Recorded history / identity / reproducible builds
Same bytes.
Different histories.
Equality is not one switch. Two builds can match byte-for-byte and still differ by toolchain, ambient record, or execution provenance. Choose the rule before reading the answer.
01 / Choose two histories
Which build records are we comparing?
All three pairs use actual captured executions. The controls change one declared dimension at a time; none of the histories is replaced by a hypothetical shortcut.
loading
02 / Choose equivalence regime F
The rule determines the answer
Each regime projects only its listed fields. Equality in one column never overwrites a different verdict in another.
loading
Differing fields
Explicitly excluded
loadingloading03 / Complete result
One experiment, fifteen honest verdicts
The matrix is the result. There is no regime-independent column hiding behind it.
04 / Rebuild contract
What was actually pinned
loading
Two fixture source files, the exact builder implementation, LF output, UTF-8 encoding, and a source-pinned timestamp are verified offline. Capture logs are evidence beside the output, not bytes inside it.
05 / Historical Load
No number is the correct result
06 / How to use this
Pick identity at the question's resolution
Distribution check
Use byte-output equivalence to ask whether a rebuild produced the exact specified artifact.
Review and audit
Use inputs, toolchain, environment, or provenance when the route to those bytes matters.
Never collapse the records
An equal verdict creates a regime-local equivalence class. It does not erase either execution event.
Evidence boundary
Direct
Source bytes, four execution records, reported runtimes, and produced artifact bytes.
Derived
Projection identities, differing-field lists, and the fifteen pair/regime verdicts.
Excluded
The session label is recorded but does not participate in normalized environment identity.
Unknown
Cross-machine, non-Darwin, other-runtime reproducibility, and builder trustworthiness.