The shield is a deployable, streaming repackaging of the framework's joint-shadow-admissibility test. Given a multi-shadow telemetry stream (M ≥ 2 channels reading what should be the same underlying cascade), it emits structural-anomaly events whenever the shadow set's anti-shadow score \(\mathfrak{A} = \mathcal{S}(\{\mathcal{B}(\Delta_{\alpha\beta})\}_{\alpha<\beta})\) exceeds the Theorem 3 precision floor \(\tau_{T3}\).
Theorem anchors
- Theorem 1 (brake threshold) — \(\beta = 1\) is the only universal threshold; the shield cuts there for every domain with no domain-specific tuning constant.
- Theorem 3 (precision floor) — \(\tau_{T3}\) is the cascade-internal noise scale below which the shield reports nothing (no false alarms below the floor).
- Theorem 7 (cadence-stamp) — every event carries the cadence at which it was detected, supporting time-windowed \(d\mathfrak{A}/dt\) drift analysis.
- Theorem 10 (anti-shadow detection) — \(\mathfrak{A}\) is the joint pair-difference brake-rate test; the operative diagnostic.
- Theorem 11 (cross-shadow universality) — \(\sigma_{\text{cross}}\) governs how fast the consensus tightens with M; the shield reports both \(\mathfrak{A}\) and \(\sigma_{\text{cross}}\).
- Theorem 12 (honest-scope meta-theorem) — every event publishes a \(\mathcal{A}\) tuple: tested shadows, excluded shadows, \(\tau_{T3}\) floor, gauge group, application mode.
- Theorem 13 (spectral primitive) — multi-band \(\mathcal{P}\) decomposition separates contamination by band before \(\mathfrak{A}\) is computed.
v2 application-layer hardenings
- L1 — continuous weights, not binary mask. Per-shadow weight \(w_\alpha = \sigma(\text{LOO\_delta}_\alpha / \sigma_{\text{LOO}}) \in [0, 1]\). Clean: ≈1.0; contaminated: ≈0; borderline: ≈0.5. v1 discarded magnitude.
- L2 — per-band \(\mathcal{P}\) decomposition. Haar DWT separates each shadow into n_bands+1 band-signals before \(\mathfrak{A}\). Min-pool across bands gives the per-shadow weight.
- L3 — joint-basket \(\mathfrak{A}\). \(\mathfrak{A}\) is a JOINT test, not per-subset: cross-pairs between coordinated and legitimate members carry the structural mismatch.
- L4 — harmonic-summed autocorrelation. For periodic / Dirac-comb cascades, jitter-resilient G_𝓟 representative; raw periodogram biases under jitter.
- L5 — elapsed-time cascade. Inter-arrival sources use elapsed-time \(\Phi\) (default), not raw arrival rates.
- L6 — compose-aware weighted fusion. When multiple shields are composed, fusion respects the cascade-internal weights.
Full scope: shield/SCOPE.md