For fixed-bucket histograms, we assumed each observation occurred at the center of each bucket, and all underflow and overflow observations occurred at the lowest/highest bucket boundary. We could probably implement similar logic for exponential histograms.
Follow-up to https://github.com/GoogleCloudPlatform/opentelemetry-operations-go/pull/777
See https://github.com/GoogleCloudPlatform/opentelemetry-operations-go/blob/c7b17b6dd7a9f3b08b8539ed846cb4547927011a/exporter/metric/metric.go#L630
For fixed-bucket histograms, we assumed each observation occurred at the center of each bucket, and all underflow and overflow observations occurred at the lowest/highest bucket boundary. We could probably implement similar logic for exponential histograms.