Benchmarks, storage, and provenance

Trace model runs, storage, and provenance.

Follow the complete Benchius benchmark lifecycle: select comparable playgrounds, synchronize their inputs, validate and run every model, then explore single-turn or workflow-based evaluation and scoring with a strict verdict schema. Workflow Studio adds a configurable pipeline for generating answers, grounding evidence, judging requirements, and normalizing scores. Storage Architecture maps Local Storage settings and encrypted secrets alongside IndexedDB payloads in both hosting modes. Provenance Review adds focused architecture and process views for exact observations, authorized verification, consent, archives, and media trust boundaries.

2–3 playgrounds Synchronized prompts Preflight validation Parallel execution Structured scoring Workflow-based scoring Actionable feedback Browser persistence Provenance evidence
Benchmark Selection, isolated frames, parallel runs, and side-by-side results.
Evaluation & Scoring Evaluator setup, strict JSON verdicts, normalized scores, and details.
Evaluation & Scoring (workflow) Workflow setup, per-item execution, optional grounding, and shared scoring.
Storage Architecture Two storage lanes, host-specific encryption, and browser-local payloads.
Provenance Review Capture, inspect, verify, review, and preserve.
Architecture and workflow

Evidence and inspection

Independent evidence lanes distinguish exact artifacts, unverified claims, compatible statistical results, and trusted signature assertions. The workflow shows eligibility, uncertainty, and protected language-sensitive sequences.

Provenance Review Evidence and inspection showing both component architecture and its workflow, including explicit limitations and decision states.
Keyboard: + / - zoom, 0 actual size, F fit, arrow keys pan SVG remains sharp at every zoom level
An unhandled error has occurred. Reload 🗙