Engine room

live · 30s

The serving infrastructure behind this site’s assistant, deliberately overengineered and left visible. Every number below is real and current; every answer links to its own trace.

All lanes nominal: the canary is green, the queue is idle, 332 answers served so far.

Request pipeline · all time

since 2026-09-05

  1. Requests

    955

    arrived at the edge

  2. Admission

    334

    rate + token budgets

    ↳ 320 shed

  3. Cache

    1143

    answered without a model

    ↳ 31 dedup reattached

  4. Guardrail

    12

    scoped refusals

    ↳ 97 failed open

  5. Generate

    332

    durable workflow, ladder armed

    ↳ 62 degraded to raw KB

  6. Verify

    332

    citations resolve

Time to first token

p50 644ms · p95 12.11s · 334 samples

target 2.5s0ms–516ms: 101 answers516ms–1.0s: 114 answers1.0s–1.5s: 3 answers1.5s–2.1s: 7 answers2.6s–3.1s: 31 answers3.1s–3.6s: 24 answers3.6s–4.1s: 1 answer4.1s–4.6s: 1 answer5.2s–5.7s: 2 answers5.7s–6.2s: 4 answers6.2s–6.7s: 1 answer6.7s–7.2s: 1 answer7.7s–8.3s: 2 answers8.3s–8.8s: 2 answers8.8s–9.3s: 9 answers9.3s–9.8s: 2 answers9.8s–10.3s: 3 answers10.3s–10.8s: 2 answers10.8s–11.4s: 2 answers11.4s–11.9s: 4 answers11.9s–12.4s: 7 answers12.4s–12.9s: 2 answers12.9s–13.4s: 3 answers13.4s–13.9s: 1 answer13.9s–14.5s: 2 answers15.5s–16.0s: 1 answer16.5s–17.0s: 1 answer17.5s–18.1s: 1 answerp50 644msp95 12.1s018.6s

Goodput

100.0% · 332 ok / 0 failed

Full answer

p50 4.11s · p95 18.95s

End-to-end through guardrail, retrieval, generation, and verification, measured inside the workflow.

Ledger · all time

332 answers served for $0.06 in model spend ($0.00018 per answer), 398,946 tokens through the gateway, 1174 requests absorbed by the cache and dedup layers before spending anything.

Recent answers

timeoutcomettfttotalcited
10-27 19:02glm-5.3-flash12.71s19.98s✓trace →
10-27 19:02glm-5.3-flash12.35s16.02s✓trace →
10-25 23:55glm-5.3-flash12.91s16.48s✓trace →
10-22 22:11glm-5.3-flash12.32s15.67s✓trace →
10-22 12:37glm-5.3-flash13.95s29.31s✓trace →

Register

primary model
zai/glm-5.3-flash
model override
none
prompt hash
45b79cdd7ce5
policy hash
6bb17bbe0908
breaker · primary
closed
breaker · fallback
closed
eval gate
27/32 · 2026-09-06
canary
up · 3.0s · 11:30 UTC
daily token ceiling
0 / 2,000,000 (0.0%)
kill switch
off

Counters accumulate for the site’s lifetime; latency distributions cover the last 500 answers; traces are kept for seven days and never contain visitor message content. The architecture behind this page is the subject of the writing.