Principles / The Context Compensation Principle
§ Principle 01 of 23
The Context Compensation Principle
As context size and entropy increase, required reasoning grows nonlinearly.
This is the foundational principle of the taxonomy. It explains why messy prompts don't just produce slightly worse output — they produce disproportionately worse output at disproportionately higher cost. The relationship between context quality and reasoning cost is not linear. Doubling the noise in a prompt doesn't double the reasoning cost — it may triple or quadruple it, because the model must not only process more information but also resolve more conflicts, disambiguate more interpretations, and filter more aggressively. This nonlinearity is why small improvements in context quality produce outsized improvements in output quality and cost.
Why it matters
Teams that understand this principle invest in context quality as a cost reduction strategy. A 20% improvement in context quality might produce a 40% reduction in reasoning cost — not because of a 1:2 ratio, but because reducing noise eliminates the cascading interpretation work that noise creates. This principle also explains why Prompt Hygiene has an ROI that far exceeds the time invested in maintaining it.
In practice
Measure your reasoning token consumption per task over time. If it's rising without tasks getting more complex, you're experiencing the Context Compensation Principle — your context is degrading and the model is compensating with more reasoning. The fix isn't a better model; it's better context.