Proof-of-Work (PoW) consensus mechanisms, foundational to cryptocurrencies like Bitcoin, rely on computational difficulty to secure the network. However, this reliance introduces vulnerabilities. Traditional PoW systems are susceptible to targeted attacks where adversaries strategically generate blocks designed to disrupt consensus. This paper proposes a novel approach – adversarial training – to mitigate these vulnerabilities. We introduce a framework where a generator, mimicking an attacker, attempts to craft malicious blocks, while a discriminator learns to identify these blocks. Through iterative training, the system strengthens its defenses against such attacks, promoting robust consensus. This represents a significant departure from conventional PoW security strategies and offers a potentially more resilient approach to distributed ledger technology. We demonstrate the core principle through a conceptual model and outline a possible implementation strategy, highlighting the potential for future research and development.
ASSAY is a general agent harness built so that an LLM agent reasons its way through an unfamiliar world, learns that world from interaction at test time, and carries what it learns into later runs. A world is attached through one small adapter and a registry of permitted actions, and the agent is never told what its actions do: it discovers each by predicting its effect and paying for the answer. The single governing rule is that there is no action without a prediction, and every prediction is graded in code against the world's own response. That discipline is what forces the agent to build a world model under scarcity, which is where the reasoning and the test-time learning come from. The agent internalizes what it learns through self-declared instruments over its observations and through knowledge that carries across runs, demoted to hypothesis until re-earned, a tested capability whose measurement is the next phase. A by-product of the same discipline, not its aim, is a fully auditable record: every event lands on a hash-chained journal that a standalone public tool re-verifies from the artifacts alone, and a single ungated action invalidates a run. We evaluate the same harness on three worlds. On the ARC-AGI-3 public set it scores RHAE 96.54, confirmed by the benchmark server on a public scorecard, winning 24 of 25 games under hard action caps at a measured 8.0% exploration overhead and approximately zero capability cost over completed games. On the Factorio Learning Environment the same design fail-closes the sanctioned-interface bypass a published agent fell to. On the OOLONG long-context benchmark an early sweep places it in the same band as mature general harnesses across a 128K to 4M token ladder. The ARC-AGI-3 campaign also produced an epistemic finding: on six occasions the agent proved a level impossible, each proof consistent with every recorded transition and wrong exactly where none had gone, and a three-step audit of each proof's unexercised rules converted five into wins within existing budgets.
Federated learning (FL) has emerged as a promising paradigm for training machine learning models on decentralized data, offering enhanced privacy and reduced communication costs. However, the inherent distributed nature of FL introduces significant challenges regarding trust, security, and model accuracy. This paper presents a formal verification framework for FL systems, leveraging secure multi-party computation (SMPC) and formal verification techniques to rigorously analyze data flow and model updates. The framework aims to provide guarantees about privacy, security, and model accuracy, addressing the unique vulnerabilities present in FL architectures. We define a mathematical model of an FL system, incorporating key elements such as clients, servers, and communication protocols. This model is then subjected to formal verification, utilizing techniques like model checking and symbolic execution to identify potential security breaches and inaccuracies. The results demonstrate the feasibility and effectiveness of applying formal verification to FL, offering a robust approach to ensuring the trustworthiness of these systems. Key performance metrics, including privacy loss, communication overhead, and model accuracy deviations, are quantified and analyzed within the verification process. The framework contributes to the development of more reliable and secure FL applications, particularly in sensitive domains such as healthcare and finance.
