feat(v1.8): privacy-friendly analytics + search insights (closes #33) (#40)

Schema (Drizzle migration 0006)
-------------------------------
- page_views: composite PK on (date, path), kind enum, optional
  refId, integer count. UPSERT path so we get one row per day per
  path with atomic increments.
- search_log: anonymized {term, noResults, date} only. No IP, no UA,
  no fingerprint.

Tracker
-------
- $lib/server/analytics/index.ts:
  - trackView(db, { path, kind, refId }, consent) — gated on
    consent.analytics from the v1.7a cookie record. UPSERT bumps
    the counter atomically. Best-effort: errors never break a public
    page render.
  - logSearch(db, term, noResults) — always written when /blog?q= is
    used. Search is functional, not analytics-gated.
- AnalyticsService: topPaths / topArticles / topSearchTerms /
  topNoResultTerms / sparkline / totalViews. All scoped to a
  trailing N days (default 30). Sparkline densifies missing days
  to 0 so the chart line stays continuous.

Instrumentation
---------------
- Public home, blog index, blog slug, page catch-all all call
  trackView. Blog index also calls logSearch when ?q= is set, with
  noResults flagged when articles.items.length === 0.

Dashboard
---------
- "Top articles (30 days)" tile: resolves refId → article title +
  link to the editor. Falls back to the raw path when refId is
  null (e.g. blog index).
- "Search insights (30 days)" tile: split into Most-searched terms
  (clickable through to /blog?q=…) and Searches with no results
  (content-gap list). Both surface a non-scary "no data yet" empty
  state.

Per-article sparkline
---------------------
- Article edit page loads a 30-day series for /{locale}/blog/{slug}
  across every supported locale, merges by date, renders a tiny SVG
  sparkline + 30-day total. Hidden when total = 0 so a fresh article
  doesn't show a flatline.

Cloudflare Web Analytics (optional)
-----------------------------------
- /cms/settings gets a "Cloudflare Web Analytics token" field.
- When set AND data.consent.analytics, the (www) layout injects the
  official beacon. Off by default. The first-party D1 counter runs
  regardless of this setting.

Also includes a small svelte-check-only typing workaround on
/cms/blocks for the cms_blocks_help paraglide message (build was
already green; the cast just silences the per-message-types noise).

i18n: 9 new keys (EN + TH).

Closes #33

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

Author: thunpisit

README.md

@@ -282,7 +282,7 @@ Khao Pad started as a CMS. Through v1.5 it became a complete content layer (writ
 | **v1.5** | Content versioning          | ✅ Shipped    | Per-article revision history, line diff, one-click restore, attribution                             |
 | **v1.6** | SEO foundations             | ✅ Shipped    | Per-page meta, sitemap, robots, JSON-LD, RSS/Atom, slug redirects, SEO scoring hint                 |
 | **v1.7** | Pages, navigation, IA       | ✅ Shipped    | Media folders, reusable blocks, cookie consent, static pages, navigation manager, seed:legal |
-| **v1.8** | Analytics & insight         | 🚧 Pending   | Privacy-friendly D1 page-views, top articles, search-term insights, per-article sparkline          |
+| **v1.8** | Analytics & insight         | ✅ Shipped    | Privacy-friendly D1 page-views, top articles, search-term insights, per-article sparkline, optional CWA |
 | **v1.9** | Performance & trust         | 🚧 Pending   | Cloudflare Images responsive `srcset`, cache headers, custom 404/500, cookie consent, health check |
 | **v2.0** | Engagement & growth         | 🚧 Pending   | Forms, newsletter, comments, webhooks, public read-only API                                         |
 

docs/MILESTONES.md

@@ -208,15 +208,23 @@ A blog isn't a website. A website needs static pages (About, Contact, Privacy),
 
