The Symmetry Trap
A spatial and allocation failure mode where the model allocates equal analytical weight across all variables regardless of which ones carry the most strategic leverage, producing comprehensively balanced output that is strategically useless.
Definition
The Symmetry Trap is a spatial and allocation failure mode where the model allocates equal analytical weight across all variables regardless of which ones carry the most strategic leverage, producing comprehensively balanced output that is strategically useless.
Why It Happens
Language models are trained on the entire internet, where comprehensive-sounding outputs are overwhelmingly favored. The default probability distribution optimizes for structural symmetry and aesthetic completeness rather than strategic utility. If three frameworks are provided, the model allocates roughly equal depth to each regardless of which one addresses the actual bottleneck.
The Recognizable Signature
Beautiful five-point analyses where every point gets equal treatment. Remove any one point and the actionability of the response doesn't change. The model worked equally hard on everything and disproportionately hard on nothing.
The Cure
The Asymmetric Triage Prior — a meta-cognitive prior that suspends uniform variable weighting and forces the model to isolate the single most asymmetric, high-leverage constraint and route disproportionate analytical density to it.
FAQ
Why does balanced analysis feel satisfying even when it's strategically useless?
Because balance signals thoroughness. The Symmetry Trap is seductive — a five-point analysis with equal depth on every point reads as comprehensive and fair. The problem is that real strategic decisions are never symmetric. Some variables dominate. Treating them as equal produces elegant analysis that points in no particular direction.
How does the Asymmetric Triage Prior cure this?
It forces the model to rank variables by leverage before allocating any analytical depth. The highest-leverage constraint gets deep treatment. Lower-leverage variables get compressed to their consensus-level state. The result is shorter but more actionable.