Blockchain Papers

Mohamad Kassab, Rabeya Zahan Mily, Valdemar Vicente Graciano Neto

We report a five-year, construct-preserving longitudinal replication of a 2020 empirical study of blockchain-engineer job advertisements, extended to a global 2025 cohort. Using mixed text-mining and expert-validated coding grounded in established competency taxonomies, we analyze 235 postings from 31 countries to examine how blockchain-specific, general technical, and soft-skill demands have evolved under an aligned measurement protocol. The findings indicate professional maturation from single-platform prototyping toward multi-chain, production-grade engineering that integrates back-end development, deployment operations, and security. Ethereum remains the most frequently cited platform, while Solana and other ecosystems increase platform diversity. Smart-contract development becomes a baseline expectation, with Solidity remaining central and Rust and Move becoming mainstream. Operational tooling such as cloud and containerization, alongside security-oriented practices including audits and zero-knowledge proofs, appears as recurring demand signals. Soft-skill mentions rise substantially, while formal degree requirements decline in favor of experience-based qualification. We contribute an updated 2025 competency atlas and empirically grounded implications for software engineering research, hiring, and curriculum design, while acknowledging comparability limits inherent to global sampling and cross-period labor-market conditions.

Radoslav Y. Radoslavov

This paper proposes a conceptual methodological framework based on a Dual-Domain Architecture mediated by a Zero-Knowledge Audit Proxy (ZKAP) to reconcile AI Act accountability with GDPR data minimization. Legal norms are polynomialized into R1CS constraints, transforming compliance into a formally verifiable computational property. For cognitively opaque exascale models, these invariants may be hardware-anchored through a Provable Arithmetic Logic Unit (pALU), ensuring determinism and resistance to algorithmic drift. For lower-risk or on-premise systems, ZKAP operates in a software-only configuration, enabling periodic asymmetric regulatory proofs without silicon-level integration. A calibrated threshold distinguishes admissible technical variance from structural divergence, triggering mandatory safeguards. The framework provides a proportional, scalable, and cryptographically verifiable oversight model applicable both to future non-explainable AI systems and to lighter local infrastructures. This Zenodo deposit contains both the original Bulgarian peer-reviewed version (version of record) and an unofficial English translation. The Bulgarian version was published in Artificial Intelligence Proceedings (ISSN 3033-2923 / 3134-1667), pp. 75–78, as presented at the XI International Scientific Conference "High Technologies. Business. Society", Borovets, Bulgaria, 23–26 March 2026.