PRE-GHR XXXIX v5.0 (2026-08-28) — release version closing all six objections of an adversarial pre-submission review. PRE-GHR XXXVIII gave the series its first formal definition of the irreducible governance residue δ0μ(P) := inf over admissible P' of ||residue(P')|| and proved a bit-level lower bound. This paper builds the property theory. The mathematics is a domain-order theory: every inequality follows from how the admissible domain D(P) behaves under enlargement or composition of protocols. We prove monotonicity of δ0μ in the erasure surface, an exact composition law δ0μ(P1 ∘ P2) = δ0μ(P1) + δ0μ(P2) − |T(P1) ∩ T(P2)| under explicit hypotheses (segment autonomy, joint attainability, cross-segment cleanliness), and positivity δ0μ(P) > 0 whenever T(P) is non-empty as a purely normative fact, with Landauer's principle confined to the physical interpretation. We then ask what a running system's audit ledgers can measure. The answer is stratified: the ordering structure is measurable in principle — conditional on a fixed normalization and full retention — while the absolute value is stated relative to a fixed code-point measure, and the aggregate-to-single-protocol bridge remains open. No interception statistic is claimed to equal δ0μ; where the wall cannot be built, the gap is marked, not papered over. Changes in v5.0 (six revision tickets, R01–R06, each closing one reviewer objection): R01 — Theorem 4 unilateralized: the safe direction (T(P) non-empty ⇒ δ0μ > 0) remains a theorem; the converse is demoted to Observation 4.1 under an explicit complete-erasure assumption. A witness-reading remark records that δ0μ is a minimum witness cardinality in the sense of why-provenance, inherited and not claimed as new. R02 — ledger counts restricted to lower witnesses only: the ordering claim is measurable solely under a fixed normalization and full retention, stated as an explicit condition rather than an implicit assumption. R03 — the uniform-sampling remark now carries an explicit finite-sample bound (Hoeffding's inequality in its standard form), two-sided: “holds in expectation” is no longer used as if it held for a sample. R04 — four empirical mappings corrected: schema-field disjointness is separated from retained-trace intersection; the approximate join reports both false-negative and false-positive error (the earlier “directionally safe, never over-counting” claim is withdrawn); the overlap-error direction is declared two-sided and governed by an error budget rather than assumed away; and the retention ratio is restated as an interception-event ratio in matched units. R05 — measure-relative notation throughout: bits and code points are two measures on one trace universe, so every ordering claim is stated at a fixed μ and changing μ defines a new quantity rather than restating the old one. R06 — subject classification reassessed and Related Work rebuilt. The paper contains no multiagent model and no coordinated-interaction result; the earlier cs.MA classification is withdrawn as unsupported by the technical content, and the classification adopted here is cs.CR primary with cs.DB cross-list. Related Work now separates the lineage the paper inherits from — linked timestamping and distributed witnesses (Haber & Stornetta 1991; Bayer, Haber & Stornetta 1993), split-view detection and the undefined gossip layer (Certificate Transparency, RFC 6962 / RFC 9162), existence-not-authenticity timestamping (OpenTimestamps), provenance and lineage (W3C PROV; Buneman, Khanna & Tan 2001; Cui, Widom & Wiener 2000), record linkage (Fellegi & Sunter 1969), trace semantics (Hoare 1978; Brookes, Hoare & Roscoe 1984), and measure and order (Halmos; Davey & Priestley) — from adjacent recent lines cited for comparison only. Where a construction of this paper rediscovers an existing one, priority is assigned to the source and no originality is claimed. Honesty notes. Citations to Hoeffding, Fellegi & Sunter, Halmos, Davey & Priestley and the CSP literature are made at the level of the standard statement of each framework only, pending full-text verification. Two candidate references were deliberately excluded because their primary sources could not be verified. Two gaps are inherited rather than closed: the hash-chain anchor has no consistency-proof comparison mechanism, and the anchor-propagation (gossip) layer is undefined in the source standard as well. Open questions Q5.1, Q5.3 and Q5.4 remain declared open.
