Measure retry waste from attempts your backend actually observed. Enter the number of retry attempts and their average priced cost. If any failed or timed-out attempt has unknown usage, the honest result is incomplete—not zero.
Calculate observed retry waste
Observed retry waste
$1.20
100 priced retry attempts × $0.012000 per retry.
The formula
Observed retry waste = priced retry attempts × average observed cost per retry.
Count every application-level attempt after the first try for the same logical request. Preserve the same logicalRequestId and increase attempt for each retry. Report success, failure, timeout, and cancellation separately.
Why failed calls cannot default to free
A timeout can occur after a provider has already processed tokens. If the final provider usage never reaches your application, the cost may be real but unknown. Keep that attempt in the retry count, mark its usage incomplete, and reconcile the gap against the provider bill before claiming a complete savings number.
What this evidence can and cannot prove
This calculator multiplies observed retry attempts by an observed average cost. It does not infer hidden provider-client retries, estimate missing token usage, or prove that removing retries would preserve product quality. Use an application-level logical request ID and attempt number to make the input auditable.