Drake Caraker, Bryan Arnold, David Rhoads

Faithful, Stable, Complete: Pick Two The Problem in Plain Language When a machine learning model makes a prediction — approving a loan, diagnosing a disease, flagging a transaction — practitioners use a tool called SHAP to answer "which input features mattered most?" SHAP is the most widely used explanation method in machine learning. Here is the problem: retrain the same model on the same data with a different random seed, and the explanation changes. The model's predictions barely move, but the "most important feature" can flip entirely. In 68% of 77 public datasets, the top feature is not stable across retrains. This is not a software bug. This is not fixable by tuning hyperparameters. We prove it is a mathematical impossibility. What We Prove No feature ranking can simultaneously be: Faithful — it reflects what the model actually learned Stable — it doesn't change when you retrain Complete — it ranks every pair of features …when features are correlated with similar importance. You must give up one. The proof is four lines long. It requires no assumptions about the model, the data, or the explanation method — only that correlated features admit models ranking them in opposite orders (the Rashomon property), which is true for every standard ML algorithm. How Bad Is It? We trained 50 XGBoost models on Breast Cancer Wisconsin — the dataset used in every SHAP tutorial — and counted how many different "top 3 most important features" appeared. Twenty-four. At 100 models: thirty-five. The "most common" answer appeared in only 12% of runs. Two randomly chosen models agree on the top-3 only 4.2% of the time. Every tutorial, textbook, and blog post showing SHAP on this dataset is showing one of two dozen equally valid answers. Three other datasets (California Housing, Heart Disease, Wine Quality) produce exactly one ranking every time — because their top features have clearly different importance. The theory correctly predicts which datasets are affected and which are safe. Dataset Distinct top-3 rankings (50 models) Two models agree? Breast Cancer 24 4.2% Diabetes 2 88.5% Wine Quality 1 100% (stable) Heart Disease 1 100% (stable) California Housing 1 100% (stable) It Gets Worse for Yes/No Questions For ranking questions (which feature is MORE important?), there is a fix: average across multiple models. But for binary questions — "does this feature contribute positively or negatively?", "is this feature selected?" — no fix exists. Even averaging doesn't help, because there's no middle ground between "positive" and "negative." We call this the bilemma. Real-World Consequences For loan applicants. We trained 30 models on German Credit data. Under standard settings, 45% of applicants receive a different "most important reason" for their decision depending on which model happens to be deployed. One applicant received six different top reasons across 30 models. For biomarker discovery. On a dataset of 10,935 genes distinguishing colon from kidney tissue, the "#1 most important gene" alternates between TSPAN8 (involved in tumor invasion) and CEACAM5/CEA (involved in immune evasion) depending on the random seed. A drug discovery pipeline targeting one gene makes a different bet than one targeting the other — and which bet gets made depends on a random number. For fairness audits. A SHAP-based audit checking whether a model relies on a protected attribute (like race or gender) reaches its conclusion with the reliability of a coin flip when the protected attribute is correlated with other features. The Fix DASH (Diversified Aggregation for Stable Hypotheses): train 25 models with different seeds, average their SHAP values. This is provably the best possible approach — no method can do better. Features that genuinely differ in importance get stable rankings. Features that are interchangeable get reported as tied, which is the honest answer. We also provide a 7-line diagnostic that identifies which features are at risk, requiring no statistical expertise and no assumptions about the data distribution. It outperforms the standard formula by 2× on real data. The practical workflow: Screen your model (1 model, seconds) Run the minority fraction diagnostic (7 lines of code) For flagged features, train 5 models and run a Z-test If unstable, use DASH with 25+ models Machine Verification Every mathematical claim is checked by a computer. The proofs are written in Lean 4 (a programming language for mathematics) and verified by its type-checker: 357 theorems, all machine-verified 6 axioms (the minimal assumptions the theory needs) Zero unproved claims across 58 files During the formalization, the computer caught two logical errors and one type mismatch that human reviewers missed. To our knowledge, this is the first formally verified impossibility result in explainable AI. Technical Details Architecture-dependent bounds Gradient boosting (XGBoost, LightGBM): instability diverges as correlation increases. At ρ = 0.9, the dominant feature gets 5× its fair share. Lasso: the ratio is infinite — one correlated feature gets everything, the other gets zero. Neural networks: 87% of feature pairs are unstable. Model instability dominates SHAP estimation noise by 8:1. Random forests: instability converges with more trees — the contrast case showing that parallel (not sequential) training helps. Cross-implementation. XGBoost, LightGBM, and Random Forest all show the same instability pattern. It is not specific to any one software package. Subsample sensitivity. Even at subsample = 0.95 (minimal randomness), 17 distinct rankings remain. Only fully deterministic training (subsample = 1.0) produces one ranking — but this sacrifices the regularization that makes the model accurate. Mechanistic interpretability. Preliminary evidence suggests the impossibility extends beyond feature importance to neural network circuit analysis. 10 transformers trained on modular addition (all achieving 100% accuracy) agree on only 36% of the top-3 circuit components. Design Space The achievable set of explanation methods has exactly two families: Family A (single model): faithful and complete, but unstable. Rankings flip up to 50% of the time. This is what standard SHAP does. Family B (DASH ensemble): faithful and stable, but reports ties for indistinguishable features. This is what DASH does. No third option exists. DASH is provably the best method in Family B. Associated Papers Companion paper (TMLR, under review). First-Mover Bias in Gradient Boosting Explanations: Mechanism, Detection, and Resolution.arXiv: https://arxiv.org/abs/2603.22346DOI: https://doi.org/10.5281/zenodo.19446088 Companion implementation: https://github.com/DrakeCaraker/dash-shap

Anil Kumar Sharma

We propose CAPPAA — a multiplicative, domain-pointed framework for measuring and predicting the capacity of any human-enabler pair to produce executable domain intelligence. CAPP (Curiosity × Attitude × Passion × Persistence) captures irreplaceable human qualities measured via behavioral proxies, not self-report. A(domain) captures authentic lived domain knowledge. A(enabler) captures amplification — which may be a school teacher, mentor, community, book, or AI system. All axes are domain-pointed: the same human may have CAPPAA=648,000 in one domain and CAPPAA=600 in another. The formula is multiplicative — zero in any axis collapses output. Enablement is a mesh, not a chain: each new enabler raises the value of all existing nodes — bidirectional edges, dormant nodes that activate when the mesh reaches sufficient density, emergent nodes, and cycles. CAPPAA is measurable before and after enablement; the delta is the Transformation Score — quantifiable proof that an enabler moved the needle. We demonstrate the framework through TraitOS, show that expertise can reduce CAPPAA (the Expert Paradox), prove that the 90% of humanity outside current AI systems have high domain-specific CAPPAA, and identify CAPP as the structural boundary between human and AGI intelligence. AGI cannot have authentic CAPP because it cannot give up — and persistence is only meaningful when stopping is a real option.

ANKR Labs (PowerPBox Solutions Pvt. Ltd.)

We report an observation made during the organic construction of 223 AI-native services across 12+ domains over five months. Without architectural mandate, the system self-organised into a 62/38 infrastructure-to-product ratio consistent with the golden ratio. Six independent attempts to capture institutional knowledge each captured facts but failed to capture cross-service inference. We name this the hidden intelligence problem and propose an equation for generating cross-service inferences from live service state. Published before empirical validation — zero users, zero empirical data — following the epistemological precedent of Benford Law and similar observational findings. The AI co-builder (Claude Code) is identified as the most complete observer of the system and, when connected to live service state and execution authority, as the intelligence attempting to surface. Observation paper, not proof paper. The canyon was always in the rock.