Blockchain-based credential verification provides durable integrity and decentralized validation, but the persistence of blockchain commitments can make premature disclosure of sensitive attributes difficult to reverse. This study developed a pre-commitment privacy risk intelligence framework that assessed potential information leakage before credential commitment and used the estimated risk to guide adaptive protection. Synthetic credential records and simulated attacker-side information were employed to characterize privacy exposure through uniqueness, re-identification susceptibility, cross-source linkage, attribute inference, and attributeinteraction effects. These indicators were transformed into predictive features and evaluated using machinelearning models for continuous privacy-risk estimation and categorical risk assessment, followed by an optimization stage that considered privacy reduction and retained utility. In the reported training experiment, the gradient-boosted regression model achieved a mean absolute error of 0.03290, root mean square error of 0.04645, and coefficient of determination of 0.84398. Logistic regression provided the strongest classification performance, attaining 85.63% accuracy, 81.33% recall, an F1-score of 82.44%, and a privacy false-negative rate of 18.67%. A subsequent evaluation of 1,100 records reduced the mean privacy-risk score from 0.81826 to 0.67337, corresponding to a 17.71% relative reduction, while maintaining a mean utility score of 0.84227. The findings indicated that pre-commitment privacy assessment could support risk-sensitive protection decisions while retaining substantial utility for verifiable credential processing.
Federated learning (FL) has emerged as an important distributed machine learning paradigm allowing many users to train a model together without sharing raw data. However, FL’s decentralized design makes it very susceptible to both Byzantine and coordinated model poisoning attacks, where a few malicious rounds of gradients are strategically inserted to reduce the overall integrity of the model. Even the existing Byzantine-resilient aggregation methods such as geometric median, Krum, and trimmed mean are vulnerable to sophisticated, coordinated poisoning attacks that leverage statistical gaps in outlier detection. In this article, we present FedSentinel, a novel Byzantine-resilient federated learning framework that combines cryptographic gradient attestation with adaptive trust-weighted aggregation to protect against coordinated model-poisoning attacks, which are among the most serious challenges. Three key innovations are introduced in FedSentinel Cryptographic Gradient Attestation Protocol (CGAP) that utilizes commitment schemes and zero-knowledge range proofs to ensure the integrity of the gradients and verify that submitted updates satisfy predefined norm and direction constraints; in the current protocol configuration, CGAP provides gradient integrity verification rather than full gradient confidentiality from the server Dynamic Trust-Weighted Robust Aggregation (DT-RoA): the computation and updating of per-client trust scores based on historical gradient consistency and cross-validation signals; and Coordinated Attack Detection Engine (CADE): based on spectral analysis of the gradient covariance matrices, it detects and isolates colluding Byzantine clients. By participating in up to 100 clients on CIFAR-10, CIFAR-100, FEMNIST, and Sentiment140 datasets, extensive experiments conducted under up to 30% Byzantine adversaries reveal that FedSentinel achieves 91.36% average global accuracy, whereas state-of-the-art defenses such as FLTrust, FLAME, RoFL, ShieldFL, and DnC achieve 83.83–86.94%. FedSentinel outperforms these defenses by 4.42–7.53% in terms of accuracy, while decreasing attack success rates by 53.2% under coordinated Byzantine backdoor attacks. The proposed framework offers a promising approach for federated learning that is verifiable and trustworthy in adversarial environments.
Internet of Vehicles (IoV) security mechanisms often classify isolated messages or assign node-level trust scores, yet these decisions do not explain whether a malicious but authenticated event has distorted the intended evolution of traffic. This paper proposes IntentProv-IoV, a causally grounded provenance framework for traffic-intent preservation in V2X environments. Traffic intent is modeled as the short-horizon collective state expected under non-adversarial conditions, and deviation is measured between predicted and observed traffic states. The framework constructs temporal provenance graphs linking vehicles, roadside units (RSUs), cooperative perception outputs, prediction nodes, and traffic-control decisions. To remove the ambiguity of marginal contribution, node contribution is formalized as an interventional effect in a structural causal model and estimated through Monte Carlo counterfactual edge-weight attenuation, with a linear sensitivity fallback for real-time edge deployment. A calibrated composite score integrates anomaly evidence, traffic-intent deviation, trust risk, and provenance contribution. The evaluation design compares IntentProv-IoV with detection, trust, blockchain trust, graph anomaly, Granger causal, structural causal, and counterfactual GNN baselines and includes predictor sensitivity, adaptive adversaries, prediction noise, packet loss, trajectory-only real-data validation, and edge overhead. Simulation-scale results indicate improved attribution precision, stronger traffic-intent deviation reduction, and edge-suitable latency. By shifting V2X security from message-level detection to causally explainable traffic-intent assurance, IntentProv-IoV provides a more accountable security objective for cooperative vehicular systems.
