By name: topology-benchmark-review description: Use when a topological result (persistence diagram, W2/landscape distance, regime detection) is asserted as correct — to check it against known published benchmarks and the project's validation table before trusting it.
Topology Benchmark Review
Use this for the Topology lane. A persistent-homology pipeline can run end-to-end and produce a plausible-looking diagram that is quantitatively wrong. Before a topological result is used in a paper or as a downstream input, check it against an external benchmark, not just internal consistency.
Core Check
- Known-result anchor. Compare against a published benchmark where one exists
(e.g. Gidea-Katz 2017 for financial pre-crash L1/L2 norm trends) or the
project's
validate-topologybenchmark table. - Sanity of the diagram. Off-diagonal features are real, not numerical noise; H0/H1/H2 counts are plausible for the construction; total persistence scales sensibly with the filtration.
- Metric direction. W2/landscape distances move in the expected direction for a known perturbation (a controlled change increases distance).
- Replication. The result reproduces across eras/cohorts where the theory predicts it should.
Output Format
MATCHES BENCHMARK / DEVIATES / NO BENCHMARK AVAILABLE, with the benchmark cited and the discrepancy quantified.
Pressure Scenario
A topological result asserted as correct on internal consistency alone, with no comparison to a published anchor — a silent quantitative error would pass.
Related Skills & Contracts
- Pairs with
validate-topology(the sanity checklist) andwasserstein-audit(W1/W2 and order/internal_p ambiguity). - Enforcing contracts:
topology-invariants/*.