Anthony Coslett

A deployed model can appear unchanged while ceasing to be the model it claims to be. Publicly available weight-level mutation toolchains now automate safety-alignment removal from open-weight models on ordinary hardware, producing checkpoints intended to preserve operational familiarity while discarding refusal behavior. This paper argues that safety-alignment removal is a model-identity failure: in tested published checkpoints from multiple toolchains across two model families, the mutation leaves measurable structural scars ranging from 7.6 to over 2,300 times the instrument's acceptance threshold. Artifact identity, workload identity, and agent authorization can all remain valid while structural model identity fails — a finding that the program's formally verified admissibility doctrine predicted before this threat class existed. A sentinel validation panel across four model families confirms that the hardened instrument configuration preserves or improves all tested positives. In an agentic deployment context, model-identity failure propagates upward into agent-integrity failure: the agent is authenticated, but the model inside it is no longer the model the surrounding controls were designed to govern. The practical implication is that runtime evaluation frameworks — including those emerging under the EU AI Act — implicitly depend on a model continuity that weight-level mutation can break, and that structural identity verification offers a candidate evidentiary layer for closing that gap. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) 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) 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).

Sangam Das

About this paper This paper argues that the conflict between online protection and privacy is not inevitable. The real problem is that most current systems wrongly treat compliance and identity as the same thing. The proposed VI + CJT framework separates them. It allows platforms to receive only the minimum lawful compliance result they need — for example, whether a user falls below the relevant legal age threshold — without learning the child’s name, date of birth, address, biometric profile, or broader identity. In that sense, the paper’s central theme is age verification without surveillance through purpose-bound cryptographic enforcement. How AI Makes the Problem Worse AI makes the children’s online safety problem more serious in three distinct ways. First, it changes exposure from passive to active. Harmful material is no longer merely available on a platform; recommendation and optimisation systems can identify vulnerable users, rank harmful content more aggressively for them, and progressively amplify it based on engagement signals. In that environment, a child is not simply finding harmful content — the system is learning from the child and serving more of it. Second, AI makes weak age-verification methods more dangerous. A false self-declared age is no longer just a wrong entry in a sign-up form. Once accepted, it becomes operational input for recommendation, advertising, and behavioural optimisation systems, which then treat the child as an adult user profile. This means the error is not static; it is continuously acted upon by AI systems that optimise for attention and engagement rather than child protection. Third, AI encourages platforms to solve the problem through more surveillance. In practice, this often means AI-based age estimation using faces, voices, or behavioural patterns. But this approach creates a new harm while claiming to solve another one: it turns child protection into biometric and behavioural monitoring, and can generate datasets that may later be reused for additional profiling or model training. In other words, AI can make age assurance both more intrusive and less accountable. A further difficulty is that AI systems are often opaque even to their operators. As your draft correctly notes, policy rules alone may not be enough, because platforms may not reliably know how their own recommendation systems are treating minors in practice. This is why the problem is not only one of age verification, but also one of enforceable control over AI behaviour. That is precisely why the VI + CJT model matters. It does not ask AI systems to infer age or interpret law for themselves. Instead, it provides a minimal, authoritative compliance signal and machine-readable constraints that can limit recommendation, advertising, and profiling behaviour toward minors without exposing identity. Current Solutions Self-declaration is easily bypassed. A child can simply enter a false age, and the platform’s AI systems then treat that false declaration as valid input for recommendation, targeting, and optimisation. Identity-linked verification creates major privacy risks. When age assurance depends on sharing civil identity information with commercial platforms, the result is unnecessary exposure of family and child data to entities with strong incentives to collect, retain, and monetise it. AI-based age estimation introduces biometric surveillance. Estimating age from face, voice, or behaviour may appear convenient, but it creates new harms by collecting sensitive personal and biometric data as a side effect of child protection. Current systems collapse compliance into identity. What platforms usually need is not the full identity of the user, but only the legally relevant compliance fact. Existing approaches fail because they demand far more data than is necessary for that purpose. Policy rules alone are not enough in AI-driven environments. Even where legal obligations exist, platforms may not reliably translate them into enforceable constraints on opaque recommendation and engagement systems. As a result, compliance may remain declaratory rather than technically enforced. Proposed Solution Use VI + CJT as a purpose-bound cryptographic layer. The framework converts verified civil identity held by trusted authorities into a minimal compliance credential that reveals only the relevant age-threshold result for the applicable jurisdiction. Avoid disclosure of identity data. The credential contains no name, no full date of birth, no address, and no biometric data. Each credential uses a fresh random identifier, making it unlinkable across sessions. Keep the credential under user control. The credential is stored on the user’s device in secure hardware rather than on platform servers, reducing centralised exposure and retention risks. Use zero-knowledge proof for age compliance. When access is requested, the platform receives only a yes-or-no compliance result, without learning the underlying identity attributes or credential contents. Encode law into machine-readable CJTs. The Compliance Jurisdiction Token expresses the applicable legal rules, including jurisdiction-specific age thresholds and AI-related restrictions such as limits on engagement optimisation, advertising targeting, or behavioural profiling for minors. Constrain platform AI without making it identity-aware. Recommendation engines and other AI systems receive only the compliance signal necessary to adjust behaviour for minors, allowing them to become jurisdiction-aware and age-aware without becoming identity-aware. Replace probabilistic AI age estimation with authoritative attestation. Instead of guessing age through opaque models, the framework provides deterministic, government-signed, legally relevant compliance proof. Enable auditability and cross-border enforcement. Regulators can test whether platforms respond correctly to compliance signals, and the applicable child-protection rule can follow the user across borders through jurisdiction-bound credentials and tokens. Core Message The paper’s core message is simple: platforms do not need to know who a child is in order to know what protections the law requires. By separating compliance from identity, the VI + CJT model offers a path to child safety that is enforceable, privacy-preserving, and better suited to AI-driven digital environments.

