| name | contradiction-finder |
| description | Use this to surface stated values that collide with measured behavior, and strengths whose shadow side shows. |
Contradiction Finder
Belief vs Behavior
The highest-value section of any audit. Find collisions between what the
subject says they value and what their behavior shows.
Only report contradictions with strong evidence. Structure examples like
this - content comes from data, never assumption:
- "X is stated as the priority, but 63% of relevant sessions go toward Y."
- "Simplicity is requested repeatedly, but scope expands after start."
- "Model A feels fastest, but tasks completed via B show less rework."
- "Testing is called mandatory, but validation collapses under deadline."
Format every survivor as a table row:
| What I appear to believe | What behavior shows | Evidence | Consequence |
Prioritize these findings. They are nearly impossible to see from inside
a single session, which is exactly why they are worth surfacing.
Strength -> Shadow Pairs
No generic strengths. Every pair needs evidence.
STRENGTH: the capability, with episodes.
WHY IT HELPS: observed outcome.
SHADOW SIDE: where the same behavior becomes excessive and costly.
EARLY WARNING: an observable signal that the shadow is starting.
Example shapes - format only, discover the real ones:
- fast experimentation -> rapid learning -> but constant switching blocks
consolidation.
- deep research -> better decisions -> but delays cheap reversible
experiments.
When It Backfires
- Manufacturing contradictions from tone instead of measured behavior.
- Listing strengths without shadow sides. If no shadow exists yet, say
so - that is also information.
One-Line Memory
The gap between said and done is where the leverage hides.
