Mind · Reasoning
SteelmanCheck
Qualify the opposition, then release only if the conclusion remains robust
Definition
Mechanism
Before finalizing a Decision or output, the Agent MUST generate a claim-specific counter-argument to its conclusion. It uses Critique to challenge the underlying Belief, Judge to score the counter-argument, and Check to test the Decision's Robustness against any counter-argument that meets `strength_threshold`. A sub-threshold counter-argument is strengthened up to `iteration_limit`; if no adequate counter-argument is produced, release is blocked. Once an adequate counter-argument exists, a Falsified robustness result requires revision or discard, Unknown blocks release pending review, and only Verified permits finalization. It prevents confirmation CognitiveBias. For adversarial contexts, see adversarial steelmanning.
Preconditions
- A Decision or output exists to challenge.
- The conclusion exposes a specific claim or rationale that can be contradicted.
Postconditions
Invariants
- Adequate opposition: before release, Judge scores the counter-argument at or above `strength_threshold`.
- The counter-argument identifies the specific claim it contradicts.
- Release discipline: only a Verified Check of the Decision's Robustness after adequate opposition permits release; Falsified requires revision or discard, and Unknown requires review.
Risks
Failure modes
Collusion
Proposer and Critic are same entity; susceptible to Strawman Waltz attack where agent generates weak counter-arguments to easily defeat them.
Relationships
Builds on
Composes with
References
Autonomous Goal-Directed Entity
Infrastructure
A subjective claim with confidence and evidence
Infrastructure
Structural processing error definition
Infrastructure
Irrevocable selection from a set of options
Infrastructure
Repeated execution cycle
Society
Resistance to failure under stress
Used by
Composes with
Dual-advocate verdict generation
Agent that architects mechanistic design proposals
Internal Thesis-Antithesis-Synthesis loop
References
Society
Mitigating groupthink via mandatory devils advocacy
Simulating diverse experts
Ordering rule: steelman opposing view before proposing, so SteelmanCheck has real targets
Reference
Parameters
iteration_limitInteger, [1, 5]- Maximum strengthening attempts before failure to produce adequate opposition blocks release
strength_thresholdFloat, [0.7, 0.95]- Minimum {{judge}} score for a counter-argument to qualify as a steelman; it is not the verdict on the {{decision}}
Signature
Check(Robustness)Critique(Belief)Judge(Critique)