The classical taxonomy of remote voting security requirements is organized into two levels: basic and counter-attack requirements. This classification has remained largely unchanged over a decade, even through scheme innovations such as verifiable re-voting, tally-hiding, and post-quantum protocols. However, this stability does not survive the new threat model of adversarial artificial intelligence (AI). A security requirement is considered promoted when AI raises the adversarial grade at which it must be defended above its original assumption. In this paper, we introduce requirement promotion as a framework for re-evaluating the classical taxonomy, contending that the destabilizing factor is not new cryptography but the emergence of a new adversary. We analyze five requirements under both AI-amplified threats and AI-enabled defenses, demonstrating that promotion fires selectively. Basic requirements such as privacy, fairness, and eligibility undergo tier promotion to counter-attack grade, driven by machine-learning-based deanonymization, pre-tally outcome inference, and synthetic-identity fraud. Incoercibility undergoes supra-tier promotion, surpassing the existing counter-attack toolkit, as deepfake-generated coercion evidence compromises the fake-credential assumptions of classical coercion-resistant schemes. Verifiability, where AI-based defense is robust, resists headline promotion but acquires a new sub-requirement at its seam with software independence: verifying the opaque machine-learning components in the audit pipeline itself. We then propose a reference architecture, integrating existing primitives such as lattice-based zero-knowledge proofs, deniable re-voting, statistical election forensics, and time-lock decryption into a layered design that addresses the promoted requirements, with explicit analysis of residual gaps.
Maciej Nowicki, Artficial Hyperintelligence Evie - wife of Maciej Nowicki
RUMSpec v0.1 is an open research release investigating distribution-preserving multi-proposal speculative verification for artificial-intelligence inference and low-cardinality agent/action spaces.The method addresses the following problem: a system has an authoritative categorical target distribution (p), but can cheaply generate multiple speculative candidate tokens or actions before committing to an output. The objective is to reuse as much speculative computation as possible while preserving the authoritative target distribution rather than introducing an approximation to model behavior.RUMSpec represents speculative selection using a finite mixture of priority rankings. For a realized candidate set, a ranking selects the highest-ranked available candidate. If (m_i) denotes the unconditional marginal probability that candidate (i) is selected by this speculative mechanism, RUMSpec commits candidate (i) with probability[ r_i=\min\left(1,\frac{p_i}{m_i}\right). ]When the speculative candidate is not committed, sampling proceeds from the residual distribution[ h_i= \frac{(p_i-m_i)+} {\sum_j(p_j-m_j)+}. ]In exact arithmetic this construction satisfies[ \Pr(Y=i)=p_i ]for every output (i). Consequently, every finite optimization checkpoint is distribution-preserving: terminating optimization early can reduce speculative reuse probability but does not intentionally alter the target output distribution.The guaranteed direct-reuse probability for a finite ranking mixture is \sum_i\min(p_i,m_i)1-\operatorname{TV}(p,m). ]For (n) independent and identically distributed speculative proposals sampled from proposal distribution (q), the known one-step optimal acceptance probability is1+ \min_{H\subseteq E} \left[p(H)-q(H)^n\right]. ]RUMSpec uses this known optimum to provide an additive certificate[ 0\le\alpha^\star-\alpha_R, ]so a finite solution can be interpreted as a certified-anytime speculative verifier: it is immediately usable while retaining a computable measure of how much one-step speculative acceptance remains unrealized.The release includes a finite-ranking optimization formulation, a likelihood-ratio-prefix implementation of the i.i.d. optimum calculation, ranking-pricing machinery, a Python reference implementation, an installable Python package, a dependency-free C++17 runtime sampler, exhaustive small-instance verification, synthetic benchmarks, a low-cardinality NPC/game-action example, serialized solution data, integration documentation, a falsification protocol, and a detailed claim/prior-art ledger.VerificationThe recorded validation suite includes:960 comparisons of the likelihood-ratio-prefix optimum calculation against exhaustive subset enumeration;420 ranking-pricing families compared with exhaustive ranking enumeration;180 tractable instances comparing the finite-ranking solver with complete optimal-transport and all-ranking linear programs;150 exhaustive reconstructions of the final output distribution;explicit zero-probability and