Repeated runs identical
Same fixture, same complete read model.
Original Evidence · Reproducible Benchmark
Product evidence should show what the system actually does — and what it does not prove. This benchmark runs controlled Friction Radar v1 evidence through the product read-model contracts and publishes the result, fixture description and limitations.
Measured result
Version 2026-08-23. Controlled deterministic code-level benchmark of Friction Radar v1 evidence handling. This is not a customer-outcome, recall, precision, or latency benchmark.
Same fixture, same complete read model.
Critical evidence from another project did not leak into the scoped result.
Only signals sharing the explicit entity were correlated.
Global score and severity remain null in Friction Radar v1.
No category aggregation is presented before validation.
Process, resource, dependency, schedule, cost, risk, decision, and quality.
Control-by-control evidence
Reproducibility
The controlled fixture contains 3 in-project signals and 1 foreign-project signal. Two signals share an explicit entity, one signal is unrelated, and the read model is executed 2 times for determinism checking.
Evidence chain
Controlled events
↓
Friction signals with evidence refs
↓
Scoped Friction Radar read model
↓
Expected-control assertions
↓
Public JSON + stated limitations
What this benchmark does not claim
Evidence standard
Product evidence is strongest when a reader can distinguish what was observed, what was inferred, what is only predicted, and what remains unknown. The benchmark follows the same discipline used by Project Friction Intelligence.
Direct output or event-backed fact produced by the controlled fixture.
Interpretation supported by evidence, but not presented as automatic proof of cause.
Forward-looking exposure that must stay labeled as prediction rather than fact.
Missing evidence stays unknown instead of being silently converted into a score.
FAQ
It demonstrates the tested Friction Radar v1 contracts for deterministic read-model generation, project isolation, explicit-entity correlation, restraint around unvalidated aggregate scores, and honest empty-state behavior.
No. The benchmark is intentionally controlled and synthetic. It does not measure real-world precision, recall, false-positive rate, causal accuracy, model quality, or customer outcomes.
The measured result and its stated methodology are published as machine-readable JSON. The implementation repository is private, so this page does not claim that the source code or regression test is publicly accessible.
Friction Radar v1 deliberately avoids publishing a global friction score, global severity, or category aggregate score until an aggregation policy has been validated. Missing validation is represented honestly rather than replaced with a fabricated number.
ProjectOps360
Process Mining reconstructs what happened. Friction Radar surfaces evidence-backed friction. Living Graph shows what the problem can affect next.