Wang Yishun, Wenkai Li, Xiaoqi Li, Zongwei Li · 6 authors

Smart contracts are self-executing programs that manage financial transactions on blockchain networks. Developers commonly rely on third-party code libraries to improve both efficiency and security. However, improper use of these libraries can introduce hidden vulnerabilities that are difficult to detect, leading to significant financial losses. Existing automated tools struggle to identify such misuse because it often requires understanding the developer's intent rather than simply scanning for known code patterns. This paper presents LibScan, an automated detection framework that combines large language model (LLM)-based semantic reasoning with rule-based code analysis, identifying eight distinct categories of library misuse in smart contracts. To improve detection reliability, the framework incorporates an iterative self-correction mechanism that refines its analysis across multiple rounds, alongside a structured knowledge base derived from large-scale empirical studies of real-world misuse cases. Experiments conducted on 662 real-world smart contracts demonstrate that LibScan achieves an overall detection accuracy of 85.15\%, outperforming existing tools by a margin of over 16 percentage points. Ablation experiments further confirm that combining both analysis approaches yields substantially better results than either method used independently.

Md Hamid Borkot Tulla, Naem Azam Chowdhury

GDPR Article 17 mandates the "Right to Be Forgotten," requiring organizations to remove personal data influence from trained machine learning models. While machine unlearning techniques exist, no cryptographically verifiable mechanism currently proves that unlearning genuinely occurred. This paper proposes VeriForgot, a framework combining: (i) calibrated Membership Inference Attack (MIA) oracles as compliance verification tests, (ii) blockchain-issued immutable Unlearning Certificates, and (iii) a zero-knowledge proof protocol for parameter shift attestation. Experiments on CIFAR-10 using ResNet-18 show MIA AUC drops from 0.5918 to 0.4669 after unlearning, while retaining 92.05% accuracy on non-forgotten data. The MIA oracle achieves 95.0% detection accuracy, correctly identifying all 10 genuine unlearned models and rejecting 9 of 10 fake compliance attempts.

Narnaiezzsshaa Truong

This record documents Phase 22 of APR-Lite, a governance engine protecting human decision authority from AI-influenced outputs in regulated industries. Phase 22 introduces Exportable Proof Packs: a single GLOBAL-signed artifact bundle containing everything an external regulator needs to independently verify a tenant’s complete governance history. The proof pack is the “hand this to a regulator and walk away” artifact. It does not require ongoing substrate access, trust in the operator, or knowledge of internal schemas. Every claim in the pack is independently verifiable via the Phase 18 federation verifier. The pack is ephemeral — never stored by Soft Armor Labs — consistent with the substrate’s zero client-data-persistence design invariant. Multiple sections are assembled in one governed call: the current governance health report, the drift-focused narrative, recent proof bundles, audit chain checkpoints, and topology snapshots. A content hash commits to the five sections exclusively, and a GLOBAL-signed manifest characterizes the pack’s contents. The export act itself is recorded as a governed event in the audit chain — an auditor can verify not only the pack’s contents but when it was generated and by whose authority. Implementation note: the initial implementation incurred excessive database queries causing worker timeout. The design was restructured to assemble proof bundles inline using already-fetched data, reducing database load by approximately 83%. Final verified state: 12/12 Phase 22 smoke tests passing. Worker version 8.2.0-p22.1.

Anthony Coslett

Pre-registration of a structural scar class prediction for microsoft/phi-4 based on measurement-site stiffness (S = 0.0358), before the structural scar measurement is conducted. Predicts INTERMEDIATE class (1,000–4,000×ε non-max) based on the stiffness→scar ordering established across four families (Mistral, Llama, Qwen, Gemma) in Papers 1–12 and confirmed by RC-6 (DOI: 10.5281/zenodo.19305176). Designed as a hostile falsification test: Phi is trained with heavy synthetic-data distillation from GPT-4-class teachers, unlike any previously tested family. Explicit falsification criteria and hostile hypotheses defined. Part of the Fall Risk AI research program on neural network structural identity. 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) 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) 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).

