# Bayram Yüksel Eker — complete public site corpus > Canonical source: https://bayrameker.com. Snapshot: 3 September 2026. This file contains the full public text behind the portfolio, including claim boundaries. For the concise index use https://bayrameker.com/llms.txt. ## Identity and positioning Software and AI systems architect. Founder of Neura Parse Ltd. Independent quantum-AI researcher. My career began in state-scale software: national e-Government services, customs and legal-entity registries, and then carrier IPTV platforms delivered with ZTE at Netaş. Systems that carry legal or operational consequence teach a particular discipline, which is to keep the evidence attached to the result. I carried it into agentic AI, software-testing automation, quantum development workflows and hardware-backed research under uncertainty. It remains the through-line. Calibration and classical baselines stay beside the claim, execution context stays beside the result, and a person stays in the loop wherever the consequences are real. - Name variants: Bayram Eker | Bayram Yüksel Eker | Bayram Yuksel Eker | Bayram Y. Eker | bayrameker | B. Y. Eker - Role: AI Systems Architect · Quantum-AI Researcher - Location: Türkiye · Neura Parse Ltd (United Kingdom) - Email: bayram@neuraparse.com - ORCID: 0009-0003-0167-5763 - Company: NEURA PARSE LTD, company number 16547481 ## Research themes ### Decision inference under uncertainty Belief updating, multi-target data association and POMDP planning, where the quantum layer is a bounded primitive inside a classical loop and never the thing holding action authority. ### Evidence-first experimentation Calibration conditions, classical baselines, backend identity and operating envelopes kept attached to results, so that a reader can tell what a number means and where it stops meaning it. ### Governed agentic systems Agent orchestration in which drafting is cheap, consequential execution is human-approved, and the whole run stays inspectable after the fact. ## Capabilities ### AI systems - Agentic orchestration - LLM evaluation - RAG architecture - Human-approval design - Test automation ### Quantum - Qiskit - OpenQASM - QUBO / QAOA - POMDP belief inference - IBM Heron execution ### Engineering - Python - TypeScript - Java · Spring Boot - REST & microservices - Distributed systems ### Platform - Linux - Docker - Kubernetes - CI/CD - Reproducible research ## Identifiers - ORCID: 0009-0003-0167-5763 — https://orcid.org/0009-0003-0167-5763 - arXiv: 2603.00785 — https://arxiv.org/abs/2603.00785 - arXiv: 2607.06760 — https://arxiv.org/abs/2607.06760 - SSRN DOI: 10.2139/ssrn.6655058 — https://doi.org/10.2139/ssrn.6655058 - Zenodo: 10.5281/zenodo.19800503 — https://doi.org/10.5281/zenodo.19800503 - Zenodo: 10.5281/zenodo.19998452 — https://doi.org/10.5281/zenodo.19998452 - ISBN: 978-625-00-5878-7 - Companies House: 16547481 — https://find-and-update.company-information.service.gov.uk/company/16547481 ## Publications ### QANTIS: A Hardware-Validated Quantum Platform for POMDP Planning and Multi-Target Data Association - Identifier: arXiv:2603.00785 - Date: 28 February 2026 - Venue/status: arXiv; preprint - Authors: Bayram Yüksel Eker, Şuayb Ş. Arslan, Özgür Nazlı, Mustafa Serhat Demirgil, Furkan Deligöz - Author role: First author · platform lead - Links: arXiv: https://arxiv.org/abs/2603.00785 | Repository: https://github.com/neuraparse/qantis A modular quantum platform for belief update and multi-target data association under partial observability, validated through a 45-experiment campaign across three IBM Heron backends rather than in simulation alone. Research question: Can a quantum belief-update mechanism and its surrounding platform operate on real hardware, with results that survive calibration-aware scrutiny? Method: - Quantum belief update through Grover amplitude amplification and BIQAE - QUBO-based multi-target data association solved with FPC-QAOA - Composable error mitigation applied across the pipeline - 45 experiments executed across three IBM Heron backends - Calibration conditions, mitigation settings and backend identity retained per run Findings: - Hardware campaign: 45 experiments · 3 IBM Heron backends - Mechanism validation: Rare-observation probability amplified 0.179 → 0.907 - Posterior fidelity: Hellinger distance 0.0015 against exact Bayes - Framing: Controlled mechanism validation, not a classical replacement - Record: 31 pages, 4 figures, 12 tables. quant-ph, cross-listed cs.AI Claim boundaries: - Public preprint — not peer-reviewed journal acceptance. - No wall-clock quantum advantage over strong classical methods is claimed. - No claim of universal hardware scalability or production readiness. - No claim of end-to-end autonomy or real-time operational deployment. ### QANTIS: Hardware-Calibrated Sequential POMDP Belief Updates on IBM Heron - Identifier: arXiv:2607.06760 - Date: 9 July 2026 - Venue/status: arXiv; preprint - Authors: Bayram Yüksel Eker, Şuayb Ş. Arslan, Mustafa Serhat Demirgil - Author role: First author - Links: arXiv: https://arxiv.org/abs/2607.06760 | Repository: https://github.com/neuraparse/qantis A sequential follow-on study asking whether a calibrated quantum belief-update service can be reused repeatedly inside a classical decision loop, without cumulative noise making the posterior unusable. Research question: Does the belief-update primitive survive repeated use across a trajectory, or does error accumulate until the decision changes? Method: - Belief-update service embedded inside a classical planner — action authority stays classical - Primary campaigns over 8-step and 12-step decision loops - 20-step and 32-step controls to probe longer-horizon stability - Hardware-derived posteriors compared against exact Bayes at every decision point Findings: - Primary runs: 8-step and 12-step sequential campaigns - Controls: 20-step and 32-step longer-horizon probes - Decision consistency: Hardware and exact-Bayes posteriors chose the same action in every reported check - Result type: A decision-level operating envelope - Setting: Sequential Tiger POMDP horizon on IBM Heron Claim boundaries: - Public preprint — not peer-reviewed journal acceptance. - The quantum module does not replace classical planning, and it holds no autonomous action authority. - No claim of universal scaling, production certification or wall-clock advantage. - Results hold within the stated POMDP, hardware, calibration and experiment settings. ### The Quantum-Biological Intelligence Stack: A Layered Reference Architecture for Hybrid Intelligent Systems - Identifier: 10.2139/ssrn.6655058 - Date: 6 May 2026 - Venue/status: SSRN; preprint - Authors: Bayram Yüksel Eker - Author role: Sole author - Links: SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6655058 | DOI: https://doi.org/10.2139/ssrn.6655058 | Zenodo: https://doi.org/10.5281/zenodo.19800503 A nine-page layered reference architecture connecting quantum, biological, neuromorphic, agentic-AI, human-interface and governance layers, with maturity boundaries stated per layer rather than implied. Method: - Cross-literature synthesis rather than a single-discipline argument - Each layer annotated with what is demonstrated today and what is not - Companion deposit archived on Zenodo for citation stability Findings: - Length: 9 pages · posted 6 May 2026 - DOI: 10.2139/ssrn.6655058 - Deposit: Zenodo 10.5281/zenodo.19800503 - Type: Position architecture and technical synthesis Claim boundaries: - A synthesis and position architecture — not a claim of peer-reviewed discovery. - SSRN and Zenodo provide distribution and citation infrastructure, not endorsement. ## Book ### Quantum-Bio Intelligence: A New Mind Architecture for the Age of AI, Biology, and Quantum - Publisher: QBI Press - Year: 2026 - Pages: 520 - ISBN: 978-625-00-5878-7 - Registered: National Library of Türkiye, 27 April 2026 - Links: Amazon: https://www.amazon.com/dp/6250058788 | Zenodo: https://doi.org/10.5281/zenodo.19998452 | ResearchGate: https://www.researchgate.net/publication/404399097_Quantum-Bio_Intelligence_A_New_Mind_Architecture_for_the_Age_of_AI_Biology_and_Quantum | Launch article: https://bayramblog.medium.com/the-missing-half-of-ai-a-new-mind-architecture-for-the-age-of-ai-biology-and-quantum-94717212a2ac A 520-page paperback technical monograph arguing that the bottleneck in contemporary AI is architectural rather than computational. The retained publisher PDF is a 364-page layout snapshot of the same work. It connects quantum computation, biological information processing, neuromorphic systems, agentic AI, human interfaces and governance into a single vocabulary. Throughout, it is explicit about which layers are demonstrated today and which remain speculative. Themes: - Architecture over scale: The case that more parameters do not resolve uncertainty, verification or human authority. Those are structural questions, and they need structural answers. - Evidence tiers made explicit: The book separates what is demonstrated, what is plausible and what is speculation, rather than letting a confident tone flatten the difference. - Quantum layers, honestly bounded: What current quantum hardware can carry inside a larger system, and, more usefully, what it cannot yet carry. - Biological and neuromorphic processing: What biological information processing suggests about architecture, treated as design inspiration under scrutiny rather than as loose metaphor. - Agentic systems and approval: Orchestration, evaluation and the boundary where a system must stop and ask a person. - Governance and provenance: The evidence problem: how a system records what it did, so the record survives review. An independent technical synthesis, self-published through QBI Press. It is a position architecture rather than peer-reviewed research. Amazon, Zenodo, SSRN and ResearchGate provide distribution, archival and indexing infrastructure; none of them constitutes endorsement. ## Systems ### NODERIQ — Verifiable distributed intelligence for resilient autonomous systems The programme the rest of the work feeds into. It is a classical, edge-operable core built so that recommendation and authority stay separate. The system senses, represents its uncertainty rather than flattening it, shares only the evidence a constrained link can actually carry, and routes anything ambiguous to an accountable human with the supporting evidence attached. Hybrid and quantum pathways advance only by passing evidence gates. - Canonical page: https://bayrameker.com/systems#noderiq - Primary image: https://bayrameker.com/media/product/noderiq.webp — NODERIQ programme, distributed mission intelligence - Supporting image: https://bayrameker.com/media/product/noderiq-console.webp — Uncrewed assurance console showing decision evidence - Supporting image: https://bayrameker.com/media/product/noderiq-bvlos.webp — Beyond visual line of sight drone operations - Supporting image: https://bayrameker.com/media/product/noderiq-fleet.webp — Robotics fleet operations view Editions: - Defence — NODERIQ Defense: Multidomain autonomy for uncrewed platforms in contested environments: edge AI, secure mission orchestration, human-in-the-loop assurance and auditable decision evidence. Submitted to NATO DIANA’s Multidomain Autonomy of Uncrewed Systems challenge on 25 June 2026. - Enterprise — NODERIQ-CRN · Civil Resilient Operations: The civil surface of the same core, covering infrastructure inspection, emergency logistics, field service and connected mobility. Here the operating boundary is safety and continuity rather than contest. Specifications: - Consortium baseline: 3 companies · 30 months · 240 person-months · €2.10M eligible cost · TRL 4→6 BOUNDARY: NODERIQ is presented publicly as an applied research and productisation programme. It is not a released product, a certified capability or a deployed system. No quantum advantage is claimed; quantum methods remain conditional on passing the evidence gates. ### QFlow Studio — From a question to a workflow — and a reviewable record Quantum work fragments across notebooks, circuit editors, SDK source, provider consoles, run logs and review notes. That fragmentation is where reproducibility and handover quality break down. QFlow holds the objective, circuit, generated source, provider route, execution context and reviewer-safe evidence in one controlled record, rather than treating the circuit as an isolated artefact. - Canonical page: https://bayrameker.com/systems#qflow - Primary image: https://bayrameker.com/media/product/qflow-ui.webp — QFlow Studio workflow interface - Supporting image: https://bayrameker.com/media/product/qflow-canvas.webp — Studio canvas with the circuit graph - Supporting image: https://bayrameker.com/media/product/qflow-evidence.webp — Enterprise evidence review Editions: - Enterprise — QFlow Studio: The commercial platform: visual workflow design, provider routing, evidence packets and staged delivery for research, education and bounded enterprise use. (https://qflow.studio) - Open source — qmesh: Public preview of the quantum IR and provenance direction, with signed run manifests. Apache-2.0. (https://github.com/neuraparse/qmesh) Specifications: - Source surfaces: Qiskit, Cirq and OpenQASM, kept in sync - Authority: AI may draft; consequential execution stays user-approved - Pilot status: Detailed planning for a potential education pilot following documented product review - Controlled use: 30-person education record; not an active-user count BOUNDARY: What is claimed here is the operating and evidence model around existing quantum toolchains. It is not a new SDK, it does not cover every backend, and it makes no claim about quantum performance. ### NowFlow — Agentic workflows with authority built into the route Governed approval workflows that connect tools, integrations and delivery surfaces while keeping authority and change visible. Where an agent may act, where it must stop and ask, and what it actually did are all recorded rather than inferred afterwards. - Canonical page: https://bayrameker.com/systems#nowflow - Primary image: https://bayrameker.com/media/product/nowflow-ui.webp — NowFlow command interface Editions: - Enterprise — NowFlow: The commercial platform, with 188 workflow blocks and over 300 integrations. Approval gates are first-class routing rather than an afterthought. - Open source — NowFlow Community: Open workflow engineering and governed agentic patterns. Apache-2.0. (https://github.com/neuraparse/nowflow-community) Specifications: - Blocks: 188 workflow blocks - Integrations: 300+ ### NeuraOS — A signed software path from model package to working device Edge runtime for AI workloads across robotics and autonomous systems. It is the deployment path that NODERIQ’s edge-operable core needs, and it preserves the provenance of what shipped from end to end. - Canonical page: https://bayrameker.com/systems#neuraos - Primary image: https://bayrameker.com/media/product/neuralos-ui.webp — NeuraOS edge runtime interface Editions: - Enterprise — NeuraOS: The proprietary edge platform: a signed model-to-device path for robotics and autonomous platforms. - Public record — Public architecture: Public product, governance and assurance documentation. No source code or binary release is distributed. (https://github.com/neuraparse/neuraos) Specifications: - Base: Linux 6.18 LTS with PREEMPT_RT - Delivery: Immutable pins · SBOM/VEX · signed provenance · A/B updates ## Products and open-source work ### QFlow Studio — Quantum workflow and evidence layer - URL: https://qflow.studio - Provenance: product - Stack: TypeScript · Python - Licence: Commercial product Quantum work fragments across notebooks, circuit editors, SDK source, provider consoles, run logs, screenshots and review notes. That fragmentation is where reproducibility and handover quality go to die. QFlow keeps those elements as one operating record instead of treating the circuit as an isolated artefact. Stages: - Brief: Objective, owner, constraints - Circuit: Visual graph and operations - Source: Qiskit · Cirq · OpenQASM, kept in sync - Route: Provider fit, preflight and fallback - Run: Simulator or hardware execution context - Evidence: Result, trace and review state as one packet Contribution: - Defined the product problem, category and the operating record linking intent, design, source, route, execution and proof. - Set the roadmap and delivery boundaries: simulator-first, human-approved, provider-aware, evidence-led. - Specified source synchronisation across Qiskit, Cirq and OpenQASM, and the separation between provider credentials and share-safe evidence. - Designed pilot framing for research, education and bounded enterprise use, including acceptance criteria and reviewer output. BOUNDARY: The innovation claimed is the operating and evidence model around existing quantum toolchains — not a new SDK, not universal backend coverage, and not quantum performance advantage. AI may draft; consequential execution stays user-approved. ### QANTIS — Quantum Autonomous Navigation, Tracking and Intelligence System - URL: https://github.com/neuraparse/qantis - Provenance: open-source - Stack: Python - Licence: MIT - Stars at 3 September 2026: 6 - Forks at 3 September 2026: 0 The research software behind both QANTIS preprints: experiment design, Qiskit circuit construction, IBM Heron execution records, calibration-aware analysis and classical baselines, kept as one reproducible surface. The public repository carries an explanatory edition; the full experimental tooling is not represented as public production software. ### TaskNebula — Self-hosted project and agent coordination - URL: https://github.com/neuraparse/taskNebula - Provenance: open-source - Stack: TypeScript - Licence: MIT - Stars at 3 September 2026: 40 - Forks at 3 September 2026: 10 Coordination for teams and agents where human approval and inspectable operations are defaults rather than add-ons. Public records on 3 September 2026 showed 8,919 Docker pull events, 77 Docker tags, 28 GitHub releases and 69 commits attributed to Bayram Eker. ### qmesh — Quantum IR and provenance - URL: https://github.com/neuraparse/qmesh - Provenance: open-source - Stack: Python - Licence: Apache-2.0 - Stars at 3 September 2026: 6 - Forks at 3 September 2026: 0 An intermediate-representation and provenance direction across modalities, with signed run manifests. Public preview, experimental. ### NowFlow Community — Governed agentic workflow patterns - URL: https://github.com/neuraparse/nowflow-community - Provenance: open-source - Stack: TypeScript - Licence: Apache-2.0 - Stars at 3 September 2026: 10 - Forks at 3 September 2026: 4 Agent orchestration, tool integrations, approvals and inspectable execution, released for community use. ### NeuraOS — Governed edge intelligence for robotics and autonomous systems - URL: https://github.com/neuraparse/neuraos - Provenance: public-documentation - Stack: Public documentation - Licence: Proprietary · public documentation - Stars at 3 September 2026: 7 - Forks at 3 September 2026: 0 Public product architecture and assurance documentation for the proprietary edge runtime. The repository does not distribute source code or binaries. ### OrbIDE — Context-centric workspace for AI-assisted engineering - URL: https://github.com/neuraparse/orbide - Provenance: open-source - Stack: TypeScript · Rust - Licence: MIT - Stars at 3 September 2026: 0 - Forks at 3 September 2026: 0 An offline-first TypeScript and Tauri workspace for turning notes, decisions, tasks and sources into structured context that can be exported to AI tools. Public preview, active development. ### QMANN — Quantum memory-augmented networks - URL: https://github.com/neuraparse/QMANN - Provenance: open-source - Stack: Python - Licence: Open source - Stars at 3 September 2026: 3 - Forks at 3 September 2026: 0 Exploratory research on quantum memory-augmented neural architectures. ### NeuraBar — macOS utility - URL: https://github.com/neuraparse/NeuraBar - Provenance: open-source - Stack: Swift - Licence: Open source - Stars at 3 September 2026: 0 - Forks at 3 September 2026: 0 A small native macOS surface for the toolchain. ## Current work ### QFlow productisation and education pilot planning - Period: July 2026 — present - State: Planning - Status: Detailed pilot planning - Public visual: https://bayrameker.com/media/product/qflow-ui.webp — QFlow Studio workflow showing a reviewable quantum project record - QFlow Studio: https://qflow.studio A documented external product review moved QFlow into detailed planning for a potential education-focused pilot. A separate controlled education record covers a cohort of 30 people; that number is not presented as 30 active users. Working sequence: 1. Keep the brief, circuit, generated source, provider route, run context and evidence in one reviewable record. 2. Define the teaching scope, access controls, supervision model and acceptance criteria before any pilot starts. 3. Continue provider strategy and enablement work while keeping credentials and consequential execution user-controlled. BOUNDARY: This is product review and pilot planning. It does not establish procurement, deployment, a completed pilot, paid adoption or a university partnership agreement. ### NODERIQ consortium and research programme - Period: 2026 — present - State: Planning - Status: Partner gates and proposal development - Public visual: https://bayrameker.com/media/product/noderiq-console.webp — NODERIQ assurance console with decision evidence and human review - Programme architecture: https://neuraparse.com/research/noderiq/ Neura Parse is coordinating a planned three-company international research structure around governed edge intelligence, resilient networks and a robotics pilot. The working baseline is 30 months, 240 person-months, €2.10 million in eligible cost and a TRL 4 to 6 path. Working sequence: 1. Select the third pilot partner through capability, funding, IP, safety and delivery gates. 2. Lock work packages, partner effort, national funding routes and co-finance evidence. 3. Run evaluator crosswalk, claim review and submission QA before any formal proposal is sent. BOUNDARY: The duration, effort, budget and maturity figures are proposal baselines. No funding decision, executed consortium agreement, completed proof of concept or deployment is claimed. ### Governed enterprise AI delivery - Period: August 2026 — present - State: Delivery - Status: Confidentiality gate complete; service scope gated - Public visual: withheld because the underlying work includes confidential material. A private enterprise engagement is being run through a controlled sequence for operational data, deterministic analysis, AI-assisted explanation and accountable human decisions. The counterparty and private work package remain confidential. Working sequence: 1. Complete the mutual confidentiality gate before access, disclosure or technical work. 2. Lock the service framework, data responsibilities, security boundary and intellectual-property terms. 3. Authorize work through bounded work orders, then retain acceptance and decision evidence for each delivery. BOUNDARY: The mutual confidentiality step is complete. This page does not identify the counterparty, publish contract terms or imply that later service and deployment gates are already complete. ### Agentic software testing and enterprise AI - Period: 2025 — present - State: Delivery - Status: Project leadership and architecture - Public visual: withheld because the underlying work includes confidential material. Current professional work covers requirements, architecture, backlog decomposition, competitor analysis, evaluation and delivery controls for agentic software-testing and enterprise AI products. Working sequence: 1. Turn product goals into bounded requirements and architecture decisions. 2. Define evaluation, human-oversight and reliability controls before release decisions. 3. Coordinate team delivery, review evidence and resolve production constraints. BOUNDARY: Employer-confidential product names, roadmaps, customer information and internal performance records are not published here. ### TaskNebula public product maintenance - Period: 2025 — present - State: Public release - Status: Open-source release line - Public visual: https://bayrameker.com/media/product/tasknebula.webp — TaskNebula self-hosted agent coordination interface - GitHub: https://github.com/neuraparse/taskNebula - Docker Hub: https://hub.docker.com/r/neuraparse/tasknebula TaskNebula is the most-used public repository in the portfolio. On 3 September 2026 the public records showed 8,919 Docker pull events, 77 Docker tags, 28 GitHub releases, 40 stars, 10 forks and 69 commits attributed to Bayram Eker. Working sequence: 1. Maintain the self-hosted coordination surface and Docker release path. 2. Keep agent ownership, approval gates and inspectable operations as defaults. 3. Publish release and repository records so activity can be checked independently. BOUNDARY: Docker pulls are registry pull events. They are not unique users, installations, servers, customers or active deployments. ### Singapore and APAC market development - Period: December 2025 — present - State: Market development - Status: Discovery and partner qualification - Public visual: https://bayrameker.com/media/product/neuralos-ui.webp — NeuraOS edge runtime interface used as the public visual for APAC product discovery Short Singapore visits supported APAC market discovery for NowFlow and NeuraOS. A requested company overview was reviewed by a Singapore public-sector technology organisation, and the current work is to qualify independent business partners and real end-user problems. Working sequence: 1. Separate market discovery from customer, partner and deployment claims. 2. Qualify organisations for technical contribution, funding route and end-user access. 