Beacon Platform Docs

Testing Strategy

2 min read

Comprehensive testing strategy for functional correctness, privacy safety, and AI quality.

Goals

  • Functional correctness across ingestion, parsing, and public APIs.
  • Privacy and safety (no PII leaks, guardrails enforced).
  • Reliability and idempotency for replays and retries.
  • AI quality and schema consistency.

Sources: beacon-platform/docs/testing.md, beacon-platform/apps/pulse-public/src/index.ts

Test levels

Unit tests

  • WhatsApp parsing formats and multi-line handling.
  • System event filtering and media placeholder handling.
  • Dedup normalization and hashing.
  • Privacy utilities (PII redaction, theme sanitization).
  • AI guardrails (schema validation, score clamping).
  • Sentiment trend math and thresholds.

Integration tests (mocked services / Miniflare)

  • Ingest worker: /presign, /upload, queue processing, daily and weekly generation, quota enforcement.
  • Public worker: /pulse.json, /pulse/history.json, /pulse/daily.json, /pulse/trends.json.
  • HTML rendering and embed responses.

End-to-end

  • Local or staging runs: upload → process → D1 → public endpoints.
  • Regeneration and cleanup paths.

Sources: beacon-platform/docs/testing.md, beacon-platform/apps/pulse-public/src/index.ts

AI testing

  • Golden outputs for fixed inputs with tolerance checks.
  • Failure scenarios: malformed output, privacy violations, quota exhaustion.

Privacy testing

  • Inject phones/emails/URLs and verify redaction.
  • Verify public endpoints never return raw messages or identifiers.
  • Validate theme sanitization and summary content.

Idempotency testing

  • Re-upload same export → no duplicates.
  • Replays and regenerations → consistent results.
  • Multiple queue deliveries → deduplicated output.

Sources: beacon-platform/apps/pulse-public/src/index.ts, beacon-platform/docs/testing.md

Sentiment trend computation

flowchart LR
  Weekly[Weekly summaries] --> Filter[Min weeks + confidence filter]
  Filter -->|sufficient| Regr[Confidence-weighted regression]
  Regr --> Classify{Slope + change thresholds}
  Classify --> Improving[Improving]
  Classify --> Declining[Declining]
  Classify --> Stable[Stable]
  Filter -->|insufficient| Insufficient[Insufficient data]

Sources: beacon-platform/apps/pulse-public/src/index.ts

CI pipeline

  • Pull request: lint, typecheck, unit, integration, privacy validation.
  • Nightly: full integration, golden-output checks, performance, idempotency.
  • Pre-release: staging smoke tests and public endpoint health.

Release requirements:

  • PR tests green.
  • Nightly tests green (last 24h).
  • Staging smoke tests pass.
  • No privacy regressions.
  • AI outputs within tolerance.
  • Documentation updated.

Sources: beacon-platform/docs/testing.md, beacon-platform/apps/pulse-public/src/index.ts

Graph View

Backlinks