 **i18n** — 18 keys in v1.7a + 23 keys in v1.7b (EN + TH).
 
-### v1.8 — Analytics and insight (pending)
+### v1.8 — Analytics and insight
 
-**Page-view counter** — Edge-side, privacy-respecting. New `analytics_events` table on D1 with a `Cloudflare Worker` middleware that increments a counter per `(path, day)` on every public page request, no IP/UA stored. Aggregates via the queue worker so the request path stays sub-millisecond. Optional integration with Cloudflare Web Analytics for the deeper view (referrers, devices, countries) — opt-in via a setting.
+Privacy-friendly editor analytics. Closes the "what's working?" gap left after v1.7. Every counter is gated on the v1.7a cookie consent — visitors who opt out write nothing. Aggregated by `(date, path)` so a busy site stays bounded. **No IP, no user agent, no fingerprint stored.**
 
-**Editor analytics on the dashboard** — Top 10 articles by views (last 30 days), top search terms (last 30 days, from `searchArticles` calls), draft-to-publish median time, articles updated this week. Sourced from `analytics_events` + `audit_log`. Adds a "Performance" card to the dashboard.
+**Schema** — Drizzle migration 0006: `page_views` (composite PK on `(date, path)`, kind enum, optional `refId`, integer `count`) and `search_log` (anonymized `term` + `noResults` flag + `date`).
 
-**Per-article analytics** — On each article edit page, a sparkline of the last 30 days of views, pulled from `analytics_events`.
+**Tracker** — `$lib/server/analytics/index.ts` exposes `trackView(db, opts, consent)` and `logSearch(db, term, noResults)`. View tracking does an UPSERT on the composite key — `INSERT … ON CONFLICT DO UPDATE SET count = count + 1` — so one row per day per path, atomic increments. Both functions are best-effort: a write failure never breaks the public page render. Tracking is instrumented on the home, blog index, blog slug, and the v1.7b page catch-all routes; the search log fires from `/blog?q=` whether or not analytics consent is given (search itself is functional, not analytics).
 
-**Search-term insights** — Log every public-facing `searchArticles` query (anonymized, just the term + date) into a new `search_log` table. Dashboard tile: "Search terms with no results" — the high-signal list of content gaps to fill.
+**`AnalyticsService`** — Read surface used by the dashboard + article edit page. Methods: `topPaths` / `topArticles` / `topSearchTerms` / `topNoResultTerms` / `sparkline(path, days)` / `totalViews(path, days)`. All scoped to the last `days` (default 30) so the queries stay fast. The sparkline densifies the result so empty days return `count = 0` and the chart line stays continuous.
+
+**Dashboard tiles** — Two new cards (admin / editor / author all see them — the data is non-sensitive aggregate). "Top articles (30 days)" resolves `refId → article` once at load time so the list shows real titles, not raw paths. "Search insights" splits into two stacks: most-searched terms (with click-through to `/blog?q=…`) and searches with no results (the content-gap list). Both surface a "no data yet" empty state that explains how to seed counters.
+
+**Per-article sparkline** — Article edit page (`/cms/articles/[id]`) loads a 30-day series for the article's slug across every supported locale, merges them by date, and renders a tiny SVG sparkline above the form alongside a 30-day total. Runs lazily — the component returns nothing if there are no views yet, so a fresh article doesn't get a flat-line chart. All best-effort: the form still loads if analytics is unavailable.
+
+**Cloudflare Web Analytics opt-in** — New `cfaToken` field in `/cms/settings`. When set AND the visitor consented to analytics, the (www) layout injects the official `https://static.cloudflareinsights.com/beacon.min.js` script with the operator's site token. Off by default; the first-party D1 counter runs regardless of this setting. No tracking code unless both conditions are met.
+
+**i18n** — 9 new keys (EN + TH).
 
 ### v1.9 — Performance and trust (pending)
 

drizzle/0006_adorable_microchip.sql

@@ -0,0 +1,16 @@
+CREATE TABLE `page_views` (
+	`date` text NOT NULL,
+	`path` text NOT NULL,
+	`kind` text NOT NULL,
+	`ref_id` text,
+	`count` integer NOT NULL,
+	PRIMARY KEY(`date`, `path`)
+);
+--> statement-breakpoint
+CREATE TABLE `search_log` (
+	`id` text PRIMARY KEY NOT NULL,
+	`term` text NOT NULL,
+	`no_results` integer NOT NULL,
+	`date` text NOT NULL,
+	`created_at` text NOT NULL
+);