English · Español
Phase 34 — Quizzes¶
🇪🇸 Espejo legible del banco de preguntas; la fuente canónica es
data/quizzes/phase-34-observability-cost.yaml.
q-34-01 — Why is request_id load-bearing?¶
Free response. Acceptable answers contain correlate.
Without request_id you cannot reconstruct the end-to-end sequence of one request — log lines from concurrent requests interleave indistinguishably.
q-34-02 — At §A13 grammar-tutor scale, what dominates self-hosted cost?¶
- Per-token API charges
- Electricity to drive the CPU during the forward pass ← correct
- GPU rental fees
- S3 storage for the log files
Why: Self-hosted has no per-token charge; the marginal cost is the few-watt-second of CPU time.
q-34-03 — Which metrics are needed for the cost-per-request dashboard?¶
- tutor_requests_total (counter) ← correct
- tutor_compute_seconds_total (counter) ← correct
- tutor_concurrent_inflight (gauge) ← correct
- cpu_thermal_throttle_events_total (counter)
q-34-04 — §A14 portal's steady-state tutor RPS?¶
- ~0.001 RPS
- ~0.3 - 1 RPS ← correct
- ~100 RPS
- ~10000 RPS
See theory/04-dollar-per-token-and-rps-budget.md for the derivation.