Detection_Capabilities

Detection Capabilities

What the impulse protocol — validated, with measured boundaries — covers. One method: non-destructive guarded pushes, ABA block design (push → pause → compare), CFAR detection with adaptive thresholds. The statistical family is radar and sonar's: constant false-alarm rate detection of a known-shape response in unknown noise.

Linear and Moderately Nonlinear Channels

    Sender parameter → [ coupling ] → Receiver objective

Fully covered. 21-cell sweep on the generic bench: detection floor between 0.1 and 0.2 coupling at noise σ=0.02; measurement error under 5%; zero false positives across all uncoupled controls. Coupling shapes (linear, saturation, threshold, polynomial) do not matter at adequate strength — the response is detected by its ABA signature, not its frequency content.

Deeply Nonlinear Cascaded Channels

    Sender parameter → [ thermal inertia ] → [ multiplicative growth,
    dead zones ] → [ phase-dependent sensitivity ] → [ sensor noise ]
    → Receiver objective

Covered — this is the greenhouse bench, the channel class that defeated every passive method. The impulse survives cascades because it does not rely on the channel preserving any signal structure, only propagating a perturbation: push hard within guardrails, measure whether the receiver's median moves and recovers. Validated bidirectionally at strong coupling, zero false positives on controls.

Non-Stationary Channels (Slow Drift)

Covered at the bench's drift rates. The CFAR reference window is local (a handful of trials), limiting exposure to slow drift; re-measurements that disagree beyond bars are flagged as drift rather than blended — the instrument reports the change instead of smearing it.

Honest Boundaries (Measured, Not Guessed)

Boundary	Where it sits
Noise floor	σ=0.05-0.10 at coupling 0.5: signal permanently below threshold — an SNR limit; more budget does not cross it
Fast phase transitions	Non-stationarity faster than the detection window remains open
Coupling-path nonlinearity	Detection is shape-agnostic; composition validated through nonlinear paths and junctions to a measured horizon — about two nonlinear hops at σ=0.02, beyond which the far end falls below the detection floor
Scale	2-6 agents validated; the coordination regime for 50+ is unbuilt

Beyond Detection: Characterization

Detection reports an edge exists. The same protocol, run to depth, measures its response curve — the full level→shift shape per (sender, parameter, channel), with uncertainty bars. Full story, calibration numbers, and composition results: Characterization.

See Interference Detection for the full method.

Real-World Channel Examples

Shared power bus — load affects voltage (linear additive)
Shared heating/cooling — waste heat through thermal inertia (nonlinear cascaded)
Shared compute — cache/memory contention (nonlinear, state-dependent)
Shared data paths — schema or queue coupling (logical coupling, any shape)

Detection_Capabilities