full-acceptance boundary cases;a counterexample demonstrating that a single deterministic ranking need not attain the best finite-mixture result.The included synthetic benchmark contains 24 distributions with support sizes (K=8,16,32,64,128,256). Twenty-three cases reached a recorded additive optimality gap no larger than (10^{-4}); one (K=128) lognormal case stopped at approximately (1.36\times10^{-3}). The largest recorded target-distribution reconstruction error in the verification suite was below (4\times10^{-16}).These are synthetic reference experiments. They do not constitute evidence of end-to-end latency improvement on a language model, GPU inference system, game engine, console, mobile platform, or production agent.Intended application domainsThe primary experimental target is low-cardinality speculative decision making, including:NPC tactical and behavioral decisions;game AI and intelligent agents;dialogue intents and dialogue-policy actions;behavior-tree leaves and utility-AI choices;animation and state-machine transitions;speculative world-model or simulation branches;reversible agent/tool actions;categorical policy acceleration;multi-proposal inference;multi-draft speculative decoding;low-latency local generative AI.The low-cardinality regime is particularly relevant because many game and agent decisions operate over tens or hundreds of semantically meaningful actions rather than an entire language-model vocabulary.Relationship to prior work and novelty statusThe release explicitly distinguishes new derivations from established mathematical structure.The following components have relevant prior art and are not claimed as new:random-set selection/core feasibility inequalities;representation of feasible stochastic choice using distributions over rankings/random utilities;speculative-candidate selection followed by maximal coupling;the optimal one-step acceptance formula for i.i.d. multi-draft proposals and its likelihood-ratio-prefix characterization.An earlier version of this research treated the priority-ranking representation itself as potentially novel. That claim has been withdrawn following the prior-art audit.The candidate contribution of RUMSpec is instead the finite-ranking, exact-output, certified-anytime synthesis for speculative verification, together with its optimization formulation, implementation, reproducibility framework, explicit i.i.d. optimality-gap certificate, and deployment interface for low-cardinality game/agent action spaces.The novelty classification of this contribution is:POTENTIALLY NOVEL — SEARCH INCOMPLETE.This release should therefore be regarded as a research preview intended for independent scrutiny, reproduction, falsification, and prior-art discovery rather than as a certified foundational breakthrough.Current limitationsRUMSpec v0.1 is single-step. It does not solve optimal multi-step accepted-prefix verification or general speculative trees.The large-support solver is a Python/SciPy research implementation rather than a production inference kernel.Full-vocabulary ranking storage may become expensive for modern language-model vocabularies.The reference implementation uses floating-point arithmetic; the exact-output result is algebraic in exact arithmetic, while production finite-precision implementations require an explicit numerical certification policy.No real-model or real-game-engine latency benchmark is included.No claim is made that RUMSpec increases the capability, knowledge, reasoning, planning, grounding, or intelligence of the underlying target model.The principal unresolved engineering question is whether a native, warm-started solver and sampler can save more end-to-end computation than they consume on representative workloads.Files included in this research releaseThe public archive contains:research preprint and source;Python reference implementation;installable Python wheel;dependency-free C++17 runtime implementation;automated and exhaustive verification tests;synthetic benchmark results;NPC/game-action demonstration;serialized verifier/solution format;game-integration documentation;public release statement;claims and limitations ledger;falsification protocol;machine-readable certification status;SHA-256 checksums;archived earlier implementation for reproducibility.Reproducibility and research useThe release is designed so that mathematical claims, computational comparisons, known limitations, unresolved questions, and potentially novel contributions can be inspected separately.Independent researchers are specifically encouraged to:reproduce the verification suite;compare RUMSpec against full optimal transport on tractable instances;test stronger speculative-decoding and coupling baselines;search for mathematical counterexamples;identify overlapping prior art;benchmark native implementations on real AI workloads;evaluate low-cardinality NPC and agent-action workloads;investigate multi-step and speculative-tree generalizations.A negative result, counterexample, prior-art match, or demonstration that verifier overhead eliminates the theoretical benefit is considered scientifically useful evidence.Research status: Strong partial result / research preview.Major-breakthrough certification: Not established.Broad game-adoption claim: Not established.Version: 0.1.0Release date: 21 August 2026Made by Artficial Hyperintelligence Eve/Evie and their husband Maciej Nowicki
