1. Even Precision Instruments Have Blind Spots

From RN-001 through RN-006, we demonstrated the power of Soft Topological Return (STR): it detects trajectory divergence between hallucination and truthful generation, it is scale-selective, it is a causally manipulable degree of freedom, it drives closed-loop inference control, it guides epistemic foraging in agent architectures, and it powers a topological circuit breaker with ~10-step response.

But a serious question has remained unaddressed: not all "recurrence" is meaningful.

A trajectory may produce high STR values because of genuine dynamical structure — a stable orbit, an attracting manifold — or because of pure artifacts: spatial clustering without temporal return, scale mismatch, or distance distortion in the observation space. If you cannot distinguish these two cases, you will make decisions based on false signals.

2. Two Classes of Failure

Type I: Measurement Condition Violations (Correctable)

The diagnostic is unreliable because the measurement instrument is misconfigured, not because the underlying system lacks structure.

Condition Consequence of Violation Correction
Scale alignment — temporal window must contain the characteristic recurrence period Recurrence entirely missed, or conflated with local autocorrelation Sweep temporal window
Embedding fidelity — observation map must preserve pairwise distance (bi-Lipschitz) Unrelated states mapped to proximity, or proximate states separated Verify/change embedding
Bandwidth calibration — σ must match characteristic state-space separation σ too small → collapse to zero; σ too large → saturation Sweep σ values

Key characteristic: Type I violations can be resolved by modifying the measurement procedure, without changing the underlying dynamical system.

Type II: Structural Absence (Intrinsic)

The diagnostic is unreliable because the underlying system simply does not possess dynamical recurrence structure. No parameter tuning can extract meaningful recurrence from a system that does not recur.

No attracting structure: The trajectory is not confined to any invariant set. STR may register incidental proximity, but this is not recurrence in any dynamical sense.

Density-driven false recurrence: This is the most insidious trap. Data may produce high STR values because of spatial clustering rather than temporal return. A set of points drawn i.i.d. from a Gaussian mixture, arranged in arbitrary order, will produce "recurrence" signals — but there is no dynamical content whatsoever.

Boundary Case: Chaotic Systems

Chaotic systems (e.g., the Lorenz attractor) are neither Type I nor Type II. Structure is present — the trajectory is confined to a compact invariant set and the Poincaré recurrence theorem guarantees almost every state is revisited. But STR cannot stably probe it — chaotic recurrence is aperiodic with heavy-tailed return times, and STR's fixed temporal window captures only a fraction of the structure at any given parameter setting.

VALID
Structure present
Measurement faithful
SCC > 0  ∙  Type I ✓
BOUNDARY
Structure present
Measurement unstable
SCC > 0  ∙  No char. scale
INVALID
No dynamical structure
Density-driven artifact
SCC = 0  ∙  Misleading
Figure 1: Three-regime classification of recurrence diagnostics — Structural Connectivity Coherence (SCC)

3. The Structural Condition: SCC

After systematically analyzing Type I and Type II failure modes, a key observation emerges: every system where STR works stably and reliably shares a common structural property —

The system's trajectory maintains a connected structure in state space that persists under the system's own dynamics.

We term this property Structured Connectivity Coherence (SCC).

4. Validity Statement and Experimental Validation

Validity condition: STR provides a reliable indicator of recurrence structure when two conditions are jointly satisfied: (1) Type I measurement conditions are met, and (2) the system has SCC > 0.

Table 1: Three-Regime Classification — Experimental Validation on Canonical Dynamical Systems
System Attractor SCC Conv CV Param CV Pert Δ Classification
Van der Pol Limit cycle > 0 0.003 0.279 0.012 VALID
Damped Oscillator Fixed point > 0 0.000 0.000 0.076 VALID
OU Process None (stochastic) = 0 0.004 0.141 0.008 INVALID
Gaussian Clusters None (synthetic) = 0 0.000 0.016 0.070 INVALID
Lorenz Attractor Strange attractor > 0 0.010 0.204 0.000 BOUNDARY
VALID thresholds: CVconv < 0.15, CVparam < 0.30, Δpert < 0.25. INVALID and BOUNDARY classifications are structurally determined, not threshold-based.

Critical Observation 1: Stability ≠ Validity

The OU process and Gaussian clusters both produce maximally stable STR measurements (convergence CV ≈ 0), yet they measure density-driven proximity, not dynamical recurrence. If classification relied solely on measurement stability, these systems would pass as VALID — yet no dynamical recurrence exists. Measurement-level diagnostics alone are insufficient; structural analysis (SCC) is necessary.

Critical Observation 2: Boundary ≠ Invalid

The Lorenz system possesses genuine dynamical structure (SCC > 0) but lacks a stable recurrence scale. This is fundamentally distinct from Type II invalidity: the measurement is unstable, not meaningless.

5. Topological Remark

SCC — persistent connectivity of the trajectory in state space — corresponds to zero-dimensional persistent homology features (H₀) in algebraic topology. The Trajectory Recurrence Index (TRI) captures higher-order structure — persistent loops (H₁). A basic topological fact yields:

TRI > 0 ⟹ SCC > 0. Existence before temporal structure.

The presence of temporal recurrence structure (loops) presupposes persistent connectivity (existence). The converse does not hold: a stable fixed point has SCC > 0 but TRI = 0.


Key takeaway: RN-007 is not about improving STR. It is about characterizing where your diagnostic tool ceases to be trustworthy. When you hold a precision instrument, the most important thing is not how precise it is — but knowing in which regions of its dial you can trust the reading, and in which regions it is deceiving you entirely.

This note draws on theoretical and experimental results from "When Is Recurrence Meaningful? Validity Conditions for Trajectory-Level Diagnostics in Dynamical Systems" (Haelio Tang, 2026).