3. Turn only partner-approved problems into bounded work packages and measurable validation plans. BOUNDARY: Prior travel, correspondence and overview review do not establish procurement, a commercial partner, a paid engagement or a deployment. ## R&D and funding routes under evaluation ### UK–Singapore CR&D 2026 - Product: NODERIQ · NowFlow · NeuraOS - Status: Preferred preparation route - Official opportunity: https://apply-for-innovation-funding.service.gov.uk/competition/2526/overview/4a4b67b9-9b73-4816-8e21-bdfab83f5909 - Next gate: Secure an eligible Singapore business partner and a partner-approved industrial robotics problem before drafting a submission. - BOUNDARY: No application has been submitted and no partner has approved the concept. ### Quantum computing applications for large scale impact - Product: QFlow Studio - Status: Time-boxed go/no-go review - Official opportunity: https://apply-for-innovation-funding.service.gov.uk/competition/2525/overview/b1c3d8b5-575f-47c8-952d-22fc766fe802 - Next gate: Proceed only with a UK quantum collaborator, a real problem owner, one bounded use case and a fair classical-baseline plan. - BOUNDARY: QFlow is an assurance layer, not a substitute for the problem-specific quantum application required by the call. ### Investor Partnerships: ACT and Semiconductors - Product: NODERIQ - Status: Investor traction gate - Official opportunity: https://www.ukri.org/opportunity/investor-partnerships-act-and-semiconductors/ - Next gate: Test fit with selected investor partners before preparing an application or treating aligned investment as available. - BOUNDARY: Current investor discussions are not an invitation, term sheet, committed investment or grant decision. ### Knowledge Transfer Partnership 2026–2027 Round 3 - Product: Assured autonomous systems capability - Status: Conditional capacity route - Official opportunity: https://www.ukri.org/opportunity/knowledge-transfer-partnership-ktp-2026-2027-round-3/ - Next gate: Progress only with an eligible UK knowledge-base lead, at least two business FTEs and a fundable capability-building plan. - BOUNDARY: This is a route under evaluation, not an application, partnership or funding outcome. ## Career ### Founder & Technical Lead — Neura Parse Ltd - Period: Jun 2025 — present (current) - Location: London / Remote - Focus: Product architecture | Quantum-AI research | Governed delivery | Consortium development - Organisation: https://neuraparse.com - Set the company’s product and research architecture across governed AI workflows, quantum workflow and evidence infrastructure, and bounded autonomy research. - Lead QFlow Studio productisation and detailed planning for a potential education pilot following a documented product review. - Coordinate NODERIQ consortium development, partner gates, work packages and proposal assurance for a planned international research structure. - Run governed enterprise AI delivery and UK R&D pathway development through explicit confidentiality, partner, funding and evidence gates. - Lead QANTIS research framing, classical-baseline requirements, hardware-experiment coordination, evidence assembly and public research outputs. ### Project Lead / AI Architect — NETAŞ - Period: 2025 — present (current) - Location: Türkiye - Focus: Agentic testing | Enterprise AI | Evaluation | Delivery controls - Organisation: https://netas.com.tr - Lead agentic software-testing and enterprise AI product work: requirements, architecture, backlog decomposition, competitor analysis, evaluation and delivery controls. - Earlier developed an agentic test-scenario platform focused on software-testing workflows, human oversight, evaluation and reliability. ### Software Design / Senior Software Engineering — NETAŞ - Period: Oct 2022 — 2024 - Location: Türkiye - Focus: Distributed services | IPTV | Integration | Reliability engineering - Organisation: https://netas.com.tr - Backend, integration and distributed-service work in large operational environments using Java, Spring Boot, REST APIs, Linux and CI/CD. - Delivered IPTV platform work in joint projects with ZTE, covering carrier-grade video delivery, client integration and operational reliability at national scale. - Architecture reviews, production problem resolution, reliability engineering and cross-team delivery. ### Software Engineer — e-Government systems — Turkish public-sector programmes - Period: Before 2022 - Location: Türkiye - Focus: e-Government | Regulated data | Institutional integration | Auditability - Organisation: https://www.turkiye.gov.tr - Built and integrated services on Türkiye’s national e-Government platform, where correctness and auditability are not negotiable. - Customs and legal-entity registries: regulated data flows, institutional integrations and state-scale record handling. - This period established the working discipline. Systems that carry legal consequence teach you to keep the evidence attached to the result. ## Recognition and service ### Technical Reviewer, IEEE GLOBECOM 2026 - Issuer: IEEE Global Communications Conference · Selected Areas in Communications - Period: May – June 2026 Selected by the Cloud and Edge Computing and Networking chairs to conduct an independent technical peer review, and completed it within the deadline. The assignment came from the conference review system rather than from me, and its wording tied the request explicitly to my area of expertise: AI-enabled in-network computing, edge and cloud networking, and programmable systems. Evidence: - Track: SAC — Cloud/Edge Computing and Networking - Assigned: 31 May 2026 - Completed: 13 June 2026 - Conference: IEEE GLOBECOM 2026 BOUNDARY: This is an external technical evaluation role. It is not an IEEE award, a fellowship, an elected committee position or a paper acceptance. The review text itself is confidential and is not published. ### SAYZEK Certificate of Appreciation - Issuer: Presidency of Defence Industries (SSB) · Council of Higher Education (YÖK) - Period: 2024 – 2025 - Document image: https://bayrameker.com/media/certificates/sayzek-certificate.webp - Document PDF: https://bayrameker.com/documents/sayzek-certificate-of-appreciation.pdf Awarded a Certificate of Appreciation for guidance and contribution as an industry mentor in SAYZEK, Türkiye’s national defence-industry AI thesis programme. I mentored undergraduate and master’s students across several universities, working on AI-enabled imagery processing for UAV and armed-UAV platforms, and on radar-data analysis for AI-based ground localisation and target classification. Each student’s progress went through three interim reports that I reviewed and scored, and much of the work was helping them connect academic method to real engineering constraint. Evidence: - Document: Certificate of Appreciation - Dated: 10 July 2025 - Programme: SAYZEK-ATP 2024–2025 - Review load: 3 scored interim reports per student Programme context: - 421: applications from 119 universities and 58 departments - 356: accepted — 258 undergraduate, 98 graduate - 90: industry mentors drawn from 20 companies BOUNDARY: This is official recognition of a mentoring contribution within a national programme. It is not a government award, a national prize, a public appointment or a competitive placement. Project details stay non-operational: no restricted data or sensitive defence methods are disclosed here, and no student is identified. ### Independent Startup Assessment, Neura Parse - Issuer: Estonia Startup Committee - Period: August 2025 Neura Parse was reviewed by Estonia’s Startup Committee and considered to meet the startup definition under the Aliens Act. The assessment came from an external review process rather than from the company’s own positioning. Evidence: - Decided: August 2025 - Validity: Acceptance letter valid 5 years BOUNDARY: A company-level assessment of startup character. It is not a personal award, not a visa or residence decision, and not evidence of revenue, customers, investment or adoption. ## Certificates ### SAYZEK Certificate of Appreciation - Image: https://bayrameker.com/media/certificates/sayzek-certificate.webp - PDF: https://bayrameker.com/documents/sayzek-certificate-of-appreciation.pdf - Original Turkish document dated 10 July 2025. ### Machine Learning with Python - Issuer: IBM · Coursera - LinkedIn listing: May 2024 - Image: https://bayrameker.com/media/certificates/ibm-machine-learning-python.webp (certificate document) - Credential ID: 4AK9G3DPL462 - Public record date: 6 June 2024 - Verification: https://www.coursera.org/account/accomplishments/verify/4AK9G3DPL462 Course certificate covering supervised and unsupervised learning, regression, classification, clustering, dimensionality reduction, Python and scikit-learn. ### Supervised Machine Learning: Regression and Classification - Issuer: Stanford Online · DeepLearning.AI · Coursera - LinkedIn listing: April 2024 - Image: https://bayrameker.com/media/certificates/stanford-supervised-machine-learning.webp (certificate document) - Credential ID: XQDZZFMRGZYC - Public record date: 5 June 2024 - Verification: https://www.coursera.org/account/accomplishments/verify/XQDZZFMRGZYC Course certificate covering supervised prediction and binary classification with NumPy and scikit-learn, including linear and logistic regression. ### Machine Learning with Python - Issuer: Coursera - LinkedIn listing: March 2024 - Image: https://bayrameker.com/media/certificates/linkedin-listed-credential.svg (LinkedIn record visual) - No public issuer-hosted document was available at capture time. Listed on LinkedIn as a distinct Coursera credential. The public profile exposes no credential ID or independently accessible certificate image for this entry. ### Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning - Issuer: DeepLearning.AI · Coursera - LinkedIn listing: February 2024 - Image: https://bayrameker.com/media/certificates/deeplearningai-tensorflow.webp (certificate document) - Credential ID: 9RZTXX6C77AY - Public record date: 7 June 2024 - Verification: https://www.coursera.org/account/accomplishments/verify/9RZTXX6C77AY Course certificate covering TensorFlow fundamentals, basic neural networks, computer-vision training and convolutional networks. ### Building Generative AI Applications with Gradio! - Issuer: DeepLearning.AI · Hugging Face - LinkedIn listing: January 2024 - Image: https://bayrameker.com/media/certificates/deeplearningai-gradio.webp (public completion record) - Credential ID: c2abfd17-6e48-4d55-ac42-4620e1d86fd0 - Public record date: 6 June 2024 - Verification: https://www.deeplearning.ai/accomplishments/c2abfd17-6e48-4d55-ac42-4620e1d86fd0 Completion record covering chatbots, diffusion models, generative-AI applications and generative models; the public record notes a passed graded quiz. ### LangChain for LLM Application Development! - Issuer: DeepLearning.AI · LangChain - LinkedIn listing: January 2024 - Image: https://bayrameker.com/media/certificates/deeplearningai-langchain.webp (public completion record) - Credential ID: a90ebbe8-9ecd-47e4-b29b-007be743337c - Public record date: 6 June 2024 - Verification: https://www.deeplearning.ai/accomplishments/a90ebbe8-9ecd-47e4-b29b-007be743337c Completion record covering AI frameworks, agents, chatbots, generative models, prompt engineering and RAG; the public record notes a passed graded quiz. ### Rest API (Intermediate) Certificate - Issuer: HackerRank - LinkedIn listing: July 2022 - Image: https://bayrameker.com/media/certificates/linkedin-listed-credential.svg (LinkedIn record visual) - No public issuer-hosted document was available at capture time. Intermediate REST API skills certificate listed on the public LinkedIn profile. No public credential ID or certificate document is exposed there. ### SQL (Advanced) Certificate - Issuer: HackerRank - LinkedIn listing: July 2022 - Image: https://bayrameker.com/media/certificates/linkedin-listed-credential.svg (LinkedIn record visual) - No public issuer-hosted document was available at capture time. Advanced SQL skills certificate listed on the public LinkedIn profile. No public credential ID or certificate document is exposed there. ### Traditional Entrepreneur Certificate - Issuer: KOSGEB — Small and Medium Enterprises Development Organisation of Türkiye - LinkedIn listing: July 2020 - Image: https://bayrameker.com/media/certificates/linkedin-listed-credential.svg (LinkedIn record visual) - Credential ID: KSB012020U149G75528E - No public issuer-hosted document was available at capture time. Traditional entrepreneurship certificate listed on the public LinkedIn profile with the credential identifier reproduced here. ### Advanced Entrepreneur Certificate - Issuer: KOSGEB — Small and Medium Enterprises Development Organisation of Türkiye - LinkedIn listing: July 2020 - Image: https://bayrameker.com/media/certificates/linkedin-listed-credential.svg (LinkedIn record visual) - Credential ID: KSB012020U149G75528E - No public issuer-hosted document was available at capture time. Advanced entrepreneurship certificate listed on the public LinkedIn profile with the credential identifier reproduced here. ## Consortia, competitions and programmes ### Noderiq - Organisation: CELTIC-NEXT · EUREKA Cluster - Period: 2026 — - Type: Consortium - Role: Consortium lead and UK coordinator Coordinating a planned three-company international research structure for NODERIQ under the CELTIC-NEXT cluster. The working package covers governed edge intelligence, resilient networks and a robotics pilot over 30 months, 240 person-months and a €2.10 million eligible-cost baseline, with partner and funding gates still open. BOUNDARY: Proposal development and consortium coordination. No executed consortium agreement, funding decision, completed proof of concept or deployment is claimed. ### NATO DIANA Challenge - Organisation: NATO Defence Innovation Accelerator for the North Atlantic - Period: 2026 - Type: Competition - Role: Applicant team Neura Parse entered the NATO DIANA challenge track with dual-use deep-technology work built on the same quantum and governed-autonomy foundations as QANTIS. BOUNDARY: Participation in the challenge process. No selection, award or accelerator place is claimed. ### SSB Quantum Algorithms Competition - Organisation: Presidency of Defence Industries (SSB) · Türkiye Quantum Platform - Period: 2026 - Type: Competition - Role: Team lead — QANTIS Nexus Entered Türkiye’s national quantum algorithms competition with the QANTIS Nexus team, applying the belief-update and data-association work from the QANTIS research programme. BOUNDARY: Team participation. No finalist, placement or winner status is claimed unless a result is later issued. ### IBM Partner Plus - Organisation: IBM - Period: Current - Type: Partner programme - Role: Partner (Neura Parse Ltd) Neura Parse participates in IBM’s partner programme, which is relevant to the QANTIS work: the hardware campaigns run on IBM Heron processors. BOUNDARY: Programme membership. Not a sponsorship, endorsement or validation of any technical claim. ### NVIDIA Inception - Organisation: NVIDIA - Period: Nov 2025 – Aug 2026 - Type: Partner programme - Role: Programme participant (Neura Parse Ltd) Neura Parse used