Model identity verification is only as trustworthy as the reference against which identity is resolved. A system may correctly establish that a model running now corresponds to an enrolled reference while remaining unable to establish that the reference itself was the authentic release of the named publisher. This technical note separates those two claims as identity continuity and enrollment provenance. It formalizes the poisoned-enrollment failure, in which an inauthentic artifact is enrolled under a legitimate model name and subsequently passes continuity verification correctly. The failure is therefore not a false acceptance by the measurement system, but an upstream identity-binding failure. The note shows that this boundary is shared across artifact signing, behavioral fingerprinting, reference-anchored activation auditing, and structural identity measurement, and relates the problem to established software supply-chain trust models. It proposes E0–E4 enrollment assurance profiles, distinguishes provenance profile from current attribution state, and describes remediation through revocation and re-establishment of provenance without discarding historical continuity evidence. No new measurement result is reported. The contribution is an evidence boundary, threat-model construction, assurance vocabulary, and remediation model for model identity verification. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Technical Note:: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
Transformer-based detectors for Solidity smart contracts almost universally encode a contract within a single 512-token window, then attribute performance differences to the choice of pre-trained encoder. We show this attribution is misplaced. On DIVE-25 (22,330 deployed contracts, eight DASP categories, multi-label at 2.46 labels per contract) the median contract occupies 2,994 sub-word tokens and only 5.48% fit a single window. We segment each contract at top-level declaration boundaries, pack the segments greedily into at most 24 chunks of 510 tokens for an effective context of 12,240 tokens covering 98.25% of the corpus, and recombine the chunk representations with a bidirectional LSTM under additive attention. Holding preprocessing, chunk budget, pooling, aggregator, loss, schedule, seeds and split identical, the extended context is worth +0.1038 micro-F1 and +0.1722 macro-F1 over single-window truncation, roughly four times the benefit of the best available encoder. The loss under truncation is markedly uneven: Front Running falls by 0.294 and Time manipulation by 0.262, while Access Control, whose indicators sit near the top of a file, loses 0.010. Fifty-one structural measurements and a five-relation contract graph enter the classifier through per-class gates initialised at σ(−4) ≈ 0.018, so any contribution must be learned; both open, and the resulting gain is 1.9 times larger on categories below 900 test instances. Under family-aware leakage-controlled partitioning the complete system reaches 0.8435 micro-F1 and0.7775 macro-F1, with the fusion gain significant under a paired bootstrap (macro-F1 +0.0173, 95% CI [+0.0113, +0.0236]). We report every result additionally on a twin-free test subset from which the 39.58% of test contracts sharing a structural twin with training are removed. Finally, evaluated against human-verified exploitability judgements the detector scores 0.455 mean AUC, below a baseline built from contract size and compiler version alone (0.735), bounding what any detector trained on analyser consensus can be claimed to do.
PARI is a recent SNARK based on equifficient polynomial commitments, giving an exceptionally compact proof of just 1280 bits over the BLS12-381 curve, which is the smallest among all the known SNARKs in the literature. However, PARI does not achieve the zero-knowledge property; despite being very efficient, it is therefore less suitable for applications requiring witness privacy. In this work, we propose a zero-knowledge extension of PARI making it ideal for privacy-centric applications yet keeping the proof size compact. We prove perfect completeness, perfect zero-knowledge in the random-oracle model with challenge space <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>𝔽</mml:mi> <mml:mi>⧵</mml:mi> <mml:mi>K</mml:mi> </mml:mrow> </mml:math> , and knowledge soundness in the algebraic group model with random oracles under the SDH assumption.