Verification plan
Reasoning depth workbench
Choose depth by evidence, not instinct.
Match the mode to the task, expose verification debt, and leave with a run brief your team can inspect.
Recommended run
Balanced mode
78fit / 100
Dominant risk
Effort
Latency
Review
Why this mode
Next action
Visible tradeoffs
Compare all three modes
Stop condition
Inspect the model
No hidden reasoning required.
Difficulty combines stakes, ambiguity, decomposition need, and novelty. Deadline, budget, privacy, and verification coverage constrain the recommendation. The lab shows inputs and score contributions, never hidden chain-of-thought.
View scoring assumptions
Deep mode gains from difficulty and available verification. Fast mode gains from deadline pressure, low stakes, and clear tasks. Balanced mode handles mixed signals. High stakes with weak verification always adds human review.
Durable output
Keep the decision, not just the answer.
Scenario is ready.