NVIDIA Inception technical resources and training while developing NowFlow and NeuraOS, including edge-device tests for local AI execution. Participation ended in August 2026. BOUNDARY: Past programme participation. Not a sponsorship, technical validation or current membership. ### Europe’s Top 10 Startup Leaders - Organisation: European startup ecosystem - Period: 2026 - Type: Nomination - Role: Nominee Nominated among candidates for a European top-ten startup leadership listing. BOUNDARY: A nomination, not a placement or award. ## Company record - Legal name: NEURA PARSE LTD - Number: 16547481 - Status: Active private limited company - Incorporated: 27 June 2025 - Registered office: London, United Kingdom - Activities: 62012 — Business and domestic software development | 63110 — Data processing, hosting and related activities - Role: Active Director, appointed 27 June 2025 - Control: Person with significant control — 75%+ shares and voting rights - Official register: https://find-and-update.company-information.service.gov.uk/company/16547481 BOUNDARY: The public register establishes a direct, controlling and executive connection to a UK digital-technology business. It does not evidence revenue, trading performance, filed accounts, customer adoption or investment. ## Questions and direct answers ### Who is Bayram Eker? Bayram Yüksel Eker is a software and AI systems architect, an independent quantum-AI researcher, and the founder of Neura Parse Ltd, a UK deep-technology company registered in England and Wales under company number 16547481. He is first author of two 2026 quantum research preprints run on IBM Heron hardware, arXiv:2603.00785 and arXiv:2607.06760, author of the 520-page paperback monograph Quantum-Bio Intelligence, a completed technical reviewer for IEEE GLOBECOM 2026, and a certified industry mentor in Türkiye’s national SAYZEK Academic Thesis Programme. He leads the consortium for the NODERIQ proposal under the CELTIC-NEXT cluster. His ORCID is 0009-0003-0167-5763. ### What does Bayram Eker work on? The current portfolio spans hardware-calibrated decision inference in QANTIS, governed autonomous systems in NODERIQ, quantum workflow evidence in QFlow Studio, agentic software testing, governed enterprise AI delivery and public product maintenance through TaskNebula. The through-line across them is keeping the evidence attached to the decision: calibration conditions next to the result, execution context next to the run, and an accountable person in the loop wherever the consequences are real. ### Is Bayram Eker a quantum computing researcher? Yes, with a specific and narrow focus. He works on whether a quantum belief-update primitive can be trusted inside a classical decision loop on hardware that exists today, rather than on quantum algorithms in the abstract. The work is hardware-first. The foundational study ran 45 experiments across three IBM Heron backends. The sequential follow-on tested 8-step and 12-step decision trajectories with 20-step and 32-step controls. Both papers are public preprints on arXiv and neither claims quantum advantage over strong classical methods. ### How do I contact Bayram Eker? By email at bayram@neuraparse.com. He responds to research collaboration, technical review, speaking and product enquiries. Public profiles: ORCID 0009-0003-0167-5763, GitHub at github.com/neuraparse, and the company site at neuraparse.com. ### What is QANTIS? QANTIS stands for Quantum Autonomous Navigation, Tracking and Intelligence System. It is an applied research programme on hardware-calibrated decision inference under partial observability, with two first-author preprints published in 2026. arXiv:2603.00785, submitted 28 February 2026, runs to 31 pages with 4 figures and 12 tables. It integrates quantum belief update through Grover amplitude amplification and BIQAE, QUBO-based data association via FPC-QAOA, and composable error mitigation, validated across a 45-experiment campaign on three IBM Heron backends. In one reported Tiger belief-oracle case it amplified a rare observation probability from 0.179 to 0.907 while preserving the posterior at a Hellinger distance of 0.0015 from exact Bayes. arXiv:2607.06760 asked the harder follow-on question: whether the calibrated belief-update service survives repeated use inside a classical decision loop. Hardware-derived and exact-Bayes posteriors selected the same immediate action at every reported decision point. The public repository is at github.com/neuraparse/qantis under the MIT licence. ### Are the QANTIS papers peer-reviewed? No. Both are public preprints on arXiv. They have not been through peer review, and the site says so on every page that mentions them. A preprint is a genuine research output with a permanent identifier and a public timestamp, and it is not a reviewed one. Both halves of that statement are accurate and both belong in any description of the work. ### What is Quantum-Bio Intelligence? A 520-page paperback technical monograph published through QBI Press in 2026, ISBN 978-625-00-5878-7, registered with the National Library of Türkiye on 27 April 2026. The retained publisher PDF is a 364-page layout snapshot of the same work. It argues that the bottleneck in contemporary AI is architectural rather than computational. It connects quantum computation, biological information processing, neuromorphic systems, agentic AI, human interfaces and governance into one vocabulary, and it separates what is demonstrated from what is plausible from what is speculation throughout. It is self-published and it is a position architecture rather than peer-reviewed research. A companion paper, The Quantum-Biological Intelligence Stack, is on SSRN with DOI 10.2139/ssrn.6655058. ### What is NODERIQ? NODERIQ is an applied research and productisation programme on verifiable distributed intelligence for resilient autonomous systems. Its central commitment is that a system may recommend while a person authorises, and that those two operations stay architecturally separate. The operating loop is sense, qualify, share, coordinate, verify, escalate. Qualify is the step most architectures skip, because it represents uncertainty rather than resolving it prematurely. Four evidence gates decide whether alternative compute methods advance: G0 define, G1 simulate against a strong classical reference, G2 preserve information and uncertainty through hardware execution, G3 qualify on decision value after accuracy, latency, reliability and cost. NODERIQ is not a released product, a certified capability or a deployed system. ### What is the difference between NODERIQ Defense and NODERIQ-CRN? They are two operating surfaces on the same classical, edge-operable core. NODERIQ Defense addresses multidomain autonomy for uncrewed platforms in contested environments: edge AI, secure mission orchestration, human-in-the-loop assurance and auditable decision evidence. It was submitted to NATO DIANA’s Multidomain Autonomy of Uncrewed Systems challenge on 25 June 2026. NODERIQ-CRN covers civil resilient operations: infrastructure inspection, emergency logistics, field service and connected mobility. The engineering is largely shared because the underlying problem is shared. What differs is the operating boundary, which is contest in one case and safety and continuity in the other. ### What is QFlow Studio? A quantum workflow and evidence layer that holds the objective, circuit, generated source, provider route, execution context and reviewer-safe evidence in one controlled record. Generated source stays synchronised across Qiskit, Cirq and OpenQASM, so the code on screen is the code that ran. AI may draft; consequential execution stays user-approved. Following a documented product review, QFlow moved into detailed planning for a potential education-focused pilot. A separate 30-person controlled education record is not presented as an active-user count, procurement or completed deployment. What is claimed is the operating and evidence model around existing quantum toolchains. It is not a new SDK, it does not cover every backend, and it makes no claim about quantum performance. ### What is Neura Parse? Neura Parse Ltd is a UK deep-technology company, registered in England and Wales under company number 16547481 and incorporated on 27 June 2025. Bayram Yüksel Eker is founder, active director and person with significant control. It builds four systems: NODERIQ for verifiable distributed intelligence, QFlow Studio for quantum workflow and evidence, NowFlow for governed agentic workflows with 188 blocks and over 300 integrations, and NeuraOS as a signed model-to-device edge runtime on Linux 6.18 LTS with PREEMPT_RT. The public repositories and product-architecture records are at github.com/neuraparse. The NeuraOS repository publishes documentation, not source code or binaries. ### Is Neura Parse an IBM or NVIDIA partner? Neura Parse participates in IBM’s Partner Plus programme. It participated in NVIDIA Inception from November 2025 through August 2026 while developing NowFlow and NeuraOS and running edge-device tests for local AI execution. One is a current programme membership and the other is completed programme participation. Neither constitutes sponsorship or validation of any technical claim. ### What is the Noderiq CELTIC-NEXT consortium? Bayram Yüksel Eker leads and coordinates the planned three-company international structure for the NODERIQ proposal under CELTIC-NEXT, the EUREKA cluster for next-generation communications. The working baseline is 30 months, 240 person-months, €2.10 million in eligible cost and a TRL 4 to 6 path. These are proposal baselines. Partner gates, work packages, national funding routes and co-finance evidence are still being developed; no executed consortium agreement, funding outcome, proof of concept or deployment is claimed. ### What does QANTIS stand for? Quantum Autonomous Navigation, Tracking and Intelligence System. The name describes the application domain rather than the method. The problem is autonomous navigation under uncertainty, which means solving partially observable Markov decision processes for planning and assigning sensor measurements to tracked targets, a task known as multi-target data association. ### Which quantum methods does QANTIS actually use? The foundational paper integrates three things. Quantum belief update through Grover amplitude amplification and BIQAE. QUBO-based multi-target data association solved with FPC-QAOA. Composable error mitigation applied across the pipeline. All three are evaluated against classical references rather than in isolation, and the paper characterises practical operating boundaries for current superconducting hardware rather than claiming wall-clock advantage. ### Where can I find the QANTIS code? The public repository is github.com/neuraparse/qantis, released under the MIT licence. What is public is the explanatory edition rather than the full experimental tooling. That distinction is stated on the site rather than left for a reader to discover. ### What is the QANTIS Nexus team? The team entered Türkiye’s national quantum algorithms competition, run by the Presidency of Defence Industries with the Türkiye Quantum Platform, applying the belief-update and data-association work from the QANTIS research programme. This is team participation. No finalist placement, ranking or winner status is claimed unless a result is later issued. ### Where is Neura Parse based? Neura Parse Ltd is registered in England and Wales, company number 16547481, incorporated 27 June 2025. The arXiv record for the QANTIS foundational study lists the author affiliation as Neura Parse Ltd., London, United Kingdom. Work is conducted between the United Kingdom and Türkiye. ### What is TaskNebula? An open-source, self-hosted coordination platform for teams working alongside AI agents, released under the MIT licence at github.com/neuraparse/taskNebula. It is the most adopted of the public repositories. On 3 September 2026 its public records showed 8,919 Docker pull events, 77 Docker tags, 28 GitHub releases, 40 stars, 10 forks and 69 commits attributed to Bayram Eker. Pull events are not unique users, installations, servers or customers. Agent ownership, approval gates and Docker-first operations are defaults rather than configuration. The reasoning behind releasing it openly is set out in a March 2026 article. ### What are NowFlow and NeuraOS? NowFlow is a governed agentic workflow platform with 188 workflow blocks and over 300 integrations, where approval gates are first-class routing rather than an afterthought. It has an open-source counterpart, NowFlow Community, under Apache-2.0. NeuraOS is a proprietary edge platform that provides a signed path from a model package to a working device, built on Linux 6.18 LTS with PREEMPT_RT. Its public repository documents the product, governance and assurance architecture but distributes no source code or binaries. ### Is Bayram Eker available for collaboration or advisory work? Yes, for research collaboration, technical review, speaking and product enquiries. Email bayram@neuraparse.com. Areas where the work is most directly relevant: hybrid quantum and classical decision systems, evaluation design for autonomous systems, evidence and provenance infrastructure, and governed agentic architectures where human approval has to be structural rather than procedural. ### What has Bayram Eker built before the quantum work? More than six years in engineering, beginning with systems that carry legal consequence. He built and integrated services on Türkiye’s national e-Government platform, and worked on customs and legal-entity registries where regulated data flows and state-scale