Open_Research

Open Research

What is genuinely open, honestly bounded. The solved problems moved to Detection Capabilities; this page is the frontier.

Resolved (context, briefly)

The method-choice question is closed. Passive statistics (FFT, mutual information, transfer entropy, Granger, CCM) failed on this problem class for a structural reason: the receiver's own exploration noise sits ~500× above any coupling signal in passive data. The shipped answer is the impulse protocol — deliberate, guarded pushes with the receiver holding still, detected by CFAR — validated across coupling shapes including deeply nonlinear cascaded channels and slow drift, zero false positives on controls, with a published boundary map. Passive methods remain interesting only as accelerators (ranking candidates for probe prioritization at scale).

Composition is closed to a measured horizon. Nonlinear composition — curves chained through nonlinear switching elements and summed at converging junctions — predicts far-end response within error bars propagated from each piece: 95 of 97 referee points across four chain seeds, a junction, and a live threshold relay, with the propagated bars validated as honest metrology. The bench capability that did not exist (nonlinearity in the edge itself) shipped with the campaign. Published as From Curves to Cascades.

Beyond the Composition Horizon

Roughly two nonlinear hops at noise σ=0.02: beyond that the far end falls below the detection floor, and composing the measured pieces is the only remaining door. Extending the horizon — a quieter instrument, evidence stacked across probes — is open.

Direct vs Transitive Edges

Distinguishing a direct edge from a two-hop path through an intermediate — the diamond topology is the test cell — remains open. Pairwise curves constrain the answer; they do not yet separate it.

Fast Non-Stationarity

Drift slower than the detection window is handled (local reference windows, drift flags on re-measurement disagreement). Phase transitions faster than the window — a system that changes regime mid-probe — remain open. Candidate directions: shorter adaptive blocks, dwell-based protocols, drift-rate estimation as a first-class measurement.

Scaling the Scan

2-6 agents validated. The O(N²) pairwise scan, turn-taking serialization, and scan-rate vs drift-rate at N >> 6 are open — partly engineering (parallel probe groups, spatial locality), partly fundamental (observability limits under drift). Group-scoped coordination and per-receiver curves (one sender walk measures all listeners) already removed one factor of N.

Prediction-Driven Rescan (the maintenance loop)

A frozen measured map enables prediction: "if node A moves to X, node D responds by Y." When prediction fails, the map is stale — and the error localizes WHERE structure changed. Rescan the neighborhood, diff against the frozen map. The prediction error is the compass: it points toward ignorance; probe where it's most wrong. The first rung is live in the engine: failed prediction triggers the re-measure walk, and the neighborhood refreshes. Open: ranking rescans by prediction error across the whole map, and fan-out from a localized miss.

Multi-Cadence Coupling Spectroscopy

The coupling structure revealed at one probe cadence is one view: slow probes see steady-state thermal coupling, fast probes see electrical/capacitive coupling. Different cadences excite different mechanisms; each yields a different graph. When one cadence suffices (dominant mechanism) the single map is honest; multiple concurrent mechanisms need multi-cadence probing, and detecting that need — via prediction error — is itself open.

The Broader Family

The impulse protocol may be one instance of a broader family: mechanisms that turn unavoidable operation into identification. Its carrier is the agents' own exploration (guarded push, held receiver, CFAR). Further carriers already run — one sender walk measuring all listeners, prediction failures localizing what gets re-measured — and cadence is an open dimension above. Two candidates, not built: coded push patterns (distinct signatures per sender, paths separable — aimed at the diamond problem and the serialization limit) and steering-as-experiment (a served wish as carrier, evidence only through reconciliation). The governor rung is experiment selection by expected information value. The family line: each new carrier unlocks the same chain where the first instance cannot reach.

Meta-Optimization of Sensing

The protocol's own parameters are tunable against measurement outcomes: probe design (amplitude, block lengths, pause spacing, walk depth), detection parameters (observation window, significance threshold, permutation count, sampling rate), observer sensitivity, guardrail thresholds, systemtender meta-configuration (trial budget, parallelism, completion criteria). The walk itself is already adaptive — it steers by measured ground, banked curves and unresolved gaps deciding where it probes next — and reconnaissance is a pluggable organ in the trial loop. Self-tuned parameters, calibration scenarios, and objective definitions do not exist yet.

Choosing experiments by expected information value — the sensing strategy choosing its own experiments — is the natural, unsolved layer above that. Prediction error is the compass, the same one the maintenance loop runs on: it points toward ignorance; probe where the map is most wrong. The order of attack follows definability: detection first (parameters well-defined, outcomes measurable), then guardrails, then systemtender meta-configuration, then probe design. Coordination parameters — how aggressively systemtenders decouple, how much they share, when they yield — sit above all of it and are not started.

Hardening the Steering

Basic steering is live: a declared wish is planned against the measured map, acted, judged against its bands, and held while the world drifts; claims that share an input are reconciled at plan time, and the holder corrects a standing wish in place. The open rungs, in order: the landed receipt for the full multi-outcome shape — it has flown end to end and concluded with an exact verdict, one line short of the band — and then many wishes coexisting on one system, kept together where possible, collisions surfaced where not. Each rung gates the next; none is assumed.