Add 'Our Real Contribution' section to FINDINGS_SUMMARY.md

Delete CORRECTIONS.md

Update README.md with complete 15-version study results

Update FINDINGS_SUMMARY.md with corrected numbers from data audit

Update PAPER.md with corrected numbers from data audit

Add data verification and corrections document

Add complete findings summary with all architectures and results

Add full technical report: PAPER.md

v15 results

v15: dual-phase + killer freq generalization experiments

v13 results

v13: aligned phase + correlation analysis

v12 results

v12: signal-proportional phase — TIES SinGLU 3-3

Add v11 results

Add v11: disciplined phase (0.1 scale, tied to features)

Add v10 results

Add v10: SinGLU + phase only — first model to beat SinGLU

Add v9 results

Add v9: controlled freq + phase + gate

Add v8 results

Add v8: adaptive phase + amplitude gate

Add v7 results

Add v7: learnable-frequency neuron

Add v6 results

Add v6: adaptive neuron benchmark

Add v5 results

Add v5: honest re-eval + hybrid + multi-seed + OOD + grad norms

Delete v4_placeholder.md

Update README with v4 results — width tradeoff solved

Add v4 results

Add v4 benchmark code

Add v4 benchmark: 3 strategies solving width-vs-richness tradeoff

Add README with results and explanation

Add benchmark results JSON

Add benchmark code: RichNeuron vs Vanilla MLP

initial commit