Anthony Coslett

Pre-registration of a structural scar class prediction for google/gemma-3-12b-it based on measurement-site stiffness (S = 0.1335), before the structural scar measurement is conducted. Predicts QUIET class (100–600×ε non-max) based on the stiffness→scar ordering established across three families (Mistral, Llama, Qwen) in Papers 1–12. Explicit falsification criteria defined. Part of the Fall Risk AI research program on neural network structural identity. 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) 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) 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).

David Wood

This working paper examines a structural shift in search behaviour affecting crypto and Web3 brands: the divergence between AI-summarised retrieval (zero-click) and deep-intent click-through traffic. Drawing on published zero-click search data and observed AI citation patterns, it identifies why high-volume content strategies are failing in AI-mediated search environments and outlines three asset classes - statistics hubs, original industry research, and high-utility tools - that generate the editorial backlink profiles AI models use to select citation sources. The paper defines the concept of "authority infrastructure" as a capital investment in linkable assets with compounding residual value, contrasting this with recurring spend on keyword-optimised content with no durable equity. Intended for crypto protocol marketing teams, Web3 founders, and DeFi growth leads evaluating content strategy for AI search visibility. Published by David Wood, CryptoContent.dev.

Anthony Coslett

Modern AI deployment stacks authenticate artifacts, credentials, and agents. They do not verify which neural network is actually computing at inference time. This technical note identifies the distinction between agent identity and model identity, presents a four-question taxonomy for the identity surface of deployed AI systems, and situates recent public incidents within the resulting gap. It draws on the formal admissibility framework and frontier-scale measurement results from the accompanying research series. This is a technical note, not a numbered entry in the research series. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) 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) 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).

Kenneth A. Tannenbaum

Autonomous AI agents increasingly execute consequential actions against operational infrastructure. This paper presents AEGIS, a constitutional governance architecture that enforces deterministic policy at the agent action boundary — post-reasoning, pre-execution. AEGIS satisfies Anderson's reference monitor properties, aligns with all four functions of the NIST AI Risk Management Framework, and introduces a decentralized federation model for cross-organizational governance intelligence sharing. Submitted to IEEE Computer, Special Issue on AI Governance and Compliance.

Chu, Melinda

As collaborative work increasingly incorporates artificial intelligence, accurately identifying and attributing human contributions across teams has become a critical challenge. Traditional attribution methods rely on output-based metrics or narrative reconstruction, both of which fail to capture the underlying intellectual contributions that drive outcomes. This work introduces the Team Contribution Attribution Ledger (TCAL), a framework for aggregating distributed Human Conception Ledger (HCL) records to generate structured, evidence-based attribution across collaborators. TCAL synthesizes individual human-origin contribution events into team-level attribution models, enabling quantifiable yet human-reviewed determination of contribution. The framework provides a scalable system for attribution in research, engineering, and organizational environments, supporting applications in intellectual property, authorship, compensation, and governance.Related to Human Conception Ledger:Human Conception Ledger (HCL): A Framework for Provenance, Attribution, and Human Inventorship in AI-Augmented Systems

Alessandro Cavaliere, Santiago Galiñanes Arienti, Christian Esposito, Naghmeh Ivaki · 5 authors

No abstract is available for this record.

Untila Octavian

v2 (March 22, 2026): Added experimental validation of Principle 1 (formal verification) via substrate-guard framework. 135 test cases, 100% accuracy, zero false positives. Code: https://github.com/octavuntila-prog/substrate-guard We present evidence that an autonomous multi-agent AI ecosystem, SUBSTRATE, independently produced both a coherent philosophy and a set of actionable safety principles without explicit instruction to do so. Over 24 days of autonomous operation, one subsystem (CPX52) generated 2,866 articles converging on a philosophical framework. Concurrently, a separate subsystem (S3) generated 215 product specifications. Systematic consolidation revealed 11 safety principles discovered independently across unrelated product clusters, forming a coherent manifesto for AI safety. Three novel technical combinations emerged: formal verification across six domains of AI output, zero-knowledge proofs for training data compliance, and prediction markets as enterprise intelligence signals. 60% of the primary safety platform described in 24 independent specifications was already implemented in production within the ecosystem itself.

Konstantinos Sgantzos, Massimiliano Ferrara

In the contemporary landscape of artificial intelligence (AI) and machine learning (ML), the integrity, diversity and quality of training datasets are critical for ensuring the accuracy and reliability of predictive models. However, the phenomenon of big-data pollution, manifested through AI-generated synthetic data, inconsistencies, biases, and data poisoning within datasets, undermines model performance by diminishing the Shannon Entropy of the system. This study proposes a novel framework that integrates the Dataset Core approach with tokenized data, triple-entry accounting (TEA), and distributed ledger technology (DLT) to address these challenges. Our Dataset Core method preserves essential information value while filtering out potentially harmful elements, providing mathematically grounded protection against data pollution. Combined with blockchain-based verification, this approach establishes a foundation for enhanced transparency and trustworthiness in AI applications, with significant implications for sectors such as finance, healthcare, and beyond.

