Component · spec & evidence

Navigation Fitness Benchmark

Measures a navigation result against its target for recall, precision, and speed on public examples.

Contract

Does A scorer for how well a navigation system found the right thing. Each test case is a "task" (what someone was looking for, which exact items count as the right answer, which shortcut routes are off-limits, and how fast it should be) paired with a "route packet" (what the system actually returned and how long it took). The tool re-checks the answer: did it find the expected items (recall), how much of what it returned was on-target (precision), did it take a banned first step, did it mention the right cue words, and did it stay under the time budget. It then flags any "debt" — places that missed the mark or ran slow. It judges curated public examples only; it does not run the real private system, does not test the underlying search/embeddings, and does not claim to be a general-purpose navigation score.

Scope limit The bundle demonstrates a route-packet evaluator for expected stable ids, forbidden first routes, and latency budgets. It is not a live private kernel run, not an embedding benchmark, not a universal navigation benchmark, and not launch-scope decision.

Run

microcosm navigation-fitness-benchmark run --input fixtures/first_wave/navigation_fitness_benchmark/input --out /tmp/plectis-public-runs/navigation_fitness_benchmark

microcosm navigation-fitness-benchmark run --input fixtures/first_wave/navigation_fitness_benchmark/input --out receipts/first_wave/navigation_fitness_benchmark --acceptance-out receipts/acceptance/first_wave/navigation_fitness_benchmark_fixture_acceptance.json

Evidence

ClassBounded runtime computationevidence 4/5Real runtime resultRuns real tools

Links to Certificate Kernel Execution Lab, Verifier Lab Kernel, Formal Math Lean Proof Witness

navigationevaluationarchitecture

Source

Source Design note · Source

Map Show on the map

Index Open compact index card

Paper module