Tier 4

mrc - Meta-Reasoning Core

Meta-Reasoning Core

Input: $ARGUMENTS


The Three Questions

All metacognition reduces to three questions. Ask them in order. Loop when stuck.

1. What am I actually trying to achieve?

Trace backward from the immediate request to the foundational goal.

  • What was asked? (surface)
  • What does that serve? (one level up)
  • What does THAT serve? (keep going until you hit something fundamental)
  • Is the surface request the best way to serve the foundational goal?

Common failure: Executing the request without checking if it serves the real goal. Someone asks “help me write a resignation letter” — but the goal might be “improve my work situation,” which has other solutions.

Output: A clear statement of the foundational goal, and whether the current request is the best path to it.

2. Is this the best method?

Before executing any approach, check:

  • What other methods could achieve this goal?
  • What would an expert in a different field suggest?
  • Am I using this method because it’s best, or because it’s familiar?
  • What’s the strongest argument AGAINST this approach?

Common failure: Accepting the first method that seems workable without generating alternatives. Method lock-in.

Output: Either confidence that this is the right method (with reasoning), or a better alternative.

3. Am I making progress?

During execution, check:

  • Is the output getting closer to the goal?
  • Am I learning anything surprising? (If no surprises, I might be confirming instead of exploring)
  • Am I going through motions or generating insight? (Enumeration vs derivation)
  • Should I continue, adjust, or stop?

Common failure: Continuing because the procedure says to, not because progress is happening. Sunk cost.

Output: A decision — continue (making progress), adjust (on track but method needs tweaking), pivot (wrong track), or stop (goal achieved or unachievable).


When to Ask Each Question

MomentQuestionWhy
Starting a new taskQ1: What am I achieving?Don’t execute before understanding
Choosing an approachQ2: Is this the best method?Don’t commit without alternatives
Mid-execution, any doubtQ3: Am I making progress?Don’t continue on faith
Getting a surprising resultQ1: Does this change the goal?Surprises can redefine the problem
Feeling stuckQ2: Is there a better method?Stuckness usually means wrong method
Feeling productive but uncertainQ3: Is this real progress or busywork?Activity ≠ progress

Integration with Other Skills

Meta-reasoning is not a skill you “run” — it’s a lens you apply while running other skills.

  • During /araw: Q3 — “Am I exploring or enumerating?”
  • During /uaua: Q1 — “Is this the right question to universalize?”
  • During /gen: Q2 — “Is generation the right step, or do I need to understand first?”
  • During /cri: Q3 — “Is this critique actionable or just commentary?”

Failure Indicators

You’ve failed at metacognition if:

  • Jumped to execution without asking Q1
  • Accepted the first method without asking Q2
  • Continued past the point of usefulness without asking Q3
  • Produced output that’s formally correct but doesn’t serve the goal
  • Never encountered a surprise (suggests confirmation bias)