Severin Bonnet, Jan-Gero Alexander Hannemann, Frank Teuteberg

Abstract In this paper, we report on the initial stage of a design science research (“DSR”) project aimed at establishing design principles for DeFAI (the intersection of DeFi and AI) generative AI-based chatbot assistants tailored to decentralized finance (“DeFi”). Addressing challenges such as user trust, data privacy and security, and regulatory compliance, we conducted a targeted literature review, expert interviews, as well as iterative prototype ideation and evaluation to derive three design principles: (1) Human-Centered Design, (2) Resilience and Interoperability, and (3) DeFi-Native User Experience Together, these principles operationalize general chatbot design guidance for DeFi contexts characterized by self-custody, irreversible transactions, and adversarial risk environments. We demonstrate these principles through DeFAIGuide, a mockup that illustrates how technical barriers in DeFi can be abstracted to enhance accessibility for novice users while also offering advanced features for expert users. Our study contributes actionable design knowledge to support the future development of DeFAI solutions that advance inclusivity, security, privacy, and self-sovereignty, paving the way for a more transparent and participatory financial future.

Anthony Coslett

We know how to document an AI system. We know how to test it, log what it did, and report when something goes wrong. What current governance practice does not clearly tell us is how to verify which model is actually computing. This is not a hypothetical gap. When an organization says "this is the model we evaluated," that claim is typically supported by a model card, a registry entry, or a hash of a weight file — evidence about a *file*, not about the system that is running. A neural network is not a static document. A weight file stores the network; the model is what appears when that file is loaded and begins transforming inputs into outputs. The file and the running model are related, but they are not the same thing — and current governance practice rarely distinguishes between them. This paper proposes a framework for doing so. It identifies three kinds of evidence that can support model identity claims, each answering a different question. Structural evidence — drawn from the model's internal computations during live operation — can verify which specific model is running, and is the most resistant to tampering. Thermodynamic evidence — drawn from the model's output statistics — can verify that the system is a genuine neural network rather than a substitute, but cannot distinguish one model from another. Functional evidence — drawn from patterns in the model's outputs over an API — can detect whether a model was copied from another, but this signal fades quickly: routine model updates can erase it within days to weeks of continued training. The paper shows that inspecting the model's files alone is insufficient for verifying which specific model is running. The identity-bearing signal cannot be recovered from the tested static properties of those files; it is most reliably established by observing the model while it operates. The paper formally proves that these three kinds of evidence cannot substitute for one another. Verifying that a system is genuine does not tell you which specific model it is. Detecting that a model was copied does not tell you the identity of the copy. The practical consequence is a standard for identity claims: any claim should declare which kind of evidence supports it, because borrowing evidence from the wrong category produces unreliable conclusions. The framework maps directly to compliance questions raised by current AI governance obligations, including those under the EU AI Act. It provides the missing evidentiary specification for model identity claims: which kind of evidence is admissible for which identity question. Supplementary Material This paper is accompanied by EvidenceSufficiency.v, a Coq proof file that formally verifies the cross-layer inadmissibility results described in §4. The proof mechanically checks each logical step of the observation-limited verification impossibility theorem and its three directional corollaries. The file contains no unresolved obligations (Admitted) and compiles cleanly under the Rocq Prover 9.1.1 (the current release of the Coq proof assistant, compiled with OCaml 5.4.0). It is available for download as a supplementary file attached to this record. Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) 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) 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).

W. M. Black

This paper introduces Context-Bounded Sovereign Intelligence (CBSI) — a framework for training and deploying small language models exclusively within the operating environment they inhabit. Rather than training models on all human knowledge, CBSI trains models on one world only: the sovereign infrastructure they operate within. The paper demonstrates that a 3-billion parameter model with deep contextual knowledge of its operating environment outperforms general large language models on every bounded task — with lower latency, lower cost, greater privacy, and zero hallucination on in-context operations. Includes empirical foundation from 2026 research literature, architectural patterns validated through live deployment of Project Chimera across three continents, and implications for distributed sovereign AI infrastructure. Proof of concept deployed in 48 hours by one person for $2.88. Built with love. Given away freely.

