Skip to main content

Level 1: First insights in ten minutes

Ten minutes on a timer: traffic in, insight out. The first level of the usage analytics onboarding path proves that requests sent through the gateway become visible, costed, and explainable within minutes. It is played against the clock, and the clock is part of the result: time-to-first-insight is one of the numbers an evaluation exists to produce.

Persona: Developer or engineering leader working in the Developer Console.

Estimated time: 10 minutes, played against a timer.

Progression: Level 1 of 3 in the usage analytics onboarding path. Completing it unlocks Level 2: Read the shadow bill.

The rules​

  • A timer starts at Step 1 and runs until the checklist at the bottom is fully ticked.
  • Each step has a minute budget and ends with a time check. A failed time check is not a failure of the run; note where the time went, because that observation is itself an evaluation finding.
  • The elapsed time of the first full run is the personal best. Every later run tries to beat it.

Outcomes​

By the end of this level:

  • A burst of requests has been sent and located in Request Logs within minutes of sending.
  • The cost, token counts, and latency of a single request have been read from its detail panel and explained.
  • The most expensive request of the run has been identified, with the reason articulated in one sentence.

Prerequisites​

  • A working API key, as issued in Make an API Call.
  • A way to send a handful of requests: a terminal with curl, an SDK snippet, or a coding agent pointed at the gateway.
  • A timer.

Step 1 (minutes 0--2): open the scoreboard​

  1. Start the timer.
  2. Sign in to the Console.
  3. Open Monitoring → Request Logs in one browser tab and Monitoring → Usage in another.

These two pages are the scoreboard for the rest of the onboarding path; everything the next two levels measure is read from them.

Time check, minute 2: both pages are open, showing either existing traffic or an empty state.

Step 2 (minutes 2--5): send a burst​

  1. Send 5--10 requests through the gateway: a short script loop, a few Playground exchanges, or one small task given to a coding agent.
  2. Vary the prompts; identical requests make the later steps less interesting.
  3. Refresh Request Logs.

Time check, minute 5: the burst appears at the top of Request Logs, newest first.

Step 3 (minutes 5--8): read one request end to end​

  1. Click any row from the burst to open its detail panel.
  2. Read, in order: the resolved model, input and output token counts, the cost calculation, and the latency breakdown (time-to-first-token and total).

Time check, minute 8: the cost of a single request can be stated in dollars and explained as tokens multiplied by the model's per-token pricing.

Step 4 (minutes 8--10): find the big ticket​

  1. Scan the burst for the request with the highest value in the Cost column.
  2. Open it and explain why it cost the most; the longest output is the usual culprit.

Time check, minute 10: the priciest request of the run is identified, and the reason fits in one sentence.

Level complete​

  • The burst was located in Request Logs within minutes of sending.
  • One request's cost, tokens, and latency were read and explained.
  • The most expensive request was identified and the reason articulated.
  • The elapsed time was recorded as the personal best.
Bonus challenge: speedrun

Run the level again from a cold start (new browser session, fresh burst). A second run under five minutes demonstrates that the workflow, not the learning curve, is what takes the time. Log both times; the gap between them is the onboarding cost for the next person.

Where to go next

Level 2: Read the shadow bill

Put a dollar figure on what subscription traffic would cost at API list prices.

Monitor traffic and usage

The full reference for Request Logs and Usage Analytics, beyond the timed tour.