THE PRODUCT / FOLLOW ONE REQUEST
The answer has a backstory.
Move from a high-level signal to the individual steps behind it. Select a scenario, then inspect a span.
Request explorer / Sample data
“Can I change the delivery address after ordering?”
Total time0.834s
Total cost$0.00460
Illustrative request. No live model calls or real customer data.
Request console
ErinTech demo API / Simulated requests
Awaiting a valid snapshot
Fixed demo snapshot. No live model calls or real customer data.
01 / LATENCY
A slow answer. A specific cause.
The slow scenario changes the model span, while retrieval and policy checks keep their original timings. This is why a request-level total is a starting point, rather than the whole diagnosis.
02 / COST
Every attempt belongs in the total.
Retries can improve the chance of a successful response, but they consume time and may incur additional charges. Inspect the failed attempt and the successful follow-up together.
03 / CONTEXT
Keep the explanation close.
A span should explain what happened, how long it took and how it relates to the surrounding work. Useful context makes investigation easier for the whole team.
NEXT / INSTRUMENTATION
Start with the events that matter.
Define request boundaries, meaningful span names and a policy for the data you retain. Build a useful picture without collecting everything.
Explore further ↗
FOLLOW THE SIGNAL
See the whole request.
From the first token to the final answer. Bring the detail into view.
Explore the product ↗
Tracefield is a fictional product. All metrics and integrations are illustrative.