mr-fang-vlsi/control-theory-veto

Control Theory Veto

Overview

This skill protects the loop from executing invalid ideas just because they are executable. It evaluates proposal validity against theory, known constraints, and policy, then emits a veto decision.

When to run

Run before:

1. launching large/long job batches,
2. changing objective/cost model assumptions,
3. making algorithmic comparison claims,
4. introducing new route/CTS/PDN policy that may break prior assumptions.

Inputs

Required:

1. current user proposal (one-paragraph objective),
2. relevant artifacts (latest summary/gate/monitor),
3. KB references used by the proposal.

Recommended:

1. primary-source/web references when assumptions may drift.
2. recent experience-layer artifacts from eda-experiment-phenomenology-analyst when similar proposals have already been attempted.

Background Knowledge Links

This skill is only valid when its veto logic is linked to explicit background knowledge:

1. relevant KB policy and physics notes,
2. local paper-derived evidence when the proposal touches theory or model assumptions,
3. scoped retrieval from eda-context-accessor when the needed KB or paper context is missing from the current brief.
4. repeated empirical experience patterns when the proposal resembles prior failed or weak-information branches.

If a proposal conflicts with linked background knowledge, the veto must say so explicitly rather than treating the conflict as a generic risk.

Decision output contract

Always output:

1. verdict: GO / CONDITIONAL / NO-GO
2. blocking_reasons: numbered list
3. assumption_table: assumption, evidence, status
4. experience_risks: repeated empirical patterns that materially changed the verdict, or none
5. safe_next_step: one minimal testable action
6. override_policy:
   • NO-GO can be bypassed only with explicit user override.
   • when override happens, mark run as veto_overridden.

Operational References

1. Load references/background-knowledge-links.md when deciding which KB docs, paper summaries, or scoped retrieval outputs are authoritative for the current proposal.
2. Load references/veto-criteria.md when classifying contradiction severity or deciding whether the result is GO, CONDITIONAL, or NO-GO.
3. Load references/decision-template.md when writing the final veto artifact.