Evidence

Proof It Works

Validated across 6 financial services domains on 30 held-out prompts. Every domain passes acceptance criteria.

Financial Services Validation

DomainTrue CoherenceStatus
Risk Analysis83.7%PASS
Regulatory Compliance83.2%PASS
Credit Assessment83.5%PASS
Market Analysis83.1%PASS
Fraud Investigation83.4%PASS
Financial Advisory83.3%PASS

What Higher Coherence Looks Like

Baseline writes about the task. The adapter writes the artifact.

PatternWith AdapterBaseline
Formal memo headers5/52/5
Facts-first ordering5/51/5
Field-labeled structure5/50/5
Placeholder discipline5/52/5
Shorter, non-redundant narrative4/51/5
Domain-appropriate artifact shape5/51/5

The Benchmark Paradox

MATH benchmark scores vs. True Coherence: r = −0.932. The models that ace benchmarks are the least coherent internally.

In coding, the adapter shows −1.2pp on HumanEval but produces qualitatively superior code. Standard AI fails by misunderstanding the problem. The adapter fails by fumbling execution of a correct understanding—a fundamentally different and more fixable failure mode.

Cross-Model Validation

Coherence adapters work across architectures. The right configuration depends on the model.

ModelBeforeAfter
Mistral-24B31.7%82.5%
Gemma 4-31B9.7%66.0%
Llama-70BSafe gainsZero degradation

See it for yourself