v0.5.0

Release benchmark results

Release gate: pass

Functional checks, microbenchmarks and E2E measurements have separate scope. Command wall time includes process startup. No causal productivity or NPU performance claim follows from functional checks or CPU microbenchmarks.

Repository: mindie-agent/profiler; version: 0.5.0
Revision: 4010d72f0b6950caf1425e3b683d495a6c76016a; dirty: False
Runner revision: 0a3c9fd81fb37a1b4a52807d35509daca777ebf2
Suite: core; environment: linux/x86_64

Case	Kind	Status	Repetitions	Metric	Median	Unit
profiler-analysis-and-attribution	microbenchmark	pass	5/5	c7_local_ideal_roofline	9.3268	us/op
profiler-analysis-and-attribution	microbenchmark	pass	5/5	c7_memory_non_additive_attribution	135.502	us/op
profiler-analysis-and-attribution	microbenchmark	pass	5/5	c7_two_rank_comparison	276.055	us/op
profiler-analysis-and-attribution	microbenchmark	pass	5/5	c7_two_rank_two_step_analysis	7667.33	us/op

What's Changed

• feat(profiler): add mindie-profiler CLI entry, mindie-sim dependency, and isolated import contract by @maoxx241 in #1
• fix(compare): require workload identity in compare CLI and preserve receipts on invalid accounting by @maoxx241 in #2
• feat(profiler): add Ascend Advisor rules, slow-rank vs slow-link attribution, and memory forensics by @maoxx241 in #3
• ci(profiler): decouple from diagnostics and add CI/release workflows by @maoxx241 in #4

New Contributors

• @maoxx241 made their first contribution in #1

Full Changelog: https://github.com/mindie-agent/profiler/commits/v0.5.0