---

name: experiments-analysis description: "Reuse OASIS legacy experiment patterns, user-generation notes, and post-simulation analysis workflows without assuming unsafe execution." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Experiments analysis

Use this sub-skill when a request is about legacy OASIS experiment families, generated-user pipelines, or post-simulation visualization and triage.

Start here

1. Identify the experiment family and artifact type.
2. Read legacy experiments.
3. Read profile generation and analysis if the request touches generated profiles, score plots, counterfactual scoring, or Neo4j follow graphs.
4. If the task starts from a SQLite database and needs schema or table inspection, hand off first to platform-actions DB summary before custom analysis.
5. If the request is really about profile validation, env lifecycle, or action arguments, route to agent-profiles, simulation-workflows, or platform-actions.

What this sub-skill owns

• Mapping legacy experiment folders to their config knobs and result artifacts.
• Explaining how to downscale or adapt a legacy config without changing the scenario unnecessarily.
• Reading generated DB, JSON, CSV, or graph outputs and deciding whether a plot, table, or graph export is the right next step.
• Flagging blocked runs caused by missing credentials, GPU or server pieces, or large-scale cost.

What it does not own

• Root install, import, or API smoke.
• Profile schema validation.
• Environment lifecycle or step-by-step execution.
• Detailed action argument matrices or recsys internals.

Safe defaults

• Treat OpenAI, Hugging Face, VLLM, Slurm, and Neo4j dependencies as conditional.
• Prefer analysis of existing outputs over rerunning simulations.
• For tiny budgets, lower num_timesteps, round_post_num, and activate_prob before changing the scenario.

Deliverables

• Adaptation notes for legacy YAMLs.
• Analysis plans for existing outputs.
• Blocker lists when a requested run is not safe to attempt.

References

• Legacy experiments
• Profile generation and analysis
• Troubleshooting