Adversarial Triad + Counterarguments

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

A counter-race report shows two workers reading zero then writing one, while a rebutter claims mutex protection. Inspect actual lock scope on one revision rather than a three-role vote.

Mechanism

Separate finding, rebuttal and adjudication with shared facts. Finder supplies feasible trigger/path, rebutter examines locks/counterevidence and adjudicator verifies both for supported/refuted/unconfirmed status. Preserve provenance/conditions without requiring a defect or treating roles as authority.

Bad example

Vote and call two-to-one proof, deleting lock evidence to make positions agree.

Good example

For the teaching counter, cite interleaving/locations and same-revision lock/caller evidence. Verify feasibility: complete protection refutes that schedule, missing protection supports it and missing material remains unconfirmed. Present positions/reasons without votes, court labels or hidden disagreement in synthesis.

Why the change matters

Roles expose assumptions about feasible ordering and coverage. Evidence-based resolution evaluates conditions rather than treating more model opinions as truth.

Observable expectation

Unlocked fixture supports the lost-update schedule; full-lock fixture refutes it; unread callers remain unconfirmed. Inspect actual conditions and both positions, not 2:1 alone.

Retain unresolved facts after adjudication. No real concurrency evaluation is run.

Limits

Roles provide neither expertise nor independence guarantees. Real locks/scheduling require evidence. Source council/provider arrangements are instances, not a requirement for three models or paid calls.