SLO/SLA Guidelines

• L1 (Global scale):

  • Corpus: 10^6–10^9 docs
  • Latency target (search): p95 200–800ms (method-dependent)
  • Freshness: batch/stream ingestion; provenance required

• L2 (Project scale):

  • Corpus: 10^2–10^4 docs
  • Latency target (hot set): p95 < 150ms
  • Freshness: near-real-time for project changes; curated relevance indexes

• L4 (Experience):

  • Write path: append-only with validation hooks
  • Read path: task trail queries p95 < 200ms

• L5 (Context assembly):

  • Budget policy (tokens/time); hard timeouts
  • Priority: L2 > L4 hints > L1; graceful degradation