File formats and codecs

This section is for reader and writer implementers. To save a Python acquisition, use Save and share your data; no codec selection is needed.

QEM: portable measurements and metadata

Read the specification in this order:

Question	Authoritative page
What does the file contain, and what must round-trip?	Container and scientific metadata
How are integer counts and float bits encoded?	Codec layouts
Which vendor fields are interpreted or retained?	Metadata mapping
How do independent readers prove correctness?	Conformance and interoperability
How does a qualified DM4 acquisition map to QEM?	DM4 conversion details

Container, metadata-schema, and codec versions identify different parts of the file contract. Follow the identifiers declared in the file; a newer prose specification does not imply that every backend supports every layout.

Runtime storage and prepared formats

A device’s resident representation and a portable acquisition file serve different purposes. Python acquisition loading uses ANS; internal layouts and prepared formats have their own callers and acceptance boundaries.

Implementation task	Contract
Understand encoded, paired, and dense layouts	Count representations
Implement the Swift/Metal prepared packed-file layout	Lossless Pack Format v1
Produce that format from a native application	Native producer
Maintain the internal count-ANS container	Count-ANS
Maintain paired resident counts	Paired-count layout

The prepared packed format is read only by the native Swift/Metal loader; io.load does not open it. Use the original measurements to export a supported QEM file. Dated codec experiments remain in Historical records.