HDBLAST + DBLAST: Kernel-free PTA robustness certificate + RS2 curvature killswitch, cross-band closure, K2/Kν universal knee transfer, and LISA-band feature-search contract (NANOGrav 15yr HD)

This deposit packages the current “next-level” **HDBLAST + DBLAST** research bundle: a falsifiable chain linking a PTA spectral knee inferred from the NANOGrav 15‑year Hellings–Downs (HD) signal to:

1) a **kernel‑free robustness certificate** against nonparametric covariance misspecification (structured adversaries; Toeplitz + ε‑budget),  
2) an **RS2 curvature / laboratory-length-scale (“ℓ”) killswitch**, and  
3) **cross‑band predictions** including a forced **LISA‑band knee scale** f* and an inferred **bulk offset** y0/ℓ via ξ = exp(−y0/ℓ).

### What’s new in this version (2026.01.22)

- **Theory Contract v3:** locks the knee definition to **f50** (half‑suppression), standardizing the closure map and eliminating “definition slippage”.
- **RS2 universal K2 knee filter:** explicit brane transfer template
  F(x)=((x^2 K2(x))/2)^2 with fixed shape invariants.
- **K2-corrected bulk‑offset inference:** applies +ln(x50)=0.2920 correction when a pipeline-reported knee corresponds to f50.
- **End‑to‑end runner v4 (SciPy‑accelerated):** produces PTA + LISA reportcards and a joint closure verdict (JSON + PDF card).
- **Robustness beyond Toeplitz:** explicit ε‑budget backstop for near‑Toeplitz covariance violations.
- **Minimal RS2 “bulk impulse → brane bang” note:** toy mechanism bridge clarifying the bulk‑to‑brane trigger vs echo-only fork.

### Reproducibility

The deposit includes runnable Python packets and expected input formats. Some steps require external PTA posterior products (not included); the runner documentation specifies inputs and formats.

### Scientific status

This package is a test harness and an explicit constraint chain for a working higher‑dimensional / brane‑inspired hypothesis. If any link fails under improved data, alternative binnings, or independent pipelines, the hypothesis is constrained or falsified.