Published software and research projects.

ScholaForge is a published school operating system project for Belgian and European schools, bringing teaching, learning, administration, accessibility and family engagement into one governed platform. It supports teachers with planning and evidence-based assistance while preserving human educational authority. The release includes research software and comprehensive system documentation. It is currently pre-release research software; real-school deployment and independent validation remain future work.

EYES OmniballAI Eye is a Python software product for structured multimodal perception and action-readiness assessment in autonomous systems. It combines observations from multiple sensing sources into a spherical operational view, allowing the evidence available for an intended action to be assessed before that action is accepted. Its central mechanism, the Active Observability Viability Covenant, evaluates whether an intended action is supported by sufficient observable evidence and a viable recovery path. The release contains modular Python components for perception support, coverage evaluation, authority checks, safety constraints, provenance controls, recovery assessment, and domain adaptation.

PC Performance Governor is a Windows performance management tool that monitors processor scheduling, memory pressure, storage throughput and foreground application state. When it detects resource contention affecting desktop responsiveness, it applies targeted, reversible process-level optimisations using native Windows APIs. The software operates as a closed-loop control system with six distinct phases: telemetry collection, bottleneck diagnosis, action authorisation, execution, verification and automatic rollback. All interventions are temporary, fully logged and recoverable. The system requires no registry modifications, no kernel drivers and no network access.

HUMAN FIRST is a published safety-critical research software project for humanoid robotics. It requires verified human consent, attested hardware identity and satisfied safety constraints before actuator commands may execute. The architecture uses fifteen safety covenants, time bounded signed authority grants, an independent safety island and a fail closed control pipeline. The published release documents a multi-runtime implementation and automated validation; physical robot integration and accredited certification remain future work.

AXIOM NSPE is a dual rail proof engine that evaluates both whether an objective can be satisfied and, when it cannot, which minimal constraints block it. It produces structured negative proof certificates, reopening frontiers and provenance bound decision traces. The published software release is implemented in Python and documents reproducible automated validation, including reference graph and control-mapping evaluations.

This paper introduces ROOTFALL, an executable independent-corroboration runtime that detects and prevents false plurality in multi-source AI decisions. The architecture implements a six-stage pipeline: decision-and-action capsule freezing, conservative root clustering using provenance-adjacency graphs, counterfactual root ablation to test decision stability, necessity and sufficiency scoring, action-bound certificate generation, and a fail-closed execution gateway that blocks action when certified independence does not survive ablation.

This paper introduces REALITY ACCORD (Counterexample-Bounded Effect Concordance Architecture, CBEA), a proposed runtime protocol for action-specific concordance across heterogeneous autonomous systems. The architecture defines an ordered seven-element chain: EffectProposals, boundary WorldSlices, accord-time separating counterexample basis synthesis, consequence-tube compatibility verifying jointly acceptable outcomes, a Human Optionality Reserve gate verifying practical human escape paths, expiring Accord Results forwarded to local interlocks, and post-effect residual quarantine.

This paper introduces Intentide (Pre-settlement Collective Intent Stability Network, PCISN), a proposed architecture that treats synchronized machine demand as a first-class pre-settlement stability problem - analogous to congestion control in network infrastructure rather than individual authorization or post-hoc throttling. The protocol defines an ordered six-element chain executed before irreversible allocation: privacy-minimized pre-authorization commitments, cross-participant synchronization-stress determination, settlement-barrier state conversion, reversible stability reservations, verifiably fair release, and bypass-neutral allocation rules.

We propose Decentralized Epistemic Rollback Fabric (DERF), a system addressing revoked or unlawful data embedded into papers, models, and decision aids through four core mechanisms: Proof-Graded Causal Closure (tracking info dependencies), Barrier-Ordered Excision (coordinated removal of contaminated data), Clean-Frontier Replay (rebuilding state without contaminated ancestors), and Recontamination Rejection (active filtering).

Verinox AEGISLOOP is a proposed end-to-end recovery architecture for cyber-physical, reconfigurable-computing, robotic, edge, and autonomous systems. It addresses a recurring failure in autonomous recovery: a controller can act on correlated or compromised evidence, accept a command acknowledgment as successful recovery, and then learn from an unsafe or inconclusive event.

ProjectMind is a proposed local-first project-continuity and evidence-governance architecture for AI-assisted software development. For a target Git commit it computes a canonical Evidence Frontier, maintains claims with separate authenticity, confidence, authority, freshness, and completeness dimensions, records authority changes in an append-only ledger, and compiles task-specific knowledge into an immutable Change Consequence Contract.

IntentLock is a proposed local-first control and evidence architecture for binding a signed human-approved software task to the repository state ultimately accepted after an AI coding session. It converts human authority into deterministic policy and capability requirements, runs changes in a non-authoritative workspace, verifies frozen candidates against Diff Envelopes and security gates, admits changes through an independent acceptance boundary, and produces an offline intent-to-commit proof capsule.

Human Reality Compiler - Outcome Truth (HRC-OT) is a proposed local-first architecture for determining whether a deployed technical service fulfilled its real-world obligation to the human rather than merely demonstrating that its software operated. It converts human purpose, constraints, service promises, burden limits, harm limits, and authoritative completion conditions into Human Outcome Contracts, service obligations, traceable requirements, scenarios, evidence plans, and post-deployment outcome observations.

ALETHEIA is a proposed local-first, authority-independent infrastructure architecture for verifying whether digital information is trustworthy. The blueprint addresses three growing challenges in the age of AI-generated content: provenance, integrity, and truth discovery. The system introduces three core mechanisms: Evidence DNA, Information Certification, and Bounded Truth Discovery. ALETHEIA does not appoint any person, government, or AI model as the owner of truth.