TDA Benchmark Evidence Map

Every performance or quality claim needs an evidence path. The benchmark suite records active behavior and keeps future acceleration work from becoming a README-only promise.

flowchart LR
  A["Claim"] --> B["Fixture"]
  B --> C["Reference output"]
  C --> D["Backend output"]
  D --> E["Equivalence check"]
  E --> F["Timing smoke"]
  F --> G["Artifact JSON/Markdown"]
  G --> H["Docs statement"]

Current Evidence Paths

Claim area	Evidence command	Boundary
Python/Rust PH behavior	python -m pytest python/tests -q	Correctness and contracts, not speedup
Native C++/ASM/CUDA/Triton source	backend-specific tests	Runtime-gated where tools/hardware are absent
E2E public claims	python benchmarks/e2e_claims.py	Smoke benchmark and claim registry
Docs and gallery	python -m mkdocs build --strict	Site builds, diagrams render through Mermaid

graph TD
  T["Tests"] --> E["E2E claims"]
  B["Benchmarks"] --> E
  D["Docs pages"] --> E
  E --> P["Published GitHub Pages"]
  E --> C["CI gate"]

Claim Boundary

The benchmark map proves that claims have a runnable verification path. It does not turn smoke timings into leaderboard results. Any speed claim must include a named baseline, hardware, data size, variance policy, and failure conditions.