A realistic production trace
Practice this questionWANT TO UNDERSTAND THE CONCEPT FIRST?Learn A realistic production trace
Illustrative timings and identifiers:
run r-1042, tenant t-demo, task claim_status, prompt policy-answer-v4
request span 0–1900 ms
authenticate/authorize 0–40 ms
intent model 40–340 ms 250 input / 35 output tokens
claim-status tool 340–560 ms backend request id api-762
object authorization 340–370 ms
claims API 370–550 ms status=in_review
answer model 560–1750 ms 420 input / 65 output tokens
first public text 1000 ms
output validation 1750–1810 ms
final event + message save 1810–1900 ms
outcome=answered, business_write=false, retries=0- The root span covers the complete request; child spans are portions of it and should not be added again to calculate total duration.
- The UI receives safe progress and text, not full claim data.
- Store usage metadata and source IDs, and keep costs computed from the provider's applicable pricing configuration.
- For an approval flow, separate traces can link through the same run ID rather than pretending the original HTTP request remained open for days.
Adapted from 13_AI.MD · “A realistic production trace”. Code examples are study fragments. Difficulty and core labels are editorial estimates; version-specific behavior should be checked against your project.