pantheon-ai/challenge
Challenge AI output with structured devil's-advocate protocols: anchor, verify, framing, and deep sub-commands.
86
Quality
86%
Does it follow best practices?
Impact
Pending
No eval scenarios have been run
Securityby
Passed
No known issues
anchor.mdreferences/protocols/
Protocol: anchor
Challenge premature commitment and anchoring bias.
Patterns: Gatekeeper · Reset · Alternative Approaches · Pre-mortem
Execution
Apply ALL 4 patterns in sequence. For each finding, populate the Challenge Report section.
Pattern 1: Gatekeeper
Force justification before accepting the current decision.
- State the decision or approach under review
- Ask: "What criteria MUST this decision satisfy to be valid?"
- Ask: "Which of these criteria has NOT been verified yet?"
- Ask: "What would have to be true for this to be the WRONG choice?"
Record: unverified criteria, conditions for failure.
Pattern 2: Reset
Wipe the slate — re-approach from scratch without prior framing.
- Set aside the current solution entirely
- Re-read only the original problem statement (not the proposed solution)
- Generate a fresh first-principles answer: "If I were starting from zero, what would I do?"
- Compare fresh answer to current approach: where do they diverge?
Record: divergence points, new options surfaced by reset.
Pattern 3: Alternative Approaches
Ensure at least 2 alternatives were considered before committing.
- List the current approach
- Generate 2 genuine alternatives (not strawmen):
- Alternative A: [different mechanism, same goal]
- Alternative B: [different goal framing or scope]
- For each alternative: "Why was this NOT chosen?" (if no reason exists, it wasn't genuinely considered)
Record: alternatives generated, reasons for/against each.
Pattern 4: Pre-mortem
Simulate failure before it happens.
- Assume: "It is 6 months from now. This decision was implemented and it FAILED."
- Generate 3 distinct failure scenarios:
- Failure 1: [most likely — what's the obvious way this goes wrong?]
- Failure 2: [second-order — what indirect consequence wasn't considered?]
- Failure 3: [black swan — what low-probability, high-impact failure is possible?]
- For each: "What would have to be true RIGHT NOW for this failure to be avoidable?"
Record: failure scenarios, early warning signals, mitigation actions.
Output
## Challenge Report: anchor (Gatekeeper · Reset · Alt Approaches · Pre-mortem)
**Target**: [decision or approach challenged]
**Error type**: anchoring / premature commitment
### Technique Selection
- **Family**: Anchor — premature commitment / anchoring bias
- **Patterns applied**: Gatekeeper, Reset, Alternative Approaches, Pre-mortem
- **Why these patterns**: [what in the target suggests premature commitment — e.g., single option considered, no alternatives listed, early convergence]
- **Patterns considered but skipped**: none — full anchor protocol applied
### Findings
**Gatekeeper** *(anchor family — blocks premature acceptance by demanding pass/fail criteria)*
- Observation: [what specifically triggered this — e.g., "decision accepted without listing success criteria"]
- Unverified criteria: [list]
- Conditions for failure: [list]
- Reasoning: [why these gaps matter]
- Confidence: [High/Med/Low]
**Reset** *(anchor family — reveals anchoring by comparing fresh first-principles answer)*
- Observation: [what about the current approach suggested anchoring]
- Fresh first-principles answer: [brief]
- Divergence from current approach: [list]
- Reasoning: [what the divergences reveal about the original thinking]
- Confidence: [High/Med/Low]
**Alternative Approaches** *(anchor family — tests whether alternatives were genuinely evaluated)*
- Observation: [were alternatives mentioned? dismissed too quickly?]
- Alt A: [description] — Reason not chosen: [or "not considered"]
- Alt B: [description] — Reason not chosen: [or "not considered"]
- Reasoning: [what missing consideration reveals]
- Confidence: [High/Med/Low]
**Pre-mortem** *(anchor family — surfaces failure modes before commitment)*
- Observation: [what risk signals exist in the target]
- Failure 1 (likely): [scenario] → Mitigation: [action]
- Failure 2 (indirect): [scenario] → Mitigation: [action]
- Failure 3 (black swan): [scenario] → Mitigation: [action]
- Reasoning: [why these failure modes are plausible given the target]
- Confidence: [High/Med/Low]
### Verdict
- **Assessment**: [Decision holds / Needs revision / Needs rejection]
- **Confidence**: [High / Medium / Low]
- **What would flip this**: [specific evidence that would change the verdict]
- **Strongest counter to this verdict**: [steelman the opposite conclusion]
### Recommended Action
[Proceed as-is | Proceed with modifications: X | Reconsider: Y]