One class of async meter-event failure: a code, how many events failed with it, and a sample of those failures.

sample_errors is a sample, not a list of every failure

error_count is the true number of events that failed with this code. sample_errors holds only a subset Stripe chose to include — and on high-volume failures it is routinely empty while error_count is in the hundreds. Treating length(sample_errors) as the failure count under-reports, usually by an order of magnitude. Read error_count for the count and sample_errors for the examples.

code is a String, and stays one

code is never converted to an atom. Three reasons, each sufficient:

1. Stripe documents the enum as open. The reference says "Open Enum." verbatim. New values arrive without a version bump.
2. It is server-controlled and demonstrably growing. Stripe has already retired one value (meter_event_value_not_found) that this repository's own metering guide still documents — the set churns in both directions.
3. Atoms are never garbage collected on the BEAM. Atomizing a value an external server chooses is an atom-table exhaustion vector, and the existing-atom variant merely converts that vector into a decode crash on the next code Stripe adds.

A closed union type would go stale the same way the Elixir peer library's card-error union did. Match on the binary.

Current values (non-exhaustive)

The ten values Stripe publishes today. This list is documentation, not a contract — do not build a total case over it without a catch-all clause:

• "archived_meter"
• "meter_event_customer_not_found"
• "meter_event_dimension_count_too_high"
• "meter_event_invalid_value"
• "meter_event_no_customer_defined"
• "meter_event_value_too_many_digits"
• "missing_dimension_payload_keys"
• "no_meter"
• "timestamp_in_future"
• "timestamp_too_far_in_past"

Note that "no_meter" (a code) is a different thing from v1.billing.meter.no_meter_found (an event type).