How to Publish a GPU Ready Rate That Readers Can Check

Steps

1. 1

   Start here

   A reliability number is only useful if readers can see what it counts. Here is the checklist we follow on our public Capacity Index, so your team can copy it. A playbook for teams that publish performance figures. Company-prepared, not independent reporting.

2. 2

   Define "ready" before you show a number

   Ours: a launch whose machine passed a health check at least once, either a healthy provider check or a healthy in-machine agent report. Acceptance by the underlying infrastructure alone does not count.

3. 3

   Split the two kinds of failure

   "No machine started" means the order could not be placed. "Machine did not become ready" means a machine started but never passed its check. They have different causes and different fixes. As of Oct. 19, 2026, 17:49 UTC, we show 679 launches: 446 ready, 103 no machine started, 130 machine did not become ready.

4. 4

   Say who ran the traffic

   We label launches as internal (490), sponsored (181) and outside (8). Most launches in our data are our own tests, and the page says so.

5. 5

   Show routine traffic separately

   A "Normal routing" view (230 launches) keeps forced test runs from blending into the headline.

6. 6

   Set a minimum sample size

   Any GPU type with fewer than 20 launches shows "Not enough data yet" instead of a rate.

7. 7

   Timestamp every number and state its scope

   Each figure carries an as-of time. Our counts start on Sep. 21, 2026, and the page notes that 32 earlier launches are excluded.

8. 8

   Publish what the results do not show

   Our reports include a section on exactly that: small samples, detection times longer than our target, and the fact that we prepared the reports ourselves. When a run was invalid because of a defect in our own test script, we marked it "Not a test result" and said why.

9. 9

   Let outsiders check your work, and show what changed

   We published Outside Tester Report 1: nine findings, nine changes.

10. 10

    What this does not claim, and sources

    These figures come from our own job records. They are not a customer success rate and have not been independently verified. Live figures and definitions: https://www.kilawattcloud.dev/capacity-index?utm_source=epismo&utm_medium=playbook&utm_campaign=ready-rate-checklist Our published reports, including the misses: https://www.kilawattcloud.dev/super-intelligence?utm_source=epismo&utm_medium=playbook&utm_campaign=ready-rate-checklist Kilawatt Cloud is a trade name of Right Recruit LLC. This is not an offer and contains no forecast.