feat: extract event descriptions from Game8 detail tables

Column tables already carried a description column, but label/value detail
tables did not — Genshin events had no blurb at all. The page follows each
detail table with a sentence of prose, so take that, skipping the
call-to-action paragraphs, which are navigation rather than description.

Summary coverage goes from 13/22 events to 22/22.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>

src/ingest/html.ts

@@ -51,10 +51,22 @@ export interface TableNode {
   rows: string[][];
 }
 
+export interface ParagraphNode {
+  kind: "p";
+  text: string;
+  /**
+   * True when the paragraph is really a call-to-action link rather than prose.
+   * Game8 wraps its "… Event Guide" buttons in the same paragraph class as body
+   * copy, so callers need to tell them apart.
+   */
+  isButton: boolean;
+}
+
 export type DocNode =
   | { kind: "h2"; text: string }
   | { kind: "h3"; text: string }
-  | TableNode;
+  | TableNode
+  | ParagraphNode;
 
 /**
  * Walk a document in source order, yielding h2/h3 headings and tables.
@@ -66,16 +78,19 @@ export function scanDocument(rawHtml: string): DocNode[] {
   const html = rawHtml.replace(/<!--[\s\S]*?-->/g, "").replace(/\s+/g, " ");
   const nodes: DocNode[] = [];
 
-  const re = /<h2\b[^>]*>([\s\S]*?)<\/h2>|<h3\b[^>]*>([\s\S]*?)<\/h3>|<table\b[^>]*>([\s\S]*?)<\/table>/gi;
+  const re =
+    /<h2\b[^>]*>([\s\S]*?)<\/h2>|<h3\b[^>]*>([\s\S]*?)<\/h3>|<table\b[^>]*>([\s\S]*?)<\/table>|<p\b[^>]*>([\s\S]*?)<\/p>/gi;
 
   for (const m of html.matchAll(re)) {
-    const [, h2, h3, table] = m;
+    const [whole, h2, h3, table, p] = m;
     if (h2 !== undefined) {
       nodes.push({ kind: "h2", text: text(h2) });
     } else if (h3 !== undefined) {
       nodes.push({ kind: "h3", text: text(h3) });
     } else if (table !== undefined) {
       nodes.push(readTable(table));
+    } else if (p !== undefined) {
+      nodes.push({ kind: "p", text: text(p), isButton: /a-btn/.test(whole) });
     }
   }
 

src/ingest/parsers/game8.ts

@@ -9,7 +9,7 @@ import {
   parseSlashDateTimeRange,
   type ParsedInstant,
 } from "../dates.ts";
-import { assertFlatTables, scanDocument } from "../html.ts";
+import { assertFlatTables, scanDocument, type DocNode } from "../html.ts";
 import type { ParseContext } from "../adapters/types.ts";
 import type { SourceParser } from "./types.ts";
 
@@ -84,7 +84,10 @@ export function parseGame8EventsPage(
   let sectionIncluded = false;
   let currentTitle: string | null = null;
 
-  for (const node of nodes) {
+  for (let i = 0; i < nodes.length; i += 1) {
+    const node = nodes[i];
+    if (node === undefined) continue;
+
     // Sections are marked by h2 on some pages and h3 on others, so inclusion is
     // tracked at whichever level actually names the section. A heading matching
     // neither list leaves the current state alone — it is an event name.
@@ -102,7 +105,7 @@ export function parseGame8EventsPage(
       continue;
     }
 
-    if (!sectionIncluded) continue;
+    if (!sectionIncluded || node.kind !== "table") continue;
 
     const fromColumns = readColumnTable(node.rows);
     if (fromColumns.length > 0) {
@@ -114,7 +117,13 @@ export function parseGame8EventsPage(
     const dates = readLabelledDates(node.pairs);
     if (dates === null) continue;
 
-    candidates.push({ title: currentTitle, summary: null, ...dates });
+    candidates.push({
+      title: currentTitle,
+      // Detail tables carry no description column, but the page follows them
+      // with a sentence of prose. That sentence is the event blurb.
+      summary: summaryAfter(nodes, i),
+      ...dates,
+    });
     // One dated table per heading; ignore follow-on reward tables until the
     // next heading.
     currentTitle = null;
@@ -123,6 +132,24 @@ export function parseGame8EventsPage(
   return dedupe(candidates.map((c) => toEvent(c, ctx)));
 }
 
+/**
+ * The event blurb that follows a detail table.
+ *
+ * Scans forward only to the next heading or table, so a description never
+ * leaks from one event onto another. Call-to-action paragraphs ("… Event
+ * Guide") are skipped — they are navigation, not description.
+ */
+function summaryAfter(nodes: DocNode[], from: number): string | null {
+  for (let i = from + 1; i < nodes.length; i += 1) {
+    const node = nodes[i];
+    if (node === undefined) break;
+    if (node.kind !== "p") break;
+    if (node.isButton || node.text.length === 0) continue;
+    return node.text.slice(0, 500);
+  }
+  return null;
+}
+
 /** Shape 1: `Event Start` / `Event End` / `Availability Period` rows. */
 function readLabelledDates(
   pairs: Array<{ label: string; value: string }>,