FlagTensor Acceptance Checklist

This checklist tracks the current compliance status against the operator library acceptance standards.

Structure & Organization

Item	Status	Notes
Unified operator registry exists	Done	conf/operators.yaml
Registry is CI/weekly/entry point	Done	tools/run_flagtensor_ci.py and run_flagtensor_weekly.py use it
tests/ directory exists as correctness entry	Done	Category-based organization: tests/{unary,binary,contraction,sparse}/. All categories migrated (28 unary, 4 binary, 5 contraction, 1 sparse). Legacy per-op ctests/ files retained as compatibility shims.
Benchmark dtype coverage	Done	float16, float32 per DEFAULT_BENCHMARK_DTYPES
Correctness dtype coverage	Done	float16, float32, bfloat16 per DEFAULT_CORRECTNESS_DTYPES
Benchmark shape coverage	Done	unary/binary: 22 shapes (14 1D pow2 + 8 multi-dimensional); contraction: 4 shape pairs; sparse: 3 shape pairs
benchmark/ supports category-level execution	Done	All four categories: test_unary_perf.py, test_binary_perf.py, test_contraction_perf.py, test_sparse_perf.py
Pre-commit configuration exists	Done	.pre-commit-config.yaml
pyproject.toml has tool configs	Done	black, isort, flake8, pytest markers

Testing Framework

Item	Status	Notes
Pytest-based correctness tests	Done	ctests/ and tests/
Shared tolerance/assertion helpers	Done	centralized in src/flagtensor/testing/ package with assertions.py, shapes.py, dtypes.py modules
tests/accuracy_utils.py compatibility layer	Done	Re-exports from flagtensor.testing
Dtype-aware tolerance policy	Done	float16, float32, bfloat16
Reference selection documented	Done	docs/testing_strategy.md
Shape coverage policy documented	Done	docs/testing_strategy.md

Performance Testing

Item	Status	Notes
Benchmark against cuTensor baselines	Done	Existing benchmark suite
Kernel/operator/wrapper modes defined	Done	docs/benchmark_strategy.md
Smoke vs full run distinction	Done	run_flagtensor_ci.py --smoke
Warmup/repetition/timing standards	Done	Consolidated in benchmark_core.py and config.py
Mode-aware CSV selection	Done	benchmark_csv_path() in CI runner
HTML/XLSX report generation	Done	HTML report tooling exists; XLSX output via write_benchmark_xlsx() in visualization.py

CI/CD & Automation

Item	Status	Notes
Correctness CI job	Done	ci.yaml correctness-smoke
Performance CI job	Done	ci.yaml perf-smoke
Weekly regression workflow	Done	weekly.yaml (registry-driven)
Quality gate (pre-commit)	Done	quality-gate.yaml
Registry consistency check	Done	quality-gate.yaml registry-consistency job
Build/package check	Done	quality-gate.yaml build-check job
Artifact upload and summary	Done	Artifacts + GITHUB_STEP_SUMMARY in CI
Acceptance-level CI workflow	Done	acceptance.yaml with category/mode filtering
CI matrix documentation	Done	docs/acceptance/ci_matrix.md
Release/publish workflow	Done	release.yaml with build, release notes, PyPI publish
Multi-backend compatibility CI	Done	compatibility.yaml with CUDA 12.1/12.4/12.6 matrix + H20 regression

Documentation & Release

Item	Status	Notes
README with usage examples	Done	README.md
Testing strategy document	Done	docs/testing_strategy.md
Benchmark strategy document	Done	docs/benchmark_strategy.md
Acceptance checklist	Done	This file
Operator coverage matrix	Done	docs/acceptance/operator_coverage.md
FlagTensor-specific accuracy/benchmark policies exist	Done	docs/acceptance/accuracy_policy.md, docs/acceptance/benchmark_policy.md
Known issues list	Done	docs/acceptance/known_issues.md
Standard acceptance commands	Done	docs/acceptance/standard_commands.md
Release note template	Done	docs/acceptance/release_notes_template.md

Known Issues

All previously documented issues have been resolved:

• exp and log: float64 fallback removed — float64 is no longer a supported dtype.

• tensor_contraction_trinary: float64 path removed; operator supports float16/float32 only.

• block_sparse_tensor_contraction float16: Fixed via dense fallback routing; float16 tests now active.