Hybrid post-quantum TLS deployments pair classical and post-quantum signature algorithms to resist future quantum adversaries. However, to the best of our knowledge, no active IETF TLS, PQUIP, or LAMPS working-group draft treats CPU-cost adversarial algorithm substitution as a first-class availability threat. We define the Cryptographic Amplification Factor ($\mathrm{CAF}$), a metric for the per-resource cost asymmetry that an adversary induces by forcing a TLS 1.3 server onto a high-cost signature algorithm (SLH-DSA) instead of a low-cost one (ECDSA or ML-DSA). We characterise four adversary classes—insider misconfiguration, supply-chain compromise, remote TLS peer, and protocol man-in-the-middle—demonstrate working proof-of-concept exploits on representative testbeds, and evaluate three defensive primitives under realistic deployment conditions. On the liboqs 0.14.0 reference implementation, ${\mathrm{CAF}}_{r_{\mathrm{cpu}}}$ reaches $\mathrm{12,255}\times$ on Apple M-series ARM64 and $\mathrm{14,272}\times$ on AWS Graviton3 (3904–$\mathrm{14,272}\times$ across five microarchitectures). Under a representative two-worker hardware security module (HSM) calibration ($\lambda =13.5$ TPS), substitution drives the modelled queue utilisation $\rho$ past the stability boundary ($1.44\times {10}^{-4}\to 1.787$), a model-independent saturation; on the testbed, we observe server-CPU saturation, accept-queue exhaustion, and dropped connections. Two enforcement defences and a Parallel Signing Architecture provably bound $\rho \le \alpha <1$ by admission control, regardless of attack rate. The exposed surface is concentrated today in high-assurance and financial-sector deployments and widens as FIPS 206 finalises.

Decision-making problems in complex environments are often characterised by uncertainty, vagueness, and dynamically evolving information. In such contexts, decision makers may express hesitant and fluctuating evaluations over time, which cannot be adequately captured by classical hesitant fuzzy frameworks. To address this limitation, time-sequential hesitant fuzzy sets (TSHFSs) have been introduced as an effective tool for modelling temporal hesitancy. However, the development of information measures for TSHFSs, particularly entropy measures for quantifying uncertainty and deriving criteria weights, remains limited. In this paper, we propose a novel class of entropy measures for TSHFSs by constructing transformation mechanisms based on proximity-driven formulations derived from similarity structures. The proposed measures are developed using arithmetic and algebraic operators to capture the dispersion of information across time sequences, enabling a more refined representation of temporal uncertainty. These entropy measures are further integrated into a multi-criteria decision-making (MCDM) framework, where they are employed to determine criteria weights under incomplete information and combined with the TOPSIS method for ranking alternatives. The effectiveness of the proposed framework is validated through comparative analysis with existing TSHFS entropy measures and sensitivity analysis under varying decision conditions. The results demonstrate that the proposed measures maintain ranking consistency while providing improved discrimination and interpretability of alternatives. In particular, the framework effectively captures fluctuating hesitancy and enhances the robustness of decision outcomes in dynamic environments. The proposed approach contributes to the advancement of TSHFS-based decision analysis by offering a mathematically grounded and practically applicable entropy-driven framework for handling time-dependent uncertainty in complex decision-making problems.

Protein structure prediction (PSP) has become a major application of artificial intelligence in computational biology, and recent advances in transformer-based large language models (LLMs) and protein language models (PLMs) have substantially improved sequence representation learning and structural inference. Despite rapid methodological progress, existing reviews remain fragmented and provide limited systematic evaluation of methodological trends, biological applicability, and validation maturity. This systematic review synthesises 38 eligible studies published between 2020 and 2025 using a PRISMA-guided methodology and introduces a Biological-Readiness Taxonomy for evaluating LLM- and PLM-based PSP approaches. The proposed framework systematically classifies existing methods according to biological knowledge source, representation mechanism, structural reasoning capability, biological prediction task, validation maturity, and interpretability. The evidence synthesis reveals a shift from sequence-only protein language models (PLMs) toward hybrid and multimodal frameworks integrate structural, evolutionary, and biological information. Although benchmark performance has improved substantially, most representative models continue to rely primarily on computational evaluation, with comparatively limited biological validation, interpretability, and experimental verification. Overall, this review presents a unified synthesis combining taxonomy-driven classification, structured comparative analysis, and critical evaluation of biological readiness for protein language models (PLMs). The proposed framework offers a structured basis for comparing existing approaches and identifies future research priorities centred on biological validation, explainability, and translational applicability.

Quantum machine learning is commonly benchmarked by matching a quantum head’s parameter count to that of a classical baseline. We show that this control is insufficient. We compare a 32-parameter parameterized quantum circuit (PQC) head against a nominally matched 31-parameter multi-layer perceptron (MLP) on three paraphrase tasks—the Microsoft Research Paraphrase Corpus (MRPC), Paraphrase Adversaries from Word Scrambling (PAWS), and Quora Question Pairs (QQP)—and document two pitfalls that inflate an apparent quantum deficit. First, under the tested protocol, unweighted binary cross-entropy on imbalanced data leads to near-all-positive prediction, producing an illusory equivalence. Second, the surviving gap traces to a readout that sits offset from the decision threshold at initialization ($⟨Z_0⟩=+0.922$), and no circuit parameter acts as a threshold shift that could recenter it. A pre-specified calibration experiment removes 61% to 76% of the gap, and a training-free probe detects the condition in one forward pass. Threshold-free metrics then changed our own conclusions. On PAWS, both heads rank barely above chance (area under the receiver operating characteristic curve, AUC, at most $0.508$), so the difference there is threshold placement. Only QQP retains a residual, and a post hoc control doubling the qubit count narrows it by $42\%$ without removing it (paired $p=0.0009$). This pattern is consistent with a capacity or optimization limitation, but does not distinguish between them. Four of five pre-specified predictions failed. Parameter roles and readout offsets must be audited, not only counts.