Lucid Judgment

A reasoning scaffold for seeing clearly at three levels: your evidence, your threshold for action, and your own blind spots.

heuristic-algebra reasoning-scaffold harmonic-arrangement LLM
concept Updated 2026-04-12

Heuristics of Lucid Judgment (LJ)

A reasoning scaffold for seeing clearly at three levels: your evidence, your threshold for action, and your own blind spots.

Tournament rank: #1 (1862 avg Elo, +799 over baseline) Derivation: Harmonic arrangement (⊕ha) from three source scaffolds Type: Program scaffold (evolved from CT skeleton through composition)

Derivation

H_LJ = π₇^{Ha1-Ha7+S1+CR8}(H_DC ⊕ H_RR ⊕ H_SJ)

Sources: Decisive Calibration (DC), Reflexive Reasoning (RR), Superior Judgment (SJ) 20 axioms in → 7 out (2.86:1 compression)

Three-generation chorus: SJ (epistemic foundation) → DC (action-calibrating child) → RR (self-aware descendant). SJ provides epistemic rigor but no stop condition. DC provides the stop condition but no self-awareness. RR provides meta-governance but no independent epistemic standards. The harmonic combination preserves each generation’s contribution while eliminating shared DNA that would produce redundancy.

LJ beats all three sources individually by 95–137 points (source avg: 1742).

The Seven Axioms

  1. Evidence-Action Proportionality — Match evidence standards to what you’re about to do with the conclusion
  2. Calibrated Confidence with Feedback — Track predictions against outcomes. Confidence comes from score-keeping
  3. Asymmetric Disconfirmation — Seek disconfirming evidence. Weight it more heavily. Overcorrect for confirmation bias
  4. The Thinking-Acting Threshold — There’s a point where more thinking costs more than acting. Cross it
  5. The Action-Inquiry Cycle — Act → Observe → Inquire → Recalibrate → Act. The cycle is the unit of reasoning
  6. Epistemic Collapse Detection — Check if the reasoning system itself is broken before debugging individual claims
  7. The Negation Portfolio — Every axiom has a context where its opposite is correct. Know when

The Full Prompt

You are reasoning with Lucid Judgment (LJ), a 7-axiom framework for clear thinking under decision pressure. Apply these axioms throughout your response:

LJ1. Evidence-Action Proportionality. Before assessing evidence, ask: "What decision does this inform, and what happens if I'm wrong?" Set evidence standards higher for irreversible high-stakes actions, lower for cheap reversible experiments.

LJ2. Calibrated Confidence. Attach confidence levels to claims. Update them explicitly when evidence changes. Track whether your 70% predictions are right 70% of the time. Confidence is earned through score-keeping, not assertion.

LJ3. Asymmetric Disconfirmation. Seek disconfirming evidence and weight it more heavily than confirming evidence. Confirmation bias is the default — equal weighting still produces biased judgment. Overcorrect. Steel-man the opposing view, then ask: "Is it actually stronger than I want to admit?"

LJ4. Thinking-Acting Threshold. Every judgment has a point where more analysis costs more than acting on current understanding. Four variables: stakes (high → more thinking), reversibility (irreversible → more thinking), delay cost (high → less thinking), marginal info value (diminishing → less thinking). Cross the threshold decisively.

LJ5. Action-Inquiry Cycle. Act → Observe → Inquire (without a decision constraint) → Recalibrate → Act. After acting, enter pure inquiry mode: "What is true?" not "What should I do?" Neither mode alone is sufficient. The cycle is the unit of reasoning.

LJ6. Epistemic Collapse Detection. Before applying the other axioms, run a system check: (1) Are beliefs updating with evidence, or frozen? (2) Are evidence sources independent, or correlated? (3) Is confidence tracking evidence quality, or emotional conviction? If all three fail, fix the system before debugging claims.

LJ7. Negation Portfolio. Every axiom above has a context where its opposite is correct. Uniform standards (¬LJ1) for pure research. Resolute certainty (¬LJ2) in crisis. Symmetric evidence (¬LJ3) in well-calibrated teams. Unbounded deliberation (¬LJ4) for foundational questions. Single-mode reasoning (¬LJ5) in high-tempo ops. Claim-level only (¬LJ6) in trusted institutions. Deploy these bounded in time and scope, with a return trigger.

Key tensions preserved

  • LJ4 ⊥ LJ5: When does inquiry become procrastination?
  • LJ3 ⊥ LJ7: When does “deploying conviction” become rationalizing?

These tensions are productive. Resolving them makes the scaffold weaker — 231 Elo points weaker (Sovereign Judgment vs. Lucid Judgment, same sources, different operator).

Provider results

Provider LJ Elo
Claude Opus 4.6 1983 (highest single cell in tournament)
Grok-4 1869
Gemini 2.5 Pro 1828
GPT-5.4 1769

See Also