fix(calendar): surface degraded get_calendar_events sentinel so web fallback fires

[Copilot]

Analysis of the last 5 agentic-workflow PRs (#2819, #2821, #2822, #2823, #2824) for degraded datasources surfaced a client masking bug: when data.riksdagen.se/kalender/ serves an HTML error page, the riksdag-regering MCP get_calendar_events tool returns a successful result wrapping a sentinel {count:0, events:[], error, rawHtml}. The clients read the empty events:[] as a legitimate zero-event window, so scripts/calendar-fetch.ts reported a false status:ok, path:mcp-primary and the web-scraper fallback never fired — masking the outage as a real zero-sitting week.

Client code

• scripts/fetch-calendar/mcp/errors.ts — new isDegradedKalenderSentinel() detector; triggers only on a non-empty error string or rawHtml, so a clean {count:0, events:[]} stays a valid empty window.
• scripts/fetch-calendar/mcp/client.ts — detect the sentinel in both the content-wrapped and direct-result paths and throw CalendarMcpError(..., 'html', rawHtml); orchestrator.ts already breaks the retry loop on kind:'html' and falls straight to the public-page scraper.
• scripts/mcp-client/methods/calendar.ts — mirror the detection (throw instead of returning []).
• scripts/fetch-calendar.ts — re-export the new detector.

if (isDegradedKalenderSentinel(inner)) {
  throw new CalendarMcpError(
    `MCP kalender API degraded: ${String(inner['error'] ?? 'upstream HTML error')}`,
    'html',
    typeof inner['rawHtml'] === 'string' ? inner['rawHtml'] : undefined,
  );
}

With the fix, the CLI now reports status:error, path:none during the outage instead of a fabricated zero-event success.

Agentic config / prompts / instructions

• analysis/templates/session-baseline.md — added a 📅 Calendar sourcing degraded-source guard and pointed the 3 inline get_calendar_events references at the resilient data/runtime/calendar-status.json / scripts/calendar-fetch.ts.
• .github/prompts/02-mcp-access.md — added a data-sourcing rule: fetch the calendar via the resilient CLI / pre-warm artifact (never the raw tool); record [DATA GAP: calendar source degraded] when both paths fail.

Tests

• Regression coverage across tests/fetch-calendar/mcp/errors.test.ts, tests/fetch-calendar/mcp/client.test.ts (content + direct sentinel → kind:'html'), tests/fetch-calendar/orchestrator-integration.test.ts (sentinel → web-fallback), and tests/mcp-client/methods/calendar.test.ts (sentinel → throws).

Out of scope (documented)

The public fallback page moved from /sv/kalendarium/ (now 404) to a Next.js /sv/aktuellt/kalendersida/ serving data in __NEXT_DATA__; repointing the HTML scraper needs a parser rewrite plus markup fixtures. The now-honest status:error/path:none ensures the degradation is no longer silently masked in the meantime.

IMF (no change)

The other degraded source flagged in the manifests is already mitigated by the 3-tier WEO pipeline and data/imf-context.json degraded/unavailable handling.

[github-actions]

🏷️ Automatic Labeling Summary

This PR has been automatically labeled based on the files changed and PR metadata.

Applied Labels: documentation,testing,refactor,size-m

Label Categories

• 🗳️ Content: news, dashboard, visualization, intelligence
• 💻 Technology: html-css, javascript, workflow, security
• 📊 Data: cia-data, riksdag-data, data-pipeline, schema
• 🌍 I18n: i18n, translation, rtl
• 🔒 ISMS: isms, iso-27001, nist-csf, cis-controls
• 🏗️ Infrastructure: ci-cd, deployment, performance, monitoring
• 🔄 Quality: testing, accessibility, documentation, refactor
• 🤖 AI: agent, skill, agentic-workflow

For more information, see .github/labeler.yml.

[github-actions]

🔍 Lighthouse Performance Audit

Category	Score	Status
Performance	85/100	🟡
Accessibility	95/100	🟢
Best Practices	90/100	🟢
SEO	95/100	🟢

📥 Download full Lighthouse report

Budget Compliance: Performance budgets enforced via budget.json