SYNTHETIC USAGE LOG — CAPACITY ARITHMETIC ONLY
This is invented data, not observed vendor usage. A learning unit is not a Make credit or an AI token. Fictional price: USD 0.01 per unit. No purchase is requested or authorised.

Batch A: 10 unique request revisions; base processing 40 units; retries 0 units; monitoring 4 units.
Batch B: 10 unique request revisions; base processing 42 units; retries 6 units; monitoring 4 units.
Batch C: 10 unique request revisions; base processing 38 units; retries 2 units; monitoring 4 units.

Planning rules for this exercise
Use the base average and retry average separately. Monitoring costs 4 units per batch of 10 requests in the original process. Round estimated retry units up to a whole unit. Add a 20 percent reserve to the subtotal and round the reserve up. Allocation is 600 units; the owner-selected operating stop is 480 projected units. The stop is deliberately lower than the allocation.

Worked estimates in Module 7 use 100 and 80 requests.
Independent estimate 1: plan 60 requests using the original process. Show base, retry, monitoring, reserve, total and fictional cost. State whether it fits both limits.
Independent estimate 2: the reviewer proposes two monitoring checks per batch rather than one. Monitoring would become 8 units per ten-request batch; all other assumptions stay the same. Estimate 100 requests again and explain the operational decision.

Runbook adaptation
The reviewer is available only at 16:00 UTC today. Eight drafts are already pending. The existing training backlog limit is ten pending items. Decide whether to start another ten-request batch, a smaller bounded batch or no new work yet. Explain the pause point, what happens to current drafts and what evidence is needed to restart. There is no single required schedule if your capacity and safety reasoning is explicit and consistent.
