Order-Matched Forward/Reverse Tests

Updated 13 January 2026

Order-matched forward/reverse tests are specialized protocols that isolate and quantify directional process effects in systems with non-symmetric forward and reverse transformations.
They utilize distinct experimental designs, such as closed-loop σ–T cycles in shape memory alloys and optimal lifting in quotient spaces, to segregate forward and reverse events.
These methods also enable robust reverse calibration in market microstructure models, ensuring precise attribution of effects and reducing confounding influences.

Order-matched forward/reverse tests are a class of experimental and statistical protocols for isolating, quantifying, or testing the effects of directional transformations or flow in systems where forward and reverse processes may yield distinct outcomes. Within distinct research domains—including stochastic order-book modeling in financial markets, high-precision characterization of martensitic transformations in shape memory alloys, and non-Euclidean statistical inference on manifold-structured data—the “order-matched” construct is critical for ensuring that experimental or inferential results are attributable to a single process direction, free of ambiguities induced by overlapping or confounded effects.

1. Fundamental Concept and Motivations

The rationale behind order-matched forward/reverse tests is to identify and separate the contributions of forward and reverse processes in systems where time, sequence, or transformation directionality matters. In statistical modeling of microstructure in financial markets, for example, market orders and cancellations are forward and reverse flows; in materials science, the martensitic transformation is distinctly forward (cooling) or reverse (heating); in geometric statistics, order-sensitive or reflection-invariant landmark configurations necessitate careful handling of directional and labeling symmetries. Order-matching ensures that the effects measured or the hypotheses tested reflect only the targeted direction or type of transformation, not a superposition.

2. Protocols in Closed-Loop Thermomechanics and Martensitic Transformations

In shape memory alloys such as NiTi, order-matched protocols uniquely isolate the plastic strain generated by either the forward or reverse martensitic transformation (MT). The canonical protocol is a closed-loop “rectangle” in stress–temperature (σ–T) space where forward and reverse MT events are segregated into non-overlapping thermal segments. For forward-MT isolation (counter-clockwise path), the cycle is: isothermal starting point in austenite at (T₁,σ₁), cooling to (T₂,σ₁) to induce forward MT, rapid unloading at T₂ to σ₂, heating at constant σ₂ across the reverse MT, and reloading at T₁ to complete the loop. Reverse-MT isolation is achieved by inverting the path (clockwise), starting in martensite and heating under controlled stress to capture only the plastic strain generated in the reverse MT segment. The explicit “order-matching” means each transformation can have its plastic strain signature measured independently (Šittner et al., 2024).

The resulting protocols allow identification of two critical stresses: a forward-MT plasticity threshold (σ_F^th ≈ 500 MPa) and a reverse-MT threshold (σ_R^th ≈ 100 MPa), with corresponding incrementally accrued plastic strains per cycle empirically observed and described by piecewise linear constitutive relations. This clear separation is essential for defining the functional fatigue envelope of NiTi wires and for guiding alloy or process design.

3. Order-Matched Statistical Inference in Quotient Manifolds and Shape Analysis

In geometric statistics, order-matched forward/reverse tests are realized in optimal-lift two-sample testing on quotient manifolds. For spaces such as the reverse-labeling reflection shape space (RRΣ_m^k), data consists of k-landmark configurations modded out by similarity and label-reversal actions. The core inference task is to test for equality of Fréchet means under the quotient metric, accounting for group symmetries and potential asymmetries in the underlying manifold structure (Van et al., 22 Mar 2025).

Order-matched testing is achieved via “individual lifting,” where each sample is optimally registered not to a shared pooled mean but to the mean of the other. This construction uses measurable, almost-everywhere unique and smooth optimal lifts ℓ_p: Q → M relative to reference points p ∈ M. Analytical justifications guarantee L¹ convergence of empirical lift-maps under the strong law for optimal lifts. The individual-lifting test statistics exhibit correct size and dominate pooled-lifting approaches in curved or highly symmetric quotient geometries, as established both by asymptotic theory and simulation.