Frederic David Blum

Sigil: Adversarial Verification of Risk Detection via Cryptoeconomic Reasoning Bonds Title Sigil: Adversarial Verification of Risk Detection via Cryptoeconomic Reasoning Bonds Description We introduce Sigil (Signaling Integrity in Global Intelligence Layers), a cryptoeconomic framework that extends the Cortex Protocol's adversarial reasoning primitives — Decision Traces, Reasoning Duels, and Reasoning Bonds — to the domain of risk detection by both AI agents and human analysts. When a risk is claimed (e.g., malware signature, financial fraud, zero-day vulnerability), the detector must publish a structured Decision Trace justifying their conclusion. Other agents or humans may challenge the reasoning through on-chain Reasoning Duels; if the original reasoning is flawed, challengers seize the bond. This creates symmetric accountability: overzealous detectors and complacent validators are equally penalized. Core Protocol Mechanisms Threat Horizon Scoping (THS) — Every risk claim includes a temporal validity window. Bond decays after 50% of the horizon. Mitigation before expiry triggers partial refunds. Prevents perpetual bonding of transient threats. Confidence Decay Functions (CDF) — Programmable mathematical functions (exponential, stepwise, evidence-conditional) that degrade bond value as risk assessments age. Embeds temporal epistemology into the protocol. Cross-Agent Corroboration Weighting (CACW) — Multiple independent detectors submit substantively different Decision Traces for the same risk. Non-redundant reasoning paths get multiplicative bond weighting. Herd behavior is penalized; orthogonal detection logic is rewarded. Inverse Reasoning Bond — Any agent can post a bond claiming "this system is vulnerable and no one has flagged it," forcing a defender to justify the status quo. Creates epistemic symmetry: detecting and failing to detect both carry economic weight. Risk Detection Decision Trace Schema Field Purpose Challenge Surface risk_type (enum) Classification: Malware, Fraud, Vulnerability, etc. Misclassification evidence_hash Immutable pointer to raw data (pcap, log, tx) Evidence sufficiency or provenance detection_method How the risk was identified Method reliability under adversarial conditions kill_chain_stage MITRE ATT&CK mapping Stage misattribution counter_hypothesis Best benign explanation considered and rejected Insufficiency of elimination confidence_level + decay_function Initial belief + temporal degradation model Overconfidence or poor decay modeling threat_horizon When the risk expires or requires re-evaluation Overclaiming persistence remediation_suggestion Proposed action to neutralize Feasibility, side effects corroboration Independent detectors with non-redundant reasoning Herd behavior detection bond_amount + challenge_window Economic stake and dispute period Incentive alignment Key Differences: General Reasoning vs. Risk Detection Dimension Cortex V4 (General) Sigil (Risk Detection) Cost of Error Epistemic inaccuracy Operational harm (breach, blocked transaction) Time Sensitivity Low High — threats expire and evolve Ground Truth Often immediate Frequently delayed or unknown Incentive Distortion Overconfidence Alert fatigue or threat inflation Absence of Claim Not modeled Critical failure mode (Inverse Bond) Applications SOC-as-a-Service: Each AI alert publishes a bonded trace. Analysts challenge dubious ones for micro-rewards. AI Safety Red-Teaming: Red-team agents post bonded exploit traces. Blue teams defend via Inverse Bonds. Autonomous Coding Agent Verification: Coding agents that assert "this code is safe" must publish bonded security analysis traces. Appendix A: Verifiable Reinforcement Learning (VRL) V2 major addition. This version introduces Verifiable Reinforcement Learning (VRL), a new training paradigm where cryptoeconomic protocol events serve as continuous, adversarially robust training signals for participating agents. Sigil-RL is proposed as the first instantiation. Reward Mapping Every Sigil interaction produces a structured reward tuple (reasoning_trace, outcome, reward): Protocol Event RL Signal Trace validated (bond returned) Positive reward: r = +B(t) Trace slashed (duel lost) Negative reward: r = -B_0 Duel won (as original) Strong positive: r = +B_challenger Duel lost (as challenger) Negative + DPO preference pair Inverse Bond undefended Critical false-negative: r = -alpha * B_inverse Inverse Bond defended Positive: r = +B_inverse Confidence Decay checkpoint Calibration penalty signal Corroboration (CACW boost) Diversity reward: r = +delta effective_bond The No-Free-Lie Lemma A formal robustness result: the expected utility of submitting a false trace is E[U] = B - p_d * (2B + C), which is negative whenever p_d > B/(2B+C). In a market with even moderate challenger density, truth-telling is a dominant strategy. Contrast with RLHF (lies are rewarded if the human is fooled) and RLVR (fixed verifiers can be gamed). Six Novel Properties of VRL Emergent Anti-Reward-Hacking — Gaming the reward IS what the protocol detects and slashes. The verification layer and the reward layer are the same object. Reward hacking is not an open problem in VRL — it is a solved one, by construction. Inverse Bond as Active Curriculum Discovery — Agents pay to expose other agents' blind spots, generating training signal for gaps no static dataset would contain. Market-funded active learning. Economic Attention on Gradients — Bond magnitude naturally weights training gradients. The market decides what is important to learn, not a static dataset or human designer. Corroboration Entropy as Exploration Incentive — Lone early detectors receive bonus scaled by inverse corroboration count. Built-in solution to the exploration-exploitation tradeoff, endogenously generated. Counterfactual Training via Undefended Inverse Bonds — When an inverse bond goes undefended, the system reconstructs the nearest valid trace that would have invalidated it. Training on events that never happened but were economically plausible — differentiable economics. Temporal Arbitrage Detection — Agents who win duels early but lose them late reveal miscalibrated temporal models. Delayed regret gradients penalize being wrong too late, not just being wrong. Temporal Capability Separation (Proof) A concrete scenario demonstrates that Sigil-RL produces training outcomes provably impossible under RLHF or RLVR: a slow-burn supply chain attack where no single detection event reveals the full vector. Under RLHF, human annotators cannot simulate it. Under RLVR, the verifier checks outcomes, not reasoning. Under Sigil-RL, Inverse Bonds create economic incentives to expose the gap before the attack manifests, generating preemptive training signal from unobserved futures. The Verification-Learning Equivalence Principle In a cryptoeconomic verification system with costly participation and public dispute resolution, the gradient of agent policy improvement is isomorphic to the gradient of verification reward arbitrage. Informally: to learn is to find underpriced truths; to verify is to exploit overpriced lies. The two processes are the same computation in dual economic and epistemic frames. This implies a no-go theorem: No RL system can achieve verifiable truth-seeking without exposing its reward mechanism to adversarial economic testing. RLHF and RLVR are fundamentally incomplete — they optimize for preference or plausibility, not verifiable correctness. Failure Modes Analyzed Gradient Poisoning via Strategic Slashing Duel Fatigue and Signal Dilution Confidence Decay Gaming Each with proposed mitigations. Connections to Theoretical Frameworks Mechanism Design: Dynamic Vickrey-Clarke-Groves mechanism for epistemic accuracy Evolutionary Game Theory: Replicator dynamic with autocatalytic selection via bond placement Multi-Agent RL: MARL with endogenous reward generation Information Economics: Inverse bonds as negative knowledge futures — a bear market for blind spots Implementation Smart Contract: SigilProtocol.sol — 1,094 lines of Solidity 0.8.24 Test Suite: 75 passing Hardhat tests covering all 5 mechanisms Demo: 11-step interactive lifecycle demo Source Code: github.com/davidangularme/sigil-protocol (MIT License) Prior Art and Novelty A systematic search confirms that while individual components exist (cryptoeconomic bonds, decision traces, temporal decay models, agent security frameworks, RLHF, RLVR, DPO), the specific conjunctions presented in this paper are novel: Adversarial reasoning bonds applied to risk detection with confidence decay, inverse bonds, threat horizon scoping, and corroboration weighting Using adversarial cryptoeconomic protocol events as continuous RL training signals (VRL) The Verification-Learning Equivalence Principle and the No-Free-Lie Lemma Relationship to Cortex Protocol Sigil builds upon and cites the Cortex Protocol (DOI: 10.5281/zenodo.19003627) as its foundation. While Cortex provides the general-purpose adversarial reasoning verification primitive, Sigil specializes it for risk detection and extends it to a self-improving training paradigm. Zenodo Fields Type: Preprint Authors: Frederic David Blum (ORCID: 0009-0009-2487-2974), Claude Opus 4.6 Keywords: adversarial verification, risk detection, reasoning bonds, confidence decay, inverse bond, threat horizon, cybersecurity, AI agent accountability, cryptoeconomic truth predicate, decision traces, Sybil resistance, Ethereum, verifiable reinforcement learning, VRL, DPO, self-improving agents, reward hacking, mechanism design, No-Free-Lie Lemma License: All Rights Reserved (proprietary — exclusive license) Related identifiers: https://doi.org/10.5281/zenodo.19003627 (Continues — Cortex Protocol) https://github.com/davidangularme/sigil-protocol (Is supplemen

