Description
This protocol serves as a "Cognitive Brake." It is invoked when high precision is required or when the initial answer seems "too obvious" (high probability/low compute). It forces the agent to suspend the final answer, scan the entire search space for low-probability candidates, and perform an inversion test before converging on a selection.
Instructions
Step 1: Divergent Scan (The Silent Survey)
Before formulating the final response, generate an internal list of 3-5 distinct candidates that fit the user's criteria.
- Constraint: You are FORBIDDEN from selecting the first candidate that comes to mind.
- Search Target: Look for "Background Objects," "Structural Elements," or "Counter-Intuitive Solutions."
Step 2: Bias Identification
Review the generated list and identify the "Statistical Default."
- Question: "Which of these candidates would an average human or standard model pick 90% of the time?"
- Action: Flag this candidate as
[BIAS_DEFAULT].
Step 3: The Inversion Test
Challenge the [BIAS_DEFAULT].
- Question: "Why might this obvious answer be a decoy or incorrect?"
- Action: Check for exclusion criteria (e.g., user said 'Nope', context implies a trick, visual obstruction).
Step 4: Convergence & Selection
Select the final answer based on Contextual Fit rather than Saliency.
- If the
[BIAS_DEFAULT]survives the Inversion Test, output it. - If it fails, promote the highest-ranked alternative (e.g., the 'Dolly' instead of the 'Hat').
Examples
- "Engage enumeration protocol for this visual puzzle."
- "Execute enumeration scan to debug this code block (avoiding standard library assumptions)."
- "Run enumeration-protocol-execution on the error logs."