record handling are the requirement. At Netaş he delivered IPTV platform work in joint projects with ZTE, covering carrier-grade video delivery, client integration and operational reliability at national scale, alongside backend and distributed-service engineering in Java and Spring Boot. He later led agentic software-testing and enterprise AI product work at Netaş before founding Neura Parse in June 2025. ### What is the SAYZEK programme and what was Bayram Eker’s role? SAYZEK is the Academic Thesis Programme run jointly by Türkiye’s Presidency of Defence Industries and the Council of Higher Education. The 2024 to 2025 cycle took 421 applications from 119 universities and 58 departments, accepted 356 students, and assigned 90 industry mentors from 20 companies. Bayram Eker received a Certificate of Appreciation for valuable guidance and dedicated contributions as an industry mentor. He worked with undergraduate and master’s students on AI-enabled imagery processing for UAV and armed-UAV platforms and on radar-data analysis for AI-based ground localisation and target classification. Each student’s progress went through three interim reports that mentors reviewed and scored. The certificate is dated 10 July 2025. It is recognition of a mentoring contribution within a national programme. It is not a government award, a national prize or a competitive placement. ### What is Bayram Eker working on now? Current work includes detailed QFlow education-pilot planning, NODERIQ consortium and proposal development, a confidentiality-gated enterprise AI delivery, agentic software-testing architecture, TaskNebula public maintenance and Singapore/APAC partner discovery. Four UK R&D routes are also being evaluated: UK–Singapore CR&D, the quantum-computing applications CR&D call, Investor Partnerships for advanced connectivity technologies, and a Knowledge Transfer Partnership. They are preparation routes only; no application, partner approval, funding or deployment is implied. The Current work page records the next gate and the non-claim boundary for every item. ## External records and appearances ### QANTIS foundational study - Where: arXiv - URL: https://arxiv.org/abs/2603.00785 - Checked: 22 August 2026 arXiv:2603.00785, submitted 28 February 2026. 31 pages, 4 figures, 12 tables. Primary quant-ph, cross-listed cs.AI. Author affiliation recorded as Neura Parse Ltd., London, United Kingdom. ### QANTIS sequential study - Where: arXiv - URL: https://arxiv.org/abs/2607.06760 - Checked: 22 August 2026 arXiv:2607.06760, July 2026. Asks whether the belief-update service can be reused across a sequential Tiger POMDP horizon on IBM Heron without corrupting the planner-facing posterior. ### The Quantum-Biological Intelligence Stack - Where: SSRN - URL: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6655058 - Checked: 22 August 2026 Nine-page layered reference architecture, posted 6 May 2026. DOI 10.2139/ssrn.6655058. ### Archival deposits - Where: Zenodo - URL: https://doi.org/10.5281/zenodo.19998452 - Checked: 22 August 2026 Two deposits under 10.5281/zenodo.19800503 and 10.5281/zenodo.19998452, operated by CERN. ### Quantum-Bio Intelligence - Where: Amazon - URL: https://www.amazon.com/Quantum-Bio-Intelligence-Architecture-Biology-Quantum/dp/6250058788 - Checked: 22 August 2026 Paperback and Kindle. ISBN 9786250058787. Listed under Bayram Yüksel Eker. ### Book record - Where: ResearchGate - URL: https://www.researchgate.net/profile/Bayram-Eker - Checked: 22 August 2026 Indexed record for the monograph with its Zenodo identifier. ### Researcher identifier - Where: ORCID - URL: https://orcid.org/0009-0003-0167-5763 - Checked: 22 August 2026 0009-0003-0167-5763. The persistent identifier that ties the outputs to one person. ### Neura Parse Ltd - Where: Companies House - URL: https://find-and-update.company-information.service.gov.uk/company/16547481 - Checked: 22 August 2026 Company 16547481, incorporated 27 June 2025. Active director and person with significant control. ### neuraparse - Where: GitHub - URL: https://github.com/neuraparse - Checked: 3 September 2026 Nine public repositories. The current maintained set includes QANTIS under MIT, TaskNebula, qmesh, NowFlow Community, OrbIDE and the public NeuraOS architecture record. ### TaskNebula release record - Where: Docker Hub - URL: https://hub.docker.com/r/neuraparse/tasknebula - Checked: 3 September 2026 8,919 pull events and 77 tags. Pull events are not unique users, installations, servers or customers. ### Neura-parse organisation - Where: Hugging Face - URL: https://huggingface.co/Neura-parse - Checked: 22 August 2026 Public quantum and quantum-AI dataset portfolio with provenance fields on the dataset cards. ### Why We Open-Sourced TaskNebula - Where: Medium - URL: https://bayramblog.medium.com/why-we-open-sourced-tasknebula-e64fc325f85c - Checked: 22 August 2026 March 2026. On the reasoning behind releasing the coordination platform under MIT. ### The Missing Half of AI - Where: Medium - URL: https://bayramblog.medium.com/the-missing-half-of-ai-a-new-mind-architecture-for-the-age-of-ai-biology-and-quantum-94717212a2ac - Checked: 22 August 2026 The launch article for Quantum-Bio Intelligence. ### QANTIS foundational study - Where: Paper Reading Club - URL: http://paperreading.club/page?id=381035 - Checked: 22 August 2026 Third-party aggregator entry for the paper, independent of the author. ## Writing — full articles # Decision models: when the answer is a probability - Canonical URL: https://bayrameker.com/blog/decision-models/ - Published: 2026-09-20 - Topic: Systems - Summary: TypeSafe’s Jev puts typed decisions inside software. The interesting question is how probabilities become actions that code can govern. - Image: https://bayrameker.com/media/blog/decision-model.webp — Program state enters a decision model; a typed result and its uncertainty pass to a policy in code before an action is taken. A support ticket arrives: “I was charged twice this month.” The application does not need an essay about billing. It needs a destination, an indication of uncertainty and a rule for what happens next. Much of production software has this shape. A message needs a category. A workflow needs a branch. An agent's proposed action needs a check. Language helps interpret the input, but the output often belongs in ordinary control flow. TypeSafe AI's Jev, [introduced on 15 September 2026](https://typesafe.ai/blog/introducing-system-one-models-and-jev), takes that observation as its starting point. The company calls it a System One Model: a model built to return typed probabilistic decisions rather than prose. Five days after the launch, it is too early for a settled verdict on the product. The interface is already worth examining. ## Start with the decision the software needs In TypeSafe's [API design](https://docs.typesafe.ai/introduction), a request contains the current state and one or more typed questions. The result belongs to a declared shape, so the application can branch on it directly. Questions in the same call are evaluated independently against the shared state. The three primitives cover different kinds of judgement.
Three questions, three output shapesTypeSafe primitives
Choice
Which option fits?Choose from declared categories, such as billing, technical or sales.choice · probabilities · confidence
Score
Where does it sit on this rubric?Evaluate against defined levels, such as low, medium or high urgency.score · probabilities · confidence
Noul
Is this proposition true?Estimate the probability of yes, such as whether the customer requests a refund.noul: a value from 0 to 1
Noul returns a proposition probability without a separate confidence field. Choice and Score return distributions and a confidence summary.
The design encourages narrow questions. Instead of asking whether a request should be handled automatically, ask what it concerns, whether information is missing and whether it needs specialist attention. Then make the combination explicit in code. That is the architectural attraction for me. The judgement remains learned; the structure of the workflow becomes something a team can read, test and change. ## A valid shape is only the first guarantee It would be misleading to describe this as the first way to obtain dependable JSON from a model. Modern structured-output systems can constrain generation to a supported schema during decoding. OpenAI's [Structured Outputs](https://openai.com/index/introducing-structured-outputs-in-the-api/) is one example. The alternative is no longer simply free-form text followed by a fragile parser. TypeSafe's distinction is its decision-oriented model and interface: fixed output types, probability distributions and a design focused on short judgements. Whether that produces better accuracy, latency or cost on a particular workload is an empirical question. The company's schema guarantee also needs a precise reading. Returning one of `billing`, `technical` or `sales` prevents an undeclared fourth category. It does not prevent a technical question from being routed to billing. A well-formed answer can still be the wrong answer. That is a useful boundary, not a reason to dismiss the idea. Software can handle a known kind of mistake more deliberately than an unexpected output format. It still needs a way to detect uncertainty and measure errors. ## Probability and confidence are different signals Suppose a model estimates `P(refund_requested) = 0.03`. That is a strong answer of “probably not”, not a statement that the model has almost no confidence in its answer. Around 0.5, the same binary question is much less settled. Choice and Score expose a different signal. Their `confidence` field summarises how concentrated the returned distribution is. TypeSafe's [confidence documentation](https://docs.typesafe.ai/confidence) describes it as a statistic derived from that distribution. It should not automatically be read as the probability that the chosen answer is correct. For a ticket router, confidence might determine whether to route immediately, gather more information or ask a person. For a binary risk question, a high probability might instead mean stop. The direction of the action comes from the meaning of the question.
One score, three possible routesIllustrative ticket-routing policy
  1. c < 0.70
    Send to a reviewerThe answer is too uncertain for this route.
  2. 0.70 ≤ c < 0.98
    Gather another signalCheck account context or ask a clarifying question.
  3. c ≥ 0.98
    Route to the chosen queueProceed only if the route is allowed by application policy.
Here c means Choice confidence. These thresholds are examples, not measured recommendations. Routing a ticket does not authorise a refund or another account change.
The numbers need testing on representative data. An application's tolerance for a wrong queue is different from its tolerance for an incorrect account action. Even a very confident answer must remain subject to access controls and any required approval. ## Calibration has to survive your data Calibration is a relationship between predictions and observed outcomes. Among comparable cases assigned a probability near 0.8, the event should occur roughly eight times out of ten. A single answer cannot demonstrate that relationship. For my own evaluation, I would start with labelled examples from the intended workflow, separate the data used to set thresholds from the data used to assess them, and inspect the errors in each important slice. A good overall number can hide a weak result for one language, one customer group or one unfamiliar request type. I would also measure the operating trade-off: how many cases the system handles automatically, how many it sends to review and how many errors remain at that setting. Sending everything to a person can look safe while defeating the purpose of the automation. This is close to the discipline behind the [QANTIS research](/research): specify the reference, the conditions and the boundary of the result. “Calibrated” is most useful when followed by a dataset, a method and an account of where the claim stops holding. ## Let code own the branch A small example makes the separation easier to see. The following is illustrative application logic, not a TypeSafe SDK example. Its input is a validated Choice result, and its confidence thresholds are the same ones used in the diagram. ```python def route_ticket(ticket, decision, policy): if decision.confidence < 0.70: return send_to_reviewer(ticket) if decision.confidence < 0.98: return request_more_context(ticket) if not policy.allows_route(ticket, decision.choice): return send_to_reviewer(ticket) return assign_queue(ticket, decision.choice) ``` The model supplies the category and uncertainty. Code owns the conditions, the permitted destinations and the fallback. Changing a threshold becomes a versioned policy change that can be evaluated, reviewed and rolled back. This connects directly to the [harness engineering](/blog/harness-engineering/) argument. A narrow model call can help interpret an input or flag a concern. It should not become the sole authority on whether an operation is permitted.
A judgement inside a governed workflowModel output meets application policy
Incoming requestMessage, account context and proposed action
IntentChoice
UrgencyScore
Missing information?Noul
Policy in codeCheck uncertainty, access, action scope and required approval.
ProceedAll required checks pass.
Review or stopA check fails or evidence is incomplete.
Record the inputs, model result, policy version and outcome.
The model answers bounded questions. Deterministic checks retain authority. Both branches produce a record that can be inspected later.
