The article is complete at 2,275 words. Here's what it covers:
- Opens with the credit scoring motivation and why Logistic Regression is constrained to linear boundaries
- Explains the IQP circuit design: three layers (Hadamard superposition, RZ phase encoding, ZZ ring entanglement), 24 gates at depth 4, no trainable parameters
- Details the nine-way comparison at equal feature budgets: linear transforms add nothing, KPCA adds 0.031 F1, IQP adds 0.055 F1
- Explains the structural reason only Logistic Regression benefits: linear models cannot build non-linear boundaries internally
- Covers feature selection results: RF importance reaches 0.523 F1, diverse/uncorrelated selection drops to 0.496, demonstrating the circuit amplifies structure rather than creating it
- Discusses the quantum-only experiment (F1 0.485, well below augmented)
- Reports feature importance: 10 of top 20 coefficients are quantum features
- Details the Benjamini-Hochberg correction across 12 tests, 3 findings survive
- Addresses limitations honestly: classical simulator, 600M circuit executions needed for hardware, n=8 trivially simulable
- Ends with practical takeaway for practitioners
All section headings are paper-specific. The article uses `IMPORTANT: no` since this is an incremental result on a standard benchmark, not a major milestone.