;
+        if (document.visibilityState === 'visible') {
+            since = Date.now();
+            timer = setTimeout(engage, ENGAGED_MS - seen);
+        }
+        else if (since) {
+            seen += Date.now() - since;
+            since = 0;
+        }
+    };
     const view = () => {
         if (location.pathname === last)
             return;
@@ -364,7 +395,15 @@ export function startBeacon(endpoint = '/api/view') {
             utm: new URLSearchParams(location.search).get('utm_source') ?? undefined,
         });
         first = false;
+        pending = last;
+        seen = since = 0;
+        watch();
     };
+    document.addEventListener('visibilitychange', watch);
+    // Not `scroll`, which a script fires as readily as a reader. Capture, so a tap on a link engages the page it leaves.
+    for (const type of ['pointerdown', 'keydown', 'wheel', 'touchstart']) {
+        addEventListener(type, e => e.isTrusted && engage(), { capture: true, passive: true });
+    }
     for (const method of ['pushState', 'replaceState']) {
         const write = history[method];
         history[method] = function (...args) {

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.d.ts

@@ -6,10 +6,9 @@ import { type Digest } from './analytics.js';
  * day from `first` (a UTC day number) to today. A count's empty day is a zero;
  * a rate's is left out, since no visits is not a bounce rate of nothing.
  */
-export declare function StatsFigures({ digest, first, apart, controls }: {
+export declare function StatsFigures({ digest, first, controls }: {
     digest: Digest;
     first: number;
-    apart: boolean;
     controls: ReactNode;
 }): import("react").JSX.Element;
 export interface PasskeyButtonProps {

package/dist/stats-figures.js

@@ -9,10 +9,11 @@ import { startAuthentication, startRegistration } from '@simplewebauthn/browser'
 import { useMemo, useState } from 'react';
 import { count } from './analytics.js';
 import { TimeChart, stepDecimals } from './chart.js';
-// Whole seconds and percentages, unless an axis tick's step needs more to tell it from the next.
+// Whole percentages, unless an axis tick's step needs more to tell it from the next.
 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
@@ -26,22 +27,24 @@ function change(now, before, { points = false, lowerIsBetter = false } = {}) {
     const text = `${n > 0 ? '+' : ''}${n}${points ? ' pp' : '%'}`;
     return n ? { text, better: n > 0 !== lowerIsBetter } : { text };
 }
-/** Visitors are one figure counting agents too, or `apart` two: people, and agents. */
-const figuresOf = ({ previous: p, ...d }, apart) => [
-    ...(apart
-        ? [
-            { label: 'People', value: count(d.visitors), delta: change(d.visitors, p.visitors), day: (r) => r.visitors },
-            { label: 'Agents', value: count(d.agents), delta: change(d.agents, p.agents), day: (r) => r.agents },
-        ]
-        : [
-            {
-                label: 'Visitors',
-                value: count(d.visitors_incl_agents),
-                delta: change(d.visitors_incl_agents, p.visitors_incl_agents),
-                day: (r) => r.visitors_incl_agents,
-            },
-        ]),
+/**
+ * People, the ones among them who engaged, and agents, never one count of both.
+ * Engaged and bounce rate are drawn from the day engagement was first counted:
+ * before it is nothing, not days of no one engaging.
+ */
+const figuresOf = ({ previous: p, ...d }) => [
+    { label: 'People', value: count(d.visitors), delta: change(d.visitors, p.visitors), day: r => r.visitors },
+    { label: 'Engaged', value: count(d.engaged), delta: change(d.engaged, p.engaged), day: r => r.engaged, from: d.engaged_from ?? '9999' },
+    { label: 'Agents', value: count(d.agents), delta: change(d.agents, p.agents), day: r => r.agents },
     { 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),
@@ -49,14 +52,7 @@ const figuresOf = ({ previous: p, ...d }, apart) => [
         day: r => (r.visits ? (100 * r.bounces) / r.visits : null),
         format: percent,
         aggregate: 'mean',
-    },
-    {
-        label: 'Visit length',
-        value: d.visit_duration == null ? '–' : duration(d.visit_duration),
-        delta: change(d.visit_duration, p.visit_duration),
-        day: r => (r.visits ? r.seconds / r.visits : null),
-        format: duration,
-        aggregate: 'mean',
+        from: d.engaged_from ?? '9999',
     },
     ...(d.follow_clicks
         ? [
@@ -70,16 +66,16 @@ const figuresOf = ({ previous: p, ...d }, apart) => [
         : []),
 ];
 const DAY = 86_400;
-const EMPTY = { date: '', visitors: 0, pageviews: 0, visits: 0, bounces: 0, seconds: 0, agents: 0, visitors_incl_agents: 0, follow_clicks: 0 };
+const EMPTY = { date: '', visitors: 0, engaged: 0, pageviews: 0, visits: 0, bounces: 0, agents: 0, follow_clicks: 0 };
 /**
  * The figures as `.stats-figures button[aria-pressed]`, each with its change
  * as `.stats-delta[data-better]` for a stylesheet to colour, and the pressed one per
  * day from `first` (a UTC day number) to today. A count's empty day is a zero;
  * a rate's is left out, since no visits is not a bounce rate of nothing.
  */
-export function StatsFigures({ digest, first, apart, controls }) {
-    const figures = useMemo(() => figuresOf(digest, apart), [digest, apart]);
-    // By name, since a window or the agents switch can arrive without a reload and the figures shift.
+export function StatsFigures({ digest, first, controls }) {
+    const figures = useMemo(() => figuresOf(digest), [digest]);
+    // By name, since a window can arrive without a reload and the figures shift (Clicks to X comes and goes).
     const [chosen, choose] = useState(figures[0].label);
     const figure = figures.find(f => f.label === chosen) ?? figures[0];
     const byDay = useMemo(() => new Map(digest.by_day.map(r => [r.date, r])), [digest.by_day]);
@@ -87,7 +83,8 @@ export function StatsFigures({ digest, first, apart, controls }) {
         const today = Math.floor(Date.now() / 1000 / DAY);
         return Array.from({ length: today - first + 1 }, (_, i) => {
             const time = (first + i) * DAY;
-            const value = figure.day(byDay.get(new Date(time * 1000).toISOString().slice(0, 10)) ?? EMPTY);
+            const date = new Date(time * 1000).toISOString().slice(0, 10);
+            const value = figure.from && date < figure.from ? null : figure.day(byDay.get(date) ?? EMPTY);
             return value == null ? null : { time, value };
         }).filter(p => p !== null);
     }, [byDay, first, figure]);

