runner-view 0.1.0

Local-first, event-driven evaluation dashboard for DeepEval

Runner View

Local-first, event-driven evaluation dashboard for DeepEval. Inspect trace runs and click to any node's detail in under 100ms — no account, no upload, nothing leaves your machine.

Launch (one command)

uvx runner-view

Opens the most recent stored run in the dashboard. With no stored runs yet, it prints an actionable message and exits cleanly.

Open a specific portable trace file:

uvx runner-view path/to/trace.rvtrace

Installed-tool form is equivalent:

uv tool install runner-view
runner-view [target]        # no arg = latest stored run; or a .rvtrace path

Runs are stored append-only at ~/.runner-view/runs.jsonl (one JSON line per run).

Develop

PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 PYTHONPATH=. pytest -q

PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 keeps the suite isolated from the project's own pytest11 entry point and sidesteps the pytest-rerunfailures socket-bind issue in sandboxed shells.

Status

Epic 2 (one-command offline launch), Epic 3 (pytest debug launcher — zero-click failure triage), and Epic 7 (Analyst Curation Layer — regression diff views and flag-to-dataset functionality) shipped.

Analyst Curation Layer Features:

• Regression Diff View: Automatically shows when 2+ runs share the same test_id
• Flag-to-Dataset Mapping: Flag spans and save them to a local dataset file
• Visual Highlighting: 🟢 Improved, 🔴 Regressed, ⚪ Unchanged metrics
• Per-Metric Comparison: Detailed comparison of scores across runs
• Group by Test ID: Runs with the same test_id are automatically grouped

Decisions and context live in _notes/.