This disclosure provides a method of accurately determining expected transaction times associated with flash memory subdivisions, such as devices, blocks or pages. By performing a test transaction to program each bit of each such unit, the maximum expected programming time of each unit may be determined in advance and used for scheduling purposes. For example, in a straightforward implementation, a relatively accurate, empirically measured time limit may be identified and used to efficiently manage and schedule flash memory transactions without awaiting ultimate resolution of attempts to write to a non-responsive page. This disclosure also provides other uses of empirically-measured maximum flash memory transaction times, including via multiple memory modes and prioritized memory; for example, if a high performance mode is desired, low variation in flash memory transaction times may be tolerated, and units not satisfying these principles may be marked relatively quickly. A mechanism is also provided for recalibrating memory previously marked. By minimizing variability, flash memory can be applied to a broader range of designs and potentially to a broader set of main memory applications.