mr-fang-vlsi/eda-experiment-phenomenology-analyst

EDA Experiment Phenomenology Analyst

When to use

Use this skill when:

1. an experiment or batch has finished and raw logs should be lifted into reusable evidence,
2. repeated experiment patterns need to become durable recommendations instead of one-off notes,
3. control-postrun-retro or control-theory-veto should consume prior empirical experience without reparsing all prior logs,
4. a new “experiment expert” style workflow needs access to an experience layer rather than raw batch files only.

Scope Boundary

This skill owns horizontal evidence lifting and experiment-experience maintenance.

It owns:

• extracting structured facts from raw experiment artifacts,
• separating result, conclusion, and experience,
• maintaining experience deltas or registry-style artifacts,
• surfacing empirically repeated patterns that downstream skills may rely on.

It does not own:

• final post-run recursion decisions (control-postrun-retro),
• final veto authority (control-theory-veto),
• workflow routing ownership,
• domain-method correctness claims by itself.

Expected Downstream Consumers

Typical consumers:

• workflow-scoped-execution for batch closeout,
• control-postrun-retro for next-step decision,
• control-theory-veto for experience-informed GO/CONDITIONAL/NO-GO,
• workflow-research-chain for durable experiment memory across stages.

Inputs

Provide or derive:

1. raw batch artifacts:

• summaries,
• manifests,
• monitor/history files,
• report files,
• stdout/stderr logs when needed,

2. experiment objective and comparison contract,
3. any existing memory or experience artifacts for the same design/branch,
4. KB references if the batch conclusion depends on policy interpretation.

Outputs

Emit the smallest set needed by downstream consumers:

1. result artifact

• structured metrics and directly observed facts
• recommended form: *.results.tsv or *.results.json

2. conclusion artifact

• batch-local mechanism judgment with cited evidence paths
• recommended form: *.conclusion.md

3. experience_delta artifact

• reusable experience items added or reinforced by this batch
• recommended form: *.experience_delta.md

4. optional registry update

• append/refresh durable experience inventory when the workflow calls for long-lived storage

Hard rules

1. Do not skip from raw logs directly to durable advice; write result and conclusion first.
2. Mark speculation as speculation; only experience with repeated or well-supported evidence should influence later veto decisions.
3. Keep batch-local conclusions separate from cross-batch experience.
4. When experience is used for veto, include the trigger pattern and linked evidence runs.

Operational References

Load only what is needed:

1. Load references/layer-contract.md when deciding how to split artifacts across log, result, conclusion, and experience.
2. Load references/consumer-integration-policy.md when wiring outputs into workflow-scoped-execution, control-postrun-retro, control-theory-veto, or workflow-research-chain.
3. Use docs/knowledge_base/templates/experiment_results_template.tsv, docs/knowledge_base/templates/experiment_conclusion_template.md, and docs/knowledge_base/templates/experiment_experience_delta_template.md as the default artifact skeletons when no more specific batch-local template already exists.