Purpose

Files and directories to inspect first

• README.md
• STATUS.md
• VALIDATION.md
• validation/seeded_vortex_phase2_3_experiment_pack.yaml
• configs/seeded_vortex_experiment_pack.schema.json
• configs/tdgl_run_provenance.schema.json
• configs/seeded_vortex_tier2.schema.json
• src/tdgl_rf/cli.py
• src/tdgl_rf/workflows/experiment_sweep.py
• src/tdgl_rf/analysis/experiment_aggregation.py
• tests/integration/test_experiment_sweep_workflows.py

Step-by-step procedure

1. Confirm the manifest is a 3.x deterministic experiment pack, not a Phase-2.2 suite and not a 4.x ensemble.
2. Read the seeded-vortex status docs so the run is interpreted as a deterministic sweep/repetition harness over committed Tier-2 observables only.
3. Run the manifest through tdgl-rf run-experiment <manifest>.
4. Inspect experiment_results.json first, then the CSV and Markdown views.
5. Spot-check one per-run provenance.json and diagnostics/seeded_vortex_tier2.json under the output tree.
6. Preserve the non-claim boundary: this harness is not a broader seeded-vortex validation artifact and does not authorize stochastic or long-time claims.

Exact commands

tdgl-rf run-experiment validation/seeded_vortex_phase2_3_experiment_pack.yaml
tdgl-rf run-experiment validation/seeded_vortex_phase2_3_experiment_pack.yaml --output-dir runs/experiment_seeded_vortex_phase2_3_experiment_pack_manual

Validation / definition of done

• The command exits successfully.
• The output directory contains experiment_results.json, experiment_results.csv, and experiment_results.md.
• Each recorded run directory contains provenance.json and diagnostics/seeded_vortex_tier2.json.
• Aggregate results reflect the expected parameter-point count and run count from the manifest.
• Summaries stay inside the deterministic same-stack sweep/repetition boundary.

Failure modes / escalation

• Manifest dispatch or schema validation failure: Fix the manifest or use qdp-contract-sync if the schema must change. Do not rely on fallback behavior.
• Non-empty output directory: Use a fresh destination; the workflow is fail-closed on output promotion.
• Missing per-run provenance or Tier-2 payload: Treat the pack as failed even if some runs completed.
• Attempted interpretation as broad validation: Stop and restate the experiment-harness boundary from VALIDATION.md and STATUS.md.

Output artifacts

• runs/experiment_<pack_name>/experiment_results.json
• runs/experiment_<pack_name>/experiment_results.csv
• runs/experiment_<pack_name>/experiment_results.md
• runs/experiment_<pack_name>/param_<hash>/<run_id>/provenance.json
• runs/experiment_<pack_name>/param_<hash>/<run_id>/diagnostics/seeded_vortex_tier2.json

Example prompts that SHOULD trigger the skill

• "Run the committed deterministic seeded-vortex experiment pack and inspect the aggregates."
• "Check whether the experiment harness still writes stable provenance for each repetition."
• "Re-run the Phase-2.3 manifest after my aggregation change."

Example prompts that should NOT trigger the skill

• "Run the Phase-2.2 seeded validation suite."
• "Add a new experiment-pack schema field."
• "Execute a stochastic 4.x ensemble manifest for publication claims."