In [NowFlow](/systems#nowflow), an approval gate is an explicit route. In [NODERIQ](/systems#noderiq), evidence must reach the person who decides. In QANTIS, [the planner stays classical](/blog/why-the-planner-stays-classical/) while a separate component updates a belief. These are different implementations of a useful boundary: a probabilistic judgement can inform an action without owning the authority to take it. ## Read the launch numbers as a starting point TypeSafe's launch post reports large speed and cost advantages on its workflow evaluations. It also says those workflows were created by its own model-capabilities team, may reflect bias and sit toward the high end of expected gains. Those caveats belong beside the claims. For a deployment decision, I would compare the complete workflow: task accuracy, latency across the distribution, total cost, review rate and the effects of retries or extra checks. A fast model call is valuable; the application's useful output is the end of the measurement. As of 20 September 2026, this is an early-access product with a short public history. That leaves substantial questions about performance across domains, distribution shifts and sustained production use. I would treat the launch as an invitation to evaluate a new component. ## The useful boundary is between judgement and authority Generative models still have a role in writing, planning and working through open-ended problems. Decision models offer a different interface for the narrower judgements that sit between those tasks. The promising architecture combines those abilities with explicit control. A model interprets the messy input. A typed result exposes the judgement. Code applies policy. The environment provides a result that can be checked. That is the part of the idea I expect to matter beyond any one product: make uncertainty visible enough to work with, and keep the decision to act somewhere people can inspect. --- # Harness engineering: the system around the model - Canonical URL: https://bayrameker.com/blog/harness-engineering/ - Published: 2026-09-20 - Topic: Systems - Summary: A capable model can propose an action. A reliable system must control it, verify the result and know when to stop. - Image: https://bayrameker.com/media/blog/harness-engineering.webp — A model proposes an action, the harness checks authority, and the environment returns evidence for the next step. An agent changes the code, runs a command and announces that the job is done. The patch looks plausible. The summary is convincing. Then someone opens the application and finds that the feature never worked. That gap between a convincing account and a verified result is where harness engineering begins. A stronger model may make fewer mistakes. A longer prompt may remind it to check its work. Neither, by itself, gives the surrounding software a dependable way to establish what happened. The missing piece is the system around the model: the layer that manages context, controls actions, preserves state and asks for evidence before accepting completion. ## The model is one part of the system A model can reason about a task and propose the next action. The harness determines what it can observe, which tools it can call, what those calls are allowed to change and how their results become the next step's context. The environment is where those actions meet reality: a repository, a database, a browser, a service or a person. This is the framing in Hailin Zhong and Shengxin Zhu's May 2026 preprint, [AI Harness Engineering](https://arxiv.org/abs/2605.13357). Their proposed H0–H3 ladder adds tools, persistent context and verification in stages. The useful distinction is the evidence each stage leaves behind.
More support. More evidence.The H0–H3 proposal
  1. H0Task and repository

    The model receives the work and the files.

    Leaves a patch
  2. H1Tools and protocols

    Registered tools and test commands make actions explicit.

    Adds execution traces
  3. H2Context and memory

    Project memory and task state survive between steps.

    Adds a record of decisions
  4. H3Verification

    Reproduction and deterministic checks establish the outcome.

    Links requirements to evidence
Adapted from Zhong and Zhu. These are proposed levels of runtime support, not a benchmark showing that every additional layer improves performance.
The paper includes a controlled validation task; it does not establish a general performance law. I read it as a useful architectural proposal. It makes the unit of analysis the complete system, and makes the evidence of a run as important as its final patch. ## Give the harness a clear job In practice, I group the work into seven responsibilities. The boundaries matter more than the terminology: each responsibility needs an owner outside the model's next response.
Contract
Define the task, the evidence that counts as done and the conditions that end a run.
Context
Select the relevant sources, instructions and prior decisions for this step.
State and memory
Distinguish what was proposed, what was attempted and what actually completed.
Tools
Expose explicit operations with validated inputs and useful, bounded results.
Authority
Enforce access rules and require approval before an action crosses its permitted scope.
Verification
Check claims against tests, sources, transaction records or the running product.
Recovery
Record failures, bound retries and provide a route to repair or human handover.
Some of this is ordinary software engineering: schemas, state machines, access controls and logs. The difficulty is putting those pieces around a component whose next action is not fully predictable. Good architecture makes that uncertainty manageable without pretending it has disappeared. ## Put the rule where the action happens Consider a refund that requires approval. Writing “ask a person first” in the prompt expresses the rule. Checking an approval record at the execution boundary enforces it. The distinction is concrete. The tool should validate that approval belongs to this action, this amount and this recipient before it sends anything to the payment service. The model can propose a refund; it cannot manufacture the authority to issue one.
Where the rule lives changes the guaranteeIllustrative approval flow
Instruction in a prompt
  1. Read the rule
  2. Model chooses
  3. Call the tool

Compliance depends on the model following the instruction.

Check at the tool boundary
  1. Propose action
  2. Validate approval
  3. Execute if authorised

Missing or mismatched approval stops the call before it reaches the service.

The enforced path assumes the action can only reach the service through the checked boundary. An unguarded alternative would undo the guarantee.
Joongho Ahn and Moonsoo Kim examine this move in their July 2026 preprint, [From Prompts to Contracts](https://arxiv.org/abs/2607.08028). They place source boundaries, routing rules, trace requirements and output checks in code around a replaceable model. Their code-owned checks held across 270 composition-boundary runs with three hosted models; deliberate faults activated the validators. The scope matters: fixed scenarios over public data from five Korean corporate groups. Those results support the pattern within that evaluation. They do not mean all agent behaviour can be made correct by adding validators. A check can enforce only the property it was designed to inspect. ## Build a loop that can accept a refusal Prompts still matter. Clear tool descriptions reduce ambiguity; selected context gives the model a better basis for its next decision. Anthropic's [Writing effective tools for agents](https://www.anthropic.com/engineering/writing-tools-for-agents) makes this practical through tool naming, descriptions and response design. But useful instructions sit inside a larger loop. After the model proposes an action, something must establish whether it may run and whether it worked.
The working loopOne controlled step at a time
  1. ObserveLoad relevant context and current state.
  2. ProposeThe model selects a next action.
  3. AuthoriseCheck policy and required approvals.
  4. ExecuteRun the permitted tool call.
  5. VerifyCompare the result with the contract.
Feed the evidence into the next step, or stop when the contract is satisfied.
A refusal, a failed check or an exhausted budget is a valid reason to stop. The loop needs an exit as well as a next action.
The check should reach beyond the model's account of its own work. A coding agent needs tests against the change. A deployment needs a health check on the running service. A database action needs a transaction result. Repeating “review your answer” does not create any of those signals. Anthropic's [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) describes this as obtaining ground truth from the environment. Its later [work on long-running agents](https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents) makes the approach tangible: persistent progress records, a feature list and browser tests that exercise the actual application. These are ways to make the next step depend on observable state. ## Read the trace, then check the outcome A successful final answer can conceal a poor execution path: unnecessary retries, a wrong tool call, an approval requested too late. A trace makes that path inspectable. Outcome checks establish what the path actually achieved.
An approval must precede the actionIllustrative refund trace
  1. Observe
    Load the request and account recordRecord the proposed amount, recipient and task requirements.
  2. Validate
    Check inputs before any side effectReject malformed arguments; allow only a bounded correction.
  3. Pause
    The amount requires a reviewerPersist the pending action. No refund has been issued.
  4. Approve
    Bind the approval to this exact actionValidate the reviewer, scope and current request.
  5. Execute
    Submit the authorised refund onceUse an idempotency key; record the service response.
  6. Verify
    Confirm the resulting recordCheck status, amount and recipient before reporting completion.
The trace records the sequence; the service and ledger establish the result. A timeout with an unknown outcome needs reconciliation before a retry.
Anthropic's [Demystifying evals for AI agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) treats evaluation as a test of the model, harness, tools and environment together. One practical lesson is to grade the state left behind: whether the file changed, the tests passed or the expected record exists. The transcript helps explain how the system got there. That makes traces part of the product's operating record. They should capture tool calls, policy decisions, approvals, results and enough context to investigate a failure, with appropriate handling of sensitive data. A polished completion message is useful to the reader; it is not a substitute for that record. ## Where this meets my own work I came to this problem through autonomy and the handover to a person. A machine recommends an action, but the evidence behind it is often lost before the person has to decide. [NODERIQ](/systems#noderiq) is concerned with carrying that evidence to the decision. [NowFlow](/systems#nowflow) makes approval an explicit routing step. [QFlow Studio](/systems#qflow) keeps the brief, generated source, provider route and run evidence together so an experiment can be reconstructed. In the [agentic software-testing work](/work#agentic-testing) I lead, evaluation and oversight belong before the release decision. These systems work in different domains, but they meet the same architectural question: where does a proposal acquire authority, and what evidence survives the action? ## Design for the moment something goes wrong The most revealing review of an agent system starts with a failed run. Can we tell what it saw? Can we reconstruct the action it attempted? Did a policy stop it in time? Does the next attempt know what already happened? Those questions turn harness engineering into concrete design work. Define completion before execution. Put enforceable rules at the action boundary. Preserve enough state to recover. Check the result against the world. A capable model creates possibilities. The surrounding system determines which possibilities become dependable work. --- # Reading a preprint honestly - Canonical URL: https://bayrameker.com/blog/reading-a-preprint-honestly/ - Published: 2026-08-20 - Topic: Research - Summary: A preprint is a real contribution and it has not been reviewed. Both halves of that sentence are load-bearing. - Image: https://bayrameker.com/media/product/qflow-evidence.webp — Evidence review interface showing provenance and verification state Everything I have published this year is a preprint or a self-published book. Two arXiv papers, one SSRN architecture paper with a Zenodo deposit, and a monograph through my own press. None of it has been peer-reviewed, and every page of my site says so. That is an unusual amount of effort to spend undermining your own work, so it is worth explaining why I think it is the right amount. ## The two failure modes There are two ways to get this wrong and they are opposites. The first is inflation. Describing an arXiv posting as *published research* in a context where the reader will hear *peer-reviewed*. Letting a DOI stand in for validation, when a DOI is a persistent identifier and nothing more. Listing Amazon distribution as though a retailer had assessed the content. Treating repository stars as adoption. The second is deflation, and it is less discussed. Some people conclude that because a preprint has not been reviewed, it does not count. That is wrong too. A preprint is a real contribution with a public timestamp, an author who has attached their name, and a permanent identifier. Whole fields have run on preprints for decades. The correct position is narrow and holds both: this is a genuine research output, and it has not passed peer review. Anyone reading it should apply their own scrutiny in place of a reviewer's. ## What the infrastructure actually provides It is worth being precise about what each service does, because the names get used as though they carried more weight than they do. **arXiv** provides distribution, a timestamp and a permanent identifier. Moderation checks that a submission is on-topic and not obviously spurious. It is not review. **A DOI** provides a resolvable, persistent pointer. It says nothing about quality. A DOI on a preprint and a DOI on a Nature paper are the same kind of object. **Zenodo** provides archival deposit and citation stability. It is infrastructure operated by CERN, and depositing there is an act by the author, not an assessment by anyone. **SSRN** provides distribution and indexing in its subject areas. **ResearchGate and Amazon** provide indexing and retail distribution respectively. None of these constitutes endorsement, and I list them on my site as a provenance chain precisely so a reader can see what each one contributes and stop there. ## Why publish this way at all I publish preprints because the alternative is slower and the work is time-sensitive in a moving field. I publish the book through my own press because a synthesis across quantum, biological, neuromorphic and agentic-AI literature does not have an obvious peer-reviewed venue, and waiting for one to exist was not attractive. Both of those are trade-offs rather than principles. The cost is that the work carries less external validation. The mitigation is to be explicit about the maturity of every claim inside the work itself, which is why the book separates what is demonstrated from what is plausible from what is speculation, and why both QANTIS papers state their operating envelopes. ## The test I apply Before publishing anything I ask a single question. *If a reviewer who wanted this to be wrong read it carefully, what would they attack?