Day2 [art-tracking] 看展追蹤器要做什麼、不做什麼：規劃時的六個取捨

{
  id: "bluerider",
  name: "Bluerider ART",
  kind: "website",
  url: "https://blueriderart.com/tw/exhibitions/",
  instructions:
    "靜態 HTML，分「現正展出 / 即將展出 / 過去展覽」。只取台北·敦南、台北·微風、台北·仁愛三個空間，排除上海、倫敦、洛杉磯。場館依標題前綴【台北·敦化】→ bluerider-dunnan、【台北·微風】或「微風廣場」→ bluerider-breeze、【台北·仁愛】→ bluerider-renai。展期格式如 2026.9.11–12.31（跨年時第二段可能省略年份）。",
}

export function groupExhibitions(items: ExhibitionDto[], by: GroupBy): GanttGroup[] {
  const byEnd = (a, b) => (a.endDate ?? "9999").localeCompare(b.endDate ?? "9999") || a.title.localeCompare(b.title);
  if (by === "none") return [{ key: "all", title: "全部", items: [...items].sort(byEnd) }];
  if (by === "venue") {
    const map = new Map<string, GanttGroup>();
    for (const e of items) {
      const g = map.get(e.venue.id) ?? { key: e.venue.id, title: e.venue.name, items: [] };
      g.items.push(e);
      map.set(e.venue.id, g);
    }
    return [...map.values()]
      .sort((a, b) => Number(b.items[0]!.venue.followed) - Number(a.items[0]!.venue.followed) || a.title.localeCompare(b.title, "zh-Hant"))
      .map((g) => ({ ...g, items: g.items.sort(byEnd) }));
  }
  // ...
}

export const sourceReport = z.object({
  crawlSourceId: z.string().min(1).max(64),
  status: sourceReportStatus, // ok | empty | error
  message: z.string().max(2000).optional(),
  itemCount: z.number().int().min(0),
  // Set when the agent found a replacement URL (see plan §6.5 step 2).
  discoveredUrl: httpUrl.optional(),
});

export const ingestPayload = z.object({
  trigger: z.enum(["routine", "manual"]),
  // true when only some sources were crawled (e.g. --source <id>); partial
  // runs never count toward archiving unseen exhibitions.
  partial: z.boolean().default(false),
  notes: z.string().max(5000).optional(),
  sourceReports: z.array(sourceReport).max(200).default([]),
  items: z.array(ingestItem).max(1000),
});

export const decisions = sqliteTable("decisions", {
  exhibitionId: text("exhibition_id").primaryKey().references(() => exhibitions.id, { onDelete: "cascade" }),
  kind: text("kind").notNull(), // interested | skip
  // ...
});

export const visits = sqliteTable("visits", {
  id: text("id").primaryKey(),
  exhibitionId: text("exhibition_id").notNull().references(() => exhibitions.id, { onDelete: "cascade" }),
  visitedAt: text("visited_at").notNull(),
  rating: integer("rating"),
  // ...
});

// Attached to an exhibition or a venue (exactly one).
export const notes = sqliteTable("notes", {
  id: text("id").primaryKey(),
  exhibitionId: text("exhibition_id").references(() => exhibitions.id, { onDelete: "cascade" }),
  venueId: text("venue_id").references(() => venues.id, { onDelete: "cascade" }),
  body: text("body").notNull(),
  // ...
});

export function phaseOf(today: string, startDate, endDate): Phase {
  if (startDate && today < startDate) return "upcoming";
  if (endDate && today > endDate) return "ended";
  if (startDate || endDate) return "ongoing";
  return "unknown";
}

export function statusOf(today: string, input: StatusInput): Status {
  const phase = phaseOf(today, input.startDate, input.endDate);
  const visited = input.visitCount > 0;
  const missed = input.decision === "interested" && phase === "ended" && !visited;
  return {
    phase,
    decision: input.decision,
    visited,
    missed,
    daysLeft: input.endDate ? daysBetween(today, input.endDate) : null,
    daysUntilStart: phase === "upcoming" && input.startDate ? daysBetween(today, input.startDate) : null,
  };
}

const byDecision = (a, b) => decisionRank(a.decision) - decisionRank(b.decision); // interested 0, none 1, skip 2

export const COMPARATORS: Record<SortKey, Cmp<Sortable>> = {
  composite: chain(byDecision, byEnd, byStart, byTitle),
  closing: chain(upcomingLast, byEnd, byDecision, byStart, byTitle),
  distance: chain(byDistance, byDecision, byEnd, byTitle),
  // ...
};

export function comparatorFor(sort: SortKey, opts: { closing?: boolean; distance?: boolean } = {}): Cmp<Sortable> {
  if (opts.closing && opts.distance) return chain(upcomingLast, byEnd, byDistance, byDecision, byTitle);
  if (opts.closing) return COMPARATORS.closing;
  if (opts.distance) return COMPARATORS.distance;
  return COMPARATORS[sort];
}

/**
 * Lowercase, half-width, no punctuation/brackets/whitespace, NFKC-folded.
 * "《少年行》— 台灣青年藝術家聯展 (2026)" and "少年行 台灣青年藝術家聯展 2026"
 * normalize to the same string.
 */
export function normalizeTitle(title: string): string {
  return toHalfWidth(title.normalize("NFKC"))
    .toLowerCase()
    .replace(/[\p{P}\p{S}\s]+/gu, "");
}

export function fingerprintOf(venueId: string, title: string, startDate: string | null | undefined): string {
  return `${venueId}|${normalizeTitle(title)}|${startDate ?? ""}`;
}

.onConflictDoUpdate({
  target: exhibitions.fingerprint,
  set: {
    title: item.title,
    sourceUrl: item.sourceUrl,
    endDate: sql`coalesce(${common.endDate}, ${exhibitions.endDate})`,
    imageSrcUrl: sql`coalesce(${item.imageUrl ?? null}, ${exhibitions.imageSrcUrl})`,
    lastSeenAt: startedAt,
    lastSeenRunId: partial ? sql`${exhibitions.lastSeenRunId}` : runId,
    archivedAt: null,
    // ...
  },
  // never overwrite a hand-entered row that won the race
  setWhere: sql`${exhibitions.manual} = 0 and ${exhibitions.lastSeenAt} < ${startedAt}`,
})

const failuresExpr = ok ? sql`0` : sql`${crawlSources.consecutiveFailures} + 1`;
const pausedExpr = ok
  ? crawlSources.status
  : sql`case when ${crawlSources.status} = 'active' and ${crawlSources.consecutiveFailures} + 1 >= ${PAUSE_AFTER_FAILURES} then 'paused' else ${crawlSources.status} end`;
// ...
consecutiveFailures: failuresExpr,
status: discovered ? "needs_review" : pausedExpr,
reviewReason: discovered ? `agent 自動改了 URL（原 ${cur.url}），請確認` : pausedReasonExpr,

export const ARCHIVE_AFTER_MISSED_RUNS = 3;

const recent = await db
  .select({ id: ingestRuns.id })
  .from(ingestRuns)
  // complete runs that finished without item/source errors
  .where(and(isNotNull(ingestRuns.finishedAt), isNull(ingestRuns.errors), eq(ingestRuns.partial, false)))
  .orderBy(desc(ingestRuns.startedAt), sql`rowid desc`)
  .limit(ARCHIVE_AFTER_MISSED_RUNS);
if (recent.length < ARCHIVE_AFTER_MISSED_RUNS) return 0;