package/dist/stats.d.ts

@@ -9,8 +9,6 @@ export interface StatsProps {
     days: number;
     /** What the digest was narrowed to, as `digest` was given it. */
     filter?: StatsFilter;
-    /** Visitors as people and agents, two figures, rather than one that counts both. */
-    apart?: boolean;
     /** The windows offered as `?days=` links on the same page. */
     ranges?: readonly number[];
     /** Every link on the page: `next/link`, say, so a window or a filter does not reload it. Each is given `prefetch={false}`. */
@@ -24,17 +22,16 @@ export interface StatsProps {
 export declare function statsQuery(params: Record<string, string | string[] | undefined>, ranges?: readonly number[]): {
     days: number;
     filter: Partial<Record<"page" | "country" | "entry" | "source" | "device", string>>;
-    apart: boolean;
 };
 /**
  * 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
+ * before, people and agents 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
- * the lists count the visitors the chart draws. The chart and the bars are `currentColor`
+ * 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 chart and the bars are `currentColor`
  * until the importing stylesheet gives them a colour.
  */
-export declare function Stats({ digest: d, days, filter, apart, ranges, link: Link }: StatsProps): import("react").JSX.Element;
+export declare function Stats({ digest: d, days, filter, ranges, link: Link }: StatsProps): import("react").JSX.Element;
 export interface StatsGateProps {
     gate: PasskeyGate;
     /** The gate's cookie from the request: `(await cookies()).get(gate.cookie)?.value`. */

package/dist/stats.js

@@ -22,14 +22,11 @@ export function statsQuery(params, ranges = STATS_RANGES) {
         if (typeof v === 'string' && v)
             filter[k] = v.slice(0, 500);
     }
-    // `apart` is `digest`'s option too, so its lists count the visitors the chart draws.
-    return { days: statsDays(params.days, ranges), filter, apart: params.agents === 'apart' };
+    return { days: statsDays(params.days, ranges), filter };
 }
-/** The page at a window, a filter and a count of agents, as a query string. */
-const href = (days, filter, apart) => {
+/** The page at a window and a filter, as a query string. */
+const href = (days, filter) => {
     const q = new URLSearchParams({ days: daysParam(days) });
-    if (apart)
-        q.set('agents', 'apart');
     for (const k of STATS_FILTERS)
         if (filter[k])
             q.set(k, filter[k]);
@@ -46,21 +43,21 @@ 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))) })] }));
 }
 /**
  * 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
+ * before, people and agents 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
- * the lists count the visitors the chart draws. The chart and the bars are `currentColor`
+ * 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 chart and the bars are `currentColor`
  * until the importing stylesheet gives them a colour.
  */
-export function Stats({ digest: d, days, filter = {}, apart = false, ranges = STATS_RANGES, link: Link = Anchor }) {
+export function Stats({ digest: d, days, filter = {}, ranges = STATS_RANGES, link: Link = Anchor }) {
     const today = Math.floor(Date.now() / DAY_MS);
     const counted = d.counted_from ? Date.parse(d.counted_from) / DAY_MS : today;
     const narrowed = STATS_FILTERS.filter(k => filter[k]);
@@ -72,12 +69,11 @@ export function Stats({ digest: d, days, filter = {}, apart = false, ranges = ST
         ? Math.max(start, counted)
         : Number.isFinite(start) ? start : Math.min(today, ...d.by_day.map(r => Date.parse(r.date) / DAY_MS));
     const rows = (list, k, m) => list.map(r => [r[k], r[m]]);
-    const at = (f, n = days, a = apart) => href(n, f, a);
+    const at = (f, n = days) => href(n, f);
     const to = (k) => (v) => at({ ...filter, [k]: v });
-    // 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') })] })] }));
+    // Each window is another digest, so a link rather than a button, pinned with the chart's own controls.
+    const controls = (_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))) }));
+    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, 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": "0.105.0",
   "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",

package/prisma/analytics.prisma

@@ -14,6 +14,8 @@ model View {
   country   String?
   device    String
   visitor   String
+  // Set once its reader has had the page visible fifteen seconds or given it input.
+  engaged   Boolean  @default(false)
   createdAt DateTime @default(now()) @db.Timestamptz(3)
 
   @@index([siteId, createdAt])