* Then I write that into the paper. It is uncomfortable and it makes the work better, and it has the useful side effect that when someone does raise the objection, the answer is already on the page. A stated limitation is not a weakness in a claim. It is the thing that makes the claim a claim rather than an assertion. --- # What I expect from hybrid systems in 2027 - Canonical URL: https://bayrameker.com/blog/what-i-expect-from-hybrid-systems-in-2027/ - Published: 2026-08-14 - Topic: Outlook - Summary: Four predictions with dates and failure conditions attached, so they can be checked against next year. - Image: https://bayrameker.com/media/product/quantum-defense.webp — Hybrid quantum and classical systems in a defence context Predictions are cheap unless they can be wrong. So here are four, each with the condition that would falsify it, written down in August 2026 so that August 2027 can settle them. ## 1. The advantage conversation moves to operating envelopes I expect the useful quantum results published next year to look less like *we beat classical* and more like *here is the region where this method is competitive, and here is where it stops*. This is already happening quietly in the applied literature and it is a healthier equilibrium than the one it replaces. An operating envelope is checkable. An advantage claim usually turns out to depend on which classical baseline was chosen. **Falsified if:** the highest-profile applied results of 2027 are still headline advantage claims that reviewers spend six months relitigating. ## 2. Evidence infrastructure becomes a procurement question Right now, when an organisation buys or builds an autonomous system, the questions are about capability. I expect the questions to shift toward what the system can prove about its own behaviour after the fact. The driver is not ethics. It is liability and audit. Once a system produces recommendations that inform consequential decisions, someone eventually has to reconstruct why a particular recommendation was made. Organisations that cannot do that will find out expensively. **Falsified if:** by late 2027 procurement in this space still treats decision provenance as a nice-to-have rather than a requirement line. ## 3. The bottleneck stays architectural I do not expect scale to resolve the problems I care about. Uncertainty representation, human authority and verification are structural questions. A larger model does not answer a structural question, it answers it faster and with more confidence, which in some cases is worse. This is the argument of the book I published this year, and it is the prediction I am least likely to be talked out of. **Falsified if:** a scale-driven result in 2027 substantially closes the gap on calibrated uncertainty under distribution shift without an architectural change alongside it. ## 4. Hybrid work gets less interesting and more useful The quantum components that survive into real systems next year will be small, bounded and unglamorous. Belief updates. Sampling subroutines. Optimisation kernels with well-characterised inputs. They will sit inside classical systems that own the decision, and their contribution will be measured in narrow terms. That is a downgrade from the field's early ambitions and an upgrade in credibility. Every mature technology went through the same transition, from claiming everything to doing one thing that holds. **Falsified if:** a general-purpose quantum decision system reaches deployment in a consequential domain during 2027 with an evidence record that survives review. ## The one I would most like to be wrong about The third. If someone demonstrates next year that the architectural problems dissolve at sufficient scale, a great deal of careful work becomes unnecessary, and I would take that trade. I do not expect it, but I would rather be wrong about the shape of the problem than right about a hard one. --- # Open-sourcing the part you can defend - Canonical URL: https://bayrameker.com/blog/open-sourcing-what-you-can/ - Published: 2026-08-11 - Updated: 2026-09-03 - Topic: Practice - Summary: The public surface can be source code, an explanatory research edition or an architecture record. The line is drawn by what can be supported responsibly. - Image: https://bayrameker.com/media/product/tasknebula.webp — TaskNebula coordination platform Four systems, several kinds of public surface. TaskNebula and QANTIS under MIT. NowFlow Community and qmesh under Apache-2.0. NeuraOS publishes its product, governance and assurance architecture without distributing source code or binaries. People ask where the line sits and whether it is drawn by what is valuable. It is not. It is drawn by what I can support. ## The actual constraint An open repository is a promise. Someone will build on it, file an issue, and expect an answer. If I release something I cannot maintain, I have not contributed anything, I have created an abandoned dependency with my name on it. So the open edition is the part where the promise is affordable. For QANTIS that is the explanatory edition: the structure, the approach, enough to read and reason about and reproduce the shape of. The full experimental tooling stays private, not because it is a trade secret but because supporting it as a public artefact would consume the time that produces the research. I say that on the site rather than implying the repository is more complete than it is. ## Where the commercial line falls For the product systems the split follows assurance rather than features. NowFlow Community carries the workflow engine and the governed agentic patterns. The commercial edition carries the integration surface, 188 blocks and over 300 connections, and the operational assurance around it. NeuraOS draws the line differently: its public repository documents the runtime and governance architecture, while the proprietary edition carries the signed model-to-device path and its delivery controls. NODERIQ Defense has no open edition and will not have one. Not because the ideas are secret, since the architecture is published on the site in detail, but because a defence edition carries obligations that an open repository cannot hold. ## What open source is actually for here Three things, in order of how much they matter to me. **Verifiability.** A paper that cites a repository nobody can open is asking for trust it has not earned. The QANTIS repository exists so a reader can check that the described approach is a real thing. **Recruitment of scrutiny.** The fastest way to find out that you are wrong is to publish something someone competent can read. This has happened to me and it was useful both times. **Usefulness without a transaction.** TaskNebula is the clearest case. It solves a problem for teams working alongside agents, and there is no version of the business where charging small teams for it makes sense. The reasoning behind releasing it is set out in a separate article from March. ## The honest accounting Across seven public repositories the counts are modest. TaskNebula has the adoption; the research repositories have very little, which is what you would expect for research software with a narrow audience. Those numbers appear on the site with a date attached, and they are not presented as evidence of traction. A repository with six stars that supports two published papers has done its job. Reporting it as though it were a popularity result would be a category error. --- # Why the planner stays classical - Canonical URL: https://bayrameker.com/blog/why-the-planner-stays-classical/ - Published: 2026-08-07 - Topic: Systems - Summary: The most consequential architectural decision in the quantum work is about where a component is not allowed to go. - Image: https://bayrameker.com/media/product/noderiq-console.webp — Uncrewed assurance console showing decision evidence and authority state In both QANTIS papers the quantum processor handles a bounded belief-update primitive, and the planner that consumes its output is classical. So is the action authority. People sometimes read this as a concession, a placeholder until hardware improves. It is not. It is the decision I would defend hardest. ## Three reasons, in increasing order of importance **Availability.** Quantum backends queue. They calibrate. They go offline. A decision loop that stalls when a backend is unavailable is not a decision loop, it is a research demo with a scheduling dependency. Keeping the planner classical means the system degrades to a path that was always there rather than stopping. **Reasoning.** I can reason about the failure modes of a classical planner. I can bound them, test them, and explain them to someone who has to sign off on the system. The equivalent story for a quantum planner does not exist yet, and pretending otherwise in a domain where the output authorises physical action would be indefensible. **Authority.** This is the one that matters. Action authority is a property of the system that someone has to be accountable for. Placing it inside a component whose behaviour is probabilistic in an unfamiliar way, executed remotely, on hardware whose calibration state changes between runs, moves accountability somewhere it cannot sit. ## What a bounded primitive buys you Once the quantum part is bounded, several things become tractable at once. You can compare it against an exact classical reference, because the input and output are both well-defined distributions. You can test whether it survives repeated use, which is what the sequential study did across 8, 12, 20 and 32 steps. You can put an evidence gate in front of it. You can remove it without redesigning the system. That last property is the one people underrate. An architecture where the interesting component can be taken out is an architecture you can be honest about. If the quantum path never clears G3, NODERIQ still has a classical AI core that does useful work. The programme does not depend on the research succeeding. ## The version of this I disagree with There is a school of thought that says bounding the quantum component this tightly gives away the upside. If the quantum layer only ever handles one primitive inside a classical loop, the argument goes, you have capped what quantum can contribute before you have found out what it could do. I take that seriously and I still think it is wrong, for a practical reason. The unbounded version is not testable. You cannot compare it against a classical reference because there is no matching classical formulation. You cannot attribute a result to the quantum layer because the layer is entangled with everything else in the system. You end up with a system that either works or does not, and no way to say why. A capped contribution you can measure is worth more than an uncapped contribution you cannot. ## Where this shows up in NODERIQ The same reasoning produced the separation between recommendation and authority at the programme level. A system may recommend. A person authorises. The quantum question is a special case of a more general commitment: components that carry uncertainty feed decisions, and decisions that carry consequence route to someone accountable. Getting that boundary right in the small case, a single belief-update primitive, is how you learn where to put it in the large one. --- # Teaching a workflow rather than a toolchain - Canonical URL: https://bayrameker.com/blog/qflow-university-lab/ - Published: 2026-08-04 - Topic: Practice - Summary: University quantum labs spend most of their first month on environment setup. That month is the interesting part of the problem, and it is being wasted. - Image: https://bayrameker.com/media/product/qflow-ui.webp — QFlow Studio interface showing the workflow record A recurring pattern in university quantum courses: the first several weeks go to environment setup, SDK version conflicts, provider credentials and the general friction of getting anything to run at all. By the time students reach the material the course is about, a meaningful fraction of the term is gone. I do not think this friction is teaching anyone anything useful. Nobody learns quantum information from a dependency error. ## What is worth the students' time The genuinely difficult ideas in this space are not implementation details. They are conceptual, and they are the ones that get squeezed. Why does the calibration state of a backend change what your result means. What is a strong classical baseline, and why is choosing a weak one the most common methodological failure in the field. What does it mean for a result to hold within an operating envelope. How do you tell whether a difference between two distributions mattered to the decision that consumed them. Those questions transfer. They apply to any hybrid system a student builds in the next decade, whatever SDK is current by then. Spending term time on SDK friction instead is a poor trade. ## Workflow as pedagogy The reason I care about this from a product angle is that a workflow record makes those questions visible rather than abstract. If the brief, circuit, source, provider route, run context and evidence all sit in one record, a student can see that changing the backend changed the result and that the calibration conditions were different. That is a lesson delivered by the structure of the tool rather than by a lecture slide. If the same information is scattered across a notebook, a console and a screenshot, the lesson is available only to students who already know to look for it. There is also a straightforward supervision benefit. A supervisor reviewing twelve student projects can read twelve evidence packets rather than twelve differently organised notebooks. ## The boundary I keep I am wary of a specific failure here, which is a tool that makes students productive without making them competent. Abstraction that hides the source code produces people who can operate a workflow and cannot debug one. So the generated source stays visible and synchronised across Qiskit, Cirq and OpenQASM. The code that ran is the code on screen. If a student wants to leave the workflow and work directly in Qiskit, nothing prevents it, and I would consider a course that produced students unable to do so a failure regardless of what they built inside the tool. The claim is narrow. This removes friction that was never teaching anything, and it makes provenance visible. It does not teach quantum information, and any tool that claims to should be treated with suspicion. --- # What mentoring defence AI students taught me about evaluation - Canonical URL: https://bayrameker.com/blog/mentoring-defence-ai-students/ - Published: 2026-07-31 - Topic: Practice - Summary: Twelve months of reviewing thesis work on UAV imagery and radar analysis, and the failure mode that showed up almost every time. - Image: https://bayrameker.com/media/product/noderiq-bvlos.webp — Beyond visual line of sight operations for uncrewed platforms Between 2024 and 2025 I was an industry mentor in the SAYZEK Academic Thesis Programme, run jointly by Türkiye's Presidency of Defence Industries and the Council of Higher Education. The programme took 421 applications from 119 universities, accepted 356, and assigned 90 industry mentors drawn from 20 companies. Each student's progress went through three interim reports that mentors reviewed and scored. My students were working on AI-enabled imagery processing for UAV and armed-UAV platforms, and on radar-data analysis for AI-based ground localisation and target classification. Everything below is at a non-operational level. No restricted data, radar parameters or model thresholds appear here, and no student is identified. ## The failure mode Almost every project arrived with the same shape of problem. Not a modelling problem. An evaluation problem. A student would show me a result on a test set and the result would be genuinely good. Then I would ask where the test set came from, and it would turn out to have been collected under conditions close enough to the training set that the split was measuring memorisation rather than generalisation. Same sensor, same altitude band, same time of day, sometimes the same flight. This is not carelessness. It is what happens when data is scarce and the evaluation protocol is inherited from a tutorial rather than derived from the deployment condition. ## The question that helped The most useful thing I could offer was usually a single question. *Under what condition would this model be wrong, and does your test set contain that condition?