Enciclopedia Boulder v1.0

Inicio · Capítulos · 14. Señales para visión artificial

15.234 14.8.24 Pipeline completo

Etapa ID Acción
1 BEN-01 Freeze intended use, operating domain, claims and prohibited interpretations.
2 BEN-02 Select immutable dataset, ontology, guideline, split and reference-policy versions.
3 BEN-03 Register benchmark profile, primary tasks, metrics, tolerances and decision gates.
4 BEN-04 Audit consent, integrity, synchronization, calibration and scorable reference tiers.
5 BEN-05 Verify grouped splits, duplicate hashes, scene leakage and test quarantine.
6 BEN-06 Build deterministic evaluation container and record all artifact hashes.
7 BEN-07 Run constant, rule, modality and oracle baselines.
8 BEN-08 Evaluate primitive pose, scene and tracking tasks.
9 BEN-09 Evaluate contact, support and force tasks by reference tier.
10 BEN-10 Evaluate phase boundaries and multilanes.
11 BEN-11 Evaluate kinematic trajectories, derivatives and summaries.
12 BEN-12 Evaluate movement/failure classification and 28 contrast sets.
13 BEN-13 Evaluate calibration, prediction sets, abstention and open-set behavior.
14 BEN-14 Run full attempt reconstruction with real upstream predictions.
15 BEN-15 Run distribution-shift views and worst-group analysis.
16 BEN-16 Run controlled corruptions, sensor dropout and calibration drift.
17 BEN-17 Execute prespecified ablations on paired units.
18 BEN-18 Measure latency, throughput, memory, energy and operational failures at quality floor.
19 BEN-19 Compute cluster-aware confidence intervals and paired comparisons.
20 BEN-20 Sample evidence ledgers and attribute end-to-end errors upstream.
21 BEN-21 Re-run audit subset from immutable artifacts to test reproducibility.
22 BEN-22 For deployment claims, execute prospective frozen field pilot.
23 BEN-23 For coach claims, execute separate blinded/controlled human-factors study.
24 BEN-24 Publish result bundle, limitations, unsupported classes and gate decisions.