In the "both shards overloaded > retry cap" case, the test asserted numRetriesDueToOverloadAttempted == min(numOverloadErrorsReceived, kRetryAttempts) per shard. This fails when one shard hits its cap first and calls stopRetrying(), cutting off the other shard after it received an error but before it attempted the corresponding retry (observed 2 retries vs 3 errors). The terminal error is also always counted without a retry. Fix: assert retries is between errors - 1 and errors, capped at kRetryAttempts. Latent since the test was introduced in SERVER-127629.