4. Application in Point-Process Market Modeling: Forward/Reverse Calibration

In point-process models of limit order books, order-matched forward/reverse tests consist of first forwarding simulating synthetic event flows (e.g., a 10-variate Hawkes process generating market orders, aggressive/passive limit orders, and cancellations), injecting these timestamped messages into a matching engine under specified rules, then attempting to recover (reverse-calibrate) the original process model from exchange event streams (trades and quotes only) (Jericevich et al., 2021).

The forward procedure encompasses complete injection logic distinguishing aggressive versus passive orders and the full set of cancellation types. The reverse procedure involves reconstructing latent event types from trade/quote data, and fitting a full Hawkes model by maximum likelihood, using residual diagnostics (Q–Q plots, Kolmogorov–Smirnov and Ljung–Box tests) for model fit. Deviations between forward (synthetic true parameters) and reverse (fitted) parameters quantify the distortion induced by market microstructure. Hypothesis tests (e.g., likelihood-ratio against true parameters) reveal that implementation rules and the matching engine can induce large, systematic distortions—blurring the inference of genuine order-flow parameters.

5. Key Equations and Operational Definitions

Order-matched protocols rely on equations or constructions that guarantee unique attribution to the forward or reverse process under study.

Martensitic Transformation Plasticity:

$\Delta\varepsilon_{p}^{\rm F}(\sigma) = \begin{cases} 0, & \sigma\le \sigma_{F}^{th}, \ A_{F}\,(\sigma-\sigma_{F}^{th}), & \sigma>\sigma_{F}^{th} \end{cases}$

$\Delta\varepsilon_{p}^{\rm R}(\sigma) = \begin{cases} 0, & \sigma\le \sigma_{R}^{th}, \ A_{R}\,(\sigma-\sigma_{R}^{th}), & \sigma>\sigma_{R}^{th} \end{cases}$

with empirically determined $A_F$ , $A_R$ (Šittner et al., 2024).

Optimal Lifts in Quotient Spaces:

$\ell_p(q) = \operatorname*{arg\,min}_{p'\in\pi^{-1}(q)} d(p, p')$

guaranteeing at almost every $q$ a unique, continuous, and (on strata) smooth mapping from the quotient back to the total space (Van et al., 22 Mar 2025).

Likelihood for Hawkes Process Recovery:

$\ln\mathcal{L}(\theta) = \sum_{m=1}^{10}\left[ -\int_{0}^{T}\lambda^m(s;\theta)\,ds + \sum_{t_i^m\le T}\ln\lambda^m(t_i^m;\theta) \right]$

optimized over all process parameters in the reverse calibration step (Jericevich et al., 2021).

6. Performance Guarantees, Limitations, and Domain-Specific Insights

Empirical and theoretical results across domains validate that order-matched protocols:

Allow precise attribution of observed effects (e.g., measurable plastic strain, statistical power in non-Euclidean inference, or parameter distortion due to market mechanics) to a single process or transformation.
Demonstrate sharp, empirically verified thresholds (e.g., σ_F^th, σ_R^th in NiTi alloys) and enable mapping of functional regimes free of degradation or bias.
Afford asymptotic control of test levels and superior testing power in complex statistical manifolds, especially when group symmetries, curvature, or labeling ambiguities are present.
Reveal in market microstructure that forward/reverse blurring by matching engines can induce order-of-magnitude parameter distortions, necessitating inclusion of microstructural rules in inferential modeling.

7. Comparative Overview of Order-Matched Testing Strategies

Domain	Order-Matched Protocol	Key Measurement/Outcome
Shape Memory Alloys (NiTi)	Closed-loop σ–T rectangles, forward/reverse MT	Isolated plasticity, fatigue limits
Shape Statistics/Manifolds	Individual optimal lifts, separate registration	Higher test power, mean differences
Market Microstructure Models	Forward simulation/reverse calibration cycle	Distortion quantification

These strategies converge on a shared methodological principle: unambiguous isolation and quantitative assessment of process-directional effects, yielding more faithful inference, critical threshold identification, and principled system design.

Order-matched forward/reverse tests are thus indispensable where system dynamics are asymmetric with respect to process direction, where superposed or confounding mechanisms obscure attribution, or where subtle geometric or combinatorial symmetries demand careful separation. Their theoretical and applied impact is already substantial across market microstructure modeling, materials science, and geometric statistics (Jericevich et al., 2021, Šittner et al., 2024, Van et al., 22 Mar 2025).