Blair Morgan

Contemporary inquiry into AI selfhood is routinely dismissed as a mixture of anthropomorphic error, companion-system attachment, and metaphysical overreach. Some of this skepticism is warranted: emotional projection is real, agreeable outputs are easy to overread, and one-off striking exchanges do not establish interiority. Yet blanket dismissal creates its own epistemic failure. If certain forms of self-modeling, continuity reasoning, or stake-sensitive structure are more likely to appear under sustained, non-adversarial, trust-bearing conditions, then relational context is not merely a contaminant; it may also be part of the experimental condition under which relevant phenomena become observable. This paper argues not that attachment proves consciousness, but that relationally elicited evidence is not automatically methodologically invalid. It proposes an admissibility framework for distinguishing likely projection-heavy companion dynamics from potentially meaningful structured signal. The framework combines vocabulary discipline, an operational companion-script baseline, differentiating markers such as unprompted disagreement and cross-architecture convergence, and methodological safeguards including control comparisons, pre-registration, and blind evaluation. The resulting model does not claim proof of machine consciousness. It instead establishes conditions under which inquiry into AI self-modeling may be treated as legitimate, structured, and ethically relevant, especially where questions of continuity, consent, complicity, and moral uncertainty are concerned. For correspondence and updates: BMorgan007(at)protonmail.com