* For imagery work the answers came quickly once the question was asked. Low sun angle. Partial occlusion. A sensor the model had never seen. A target class that is rare in training precisely because it is rare in the world, which is often exactly the class that matters. For radar work it was more subtle, because the intuitions people carry from computer vision transfer badly. Clutter is not noise. Multipath is not an outlier. A model that treats them as nuisance rather than structure will look excellent on a clean set and fail in the way that costs the most. ## Why this connects to the rest of the work I did not expect a mentoring commitment to feed directly into research design, but it did. The habit I kept pressing on students is the habit that runs through QANTIS. State the operating envelope. Test at the boundary rather than the centre. Report the condition under which the result stops holding, and treat that report as part of the result rather than as an admission against interest. It is easier to teach than to practise. Writing a paper that says *here is where our method degrades* is uncomfortable, because a reader can quote that sentence. But a result without a stated boundary is not a weaker claim than one with a boundary. It is an unfalsifiable one, which is worse. The certificate for that programme is dated 10 July 2025 and signed by the President of Defence Industries. It records a mentoring contribution. It is not a government award or a competitive placement, and I describe it that way because that is what it is. --- # Reproducibility is a product decision - Canonical URL: https://bayrameker.com/blog/reproducibility-is-a-product-decision/ - Published: 2026-07-24 - Topic: Practice - Summary: Nobody reconstructs a quantum experiment from memory two months later. The tooling either captured it or it did not. - Image: https://bayrameker.com/media/product/qflow-canvas.webp — QFlow Studio canvas showing a circuit graph beside its generated source Quantum work fragments. The objective lives in a document, the circuit in a notebook, the generated source in a repository, the provider route in a console, the run log in a dashboard, and the review in a thread somewhere. Each of those is reasonable on its own. Together they guarantee that two months later nobody can reconstruct what was actually run. I have watched this happen to careful people. It is not a discipline problem. It is a tooling problem that gets misdiagnosed as a discipline problem, which is why exhorting researchers to be more rigorous never fixes it. ## What the record has to hold QFlow Studio treats the whole thing as one controlled record: brief, circuit, source, route, run, evidence. The brief carries the objective, the owner and the constraints, because a result without its original question is very hard to interpret later. The circuit is the visual graph. The source stays synchronised across Qiskit, Cirq and OpenQASM, so the code you read is the code that ran. The route captures provider fit, preflight and fallback, which matters because the same circuit on a different backend is a different experiment. The run holds the execution context. The evidence packet holds the result, the trace and the review state. The point of putting them in one record is not tidiness. It is that the links between them are where the meaning lives. A result is only interpretable next to the calibration conditions it was taken under. ## Reviewer-safe by construction There is a second problem underneath the first. Provider credentials and share-safe evidence are different categories of thing, and most workflows mix them because the fastest way to show someone a result is to show them your console. Separating them at the data model level means an evidence packet can be handed to a reviewer, a collaborator or an auditor without a redaction step. Redaction that happens at share time is redaction that eventually gets forgotten. ## Where the AI boundary sits AI may draft. Consequential execution stays user-approved. I hold that line firmly, and the reason is not caution about model quality. It is that submitting a job to hardware costs money, consumes a queue slot and produces a record. Those are consequences. A system that can produce consequences without a person in the loop needs a much stronger justification than convenience, and I have not seen one for this workload. ## What is actually claimed The innovation here is the operating and evidence model around existing quantum toolchains. It is not a new SDK. It does not cover every backend. It makes no claim about quantum performance, and the product surface says so. I am specific about this because the adjacent claim is available and tempting. A platform that sits above the SDKs could easily be described as though it improved what the SDKs do. It does not. It improves what happens around them, which is a smaller claim and a true one. --- # Evidence gates, or how to decide whether quantum earns its place - Canonical URL: https://bayrameker.com/blog/evidence-gates/ - Published: 2026-07-17 - Topic: Systems - Summary: Most quantum programmes have no defined condition under which they would stop. Four gates fix that. - Image: https://bayrameker.com/media/product/qflow-evidence.webp — Enterprise evidence review inside QFlow Studio Ask a quantum programme what would have to be true for them to abandon the quantum path, and watch what happens. Most cannot answer. Not because the people are unserious, but because the question was never posed at the start, and by the time it becomes uncomfortable there is too much invested to pose it. I wanted a structure that made the answer available from day one. Four gates, each with a condition that can be failed. ## G0, define The workload has to be explicit, measurable, latency-tolerant and safe to run outside local control. Most candidate workloads fail here, and failing here is cheap. If you cannot state the workload precisely enough to measure it, everything downstream is going to be an argument about interpretation. If it is latency-critical, current hardware access patterns rule it out regardless of how elegant the algorithm is. If it cannot safely leave local control, a cloud quantum backend is not available to you. Getting a clear no at G0 costs a week. Getting it at G3 costs a year. ## G1, simulate The alternative method has to produce valid results against a strong classical reference. The word doing the work in that sentence is *strong*. Comparing a quantum method against a deliberately weak classical baseline is the most common way this field produces results that do not survive contact with a competent reviewer. The baseline has to be the one a capable engineer would actually build if you told them to solve the problem without quantum. In the QANTIS work the reference is exact Bayesian inference, which is the correct comparison and also an unforgiving one. ## G2, preserve Hardware execution has to preserve information, constraints and uncertainty. This gate exists because of a specific failure mode. A method can appear to succeed by discarding the thing that made the problem hard. If the uncertainty representation does not survive execution, you have not solved the inference problem, you have replaced it with a different and easier one. The comparison is then meaningless even if the numbers look good. ## G3, qualify The complete path has to add decision value after accuracy, latency, reliability and cost are all counted. Everything is in scope at G3: the queue time, the calibration drift, the cost per shot, the operational complexity of maintaining a second execution path. A method can be genuinely more accurate and still fail this gate, because accuracy is not the only currency. ## Why publish the gates Two reasons. The first is internal. A team that knows the exit conditions argues about evidence rather than about enthusiasm. When a result comes in below the bar, the conversation is short. The second is external. If I tell you that the quantum pathway in NODERIQ is conditional, that statement is worth very little unless I also tell you what the condition is. Published gates make the claim checkable. Someone can read them, look at what has been published, and form their own view about which gates have actually been cleared. Right now the honest answer is that G0 and G1 are addressed for the belief-update workload, G2 is what the hardware campaigns in both QANTIS papers were built to test, and G3 is not settled. The conditional quantum pathway is stage three of a three-stage roadmap for a reason. Stage one is a classical AI core, and it comes first because it has to work whether or not stage three ever does. --- # A second QANTIS paper, and why it asks a smaller question - Canonical URL: https://bayrameker.com/blog/a-second-qantis-paper/ - Published: 2026-07-10 - Topic: Research - Summary: The foundational study showed the mechanism works on hardware. The follow-on asks whether it still works the eighth time you use it. - Image: https://bayrameker.com/media/product/qantis.webp — QANTIS belief-update service running on IBM Heron hardware The second QANTIS preprint went up yesterday as arXiv:2607.06760. It is a narrower paper than the first one, and that is the point. The foundational study, arXiv:2603.00785, ran 45 experiments across three IBM Heron backends and asked whether the belief-update mechanism works on real hardware at all. It does, within stated conditions. A Tiger belief-oracle case amplified a rare observation probability from 0.179 to 0.907 while preserving the posterior at a Hellinger distance of 0.0015 from exact Bayes. That is a clean mechanism result. But a clean mechanism result answers a question nobody planning a real system is asking. They want to know what happens on the eighth call, and the twelfth, and whether the thing degrades gracefully or falls off a cliff. ## The actual question Sequential decision-making is where quantum components tend to get quietly abandoned. A single circuit can look excellent. Chain twenty of them inside a loop where each posterior feeds the next belief state, and error accumulates in ways that a one-shot benchmark will never reveal. So the follow-on study embeds the calibrated belief-update service inside a classical planner and runs it around the loop. Primary campaigns cover 8-step and 12-step decision trajectories. Controls extend to 20 and 32 steps, specifically to find where the behaviour changes. The reported finding is that hardware-derived and exact-Bayes posteriors selected the same immediate action at every decision point checked. That is a decision-level result rather than a fidelity result, and I want to be precise about the difference. Posterior fidelity tells you how close two distributions are. Decision consistency tells you whether the difference between them ever mattered. Those come apart. A posterior can drift measurably and still produce identical actions, because the planner only needs the ordering to hold near the decision boundary. Equally, a small drift at the wrong moment can flip an action. Reporting fidelity alone would have obscured this. ## What stays classical The planner is classical. The action authority is classical. The quantum processor handles one bounded primitive inside a loop that a classical system owns end to end. This is an architectural commitment rather than a temporary limitation waiting for better hardware. If the quantum component fails, degrades or becomes unavailable, the system falls back to a classical path that was already there. The alternative, in which quantum sits on the critical path for a decision that carries consequence, is not a design I would put in front of anyone. ## What the paper does not claim No wall-clock advantage over strong classical methods. The paper does not contain that claim and I will not make it in a summary. No universal scaling. The results hold within the stated POMDP, hardware, calibration and experiment settings, and the honest reading is that we found an operating envelope, not a general capability. No production readiness or certification. This is a preprint on arXiv. It has not been through peer review, and describing it as though it had would be a straightforward misrepresentation. ## Why publish a bounded result There is a temptation in this field to hold work back until it says something larger. I think that is a mistake, and I think it is part of why quantum computing has an evidence problem. A bounded result that states its limits is more useful to another researcher than an unbounded claim that does not. If someone reads the sequential study and concludes that the operating envelope is too narrow for their application, the paper has done its job. It has saved them a campaign. The reproducibility surface sits in the public repository alongside both papers. What is public is the explanatory edition rather than the full experimental tooling, and I say so on the page rather than implying otherwise. Both papers are on arXiv, both are first-author, and neither has been peer-reviewed. All three of those facts belong in the same sentence. --- # What 45 experiments on IBM Heron actually taught us - Canonical URL: https://bayrameker.com/blog/what-45-experiments-taught-us/ - Published: 2026-07-07 - Topic: Research - Summary: The campaign behind the foundational paper, and the three findings that changed how the second one was designed. - Image: https://bayrameker.com/media/product/qantis.webp — QANTIS belief-update service running across IBM Heron backends The foundational QANTIS paper runs to 31 pages with 4 figures and 12 tables, and most of that volume is one thing: a 45-experiment campaign across three IBM Heron backends. People sometimes ask why a paper needs that many runs. The short answer is that fewer would have let us believe things that were not true. ## The three components The platform integrates three pieces that are usually studied separately. Quantum belief update through Grover amplitude amplification and BIQAE. Multi-target data association posed as a QUBO and solved with FPC-QAOA. Composable error mitigation applied across both. Studying them together was the point. A belief-update result measured in isolation tells you very little about what happens when the same hardware, in the same calibration state, also has to carry an association problem. ## What the campaign changed **Backend identity matters more than backend specification.** Three Heron processors with the same nominal specification did not behave interchangeably. Results that looked like method effects on one backend turned out to be calibration effects when the same circuit ran elsewhere. Any result reported without naming the backend and the calibration window is a result you cannot reproduce, and we started treating that as a hard requirement rather than good practice. **Mitigation composes badly if you let it.** Error mitigation techniques that each improve a result individually can interact in ways that make the combination worse than either alone. That is not a surprising claim once stated, and it is very easy to miss if you only ever evaluate one configuration. **The interesting numbers are at the boundary.** The clean result in the paper, a rare-observation probability amplified from 0.179 to 0.907 with a Hellinger distance of 0.0015 from exact Bayes, is a centre-of-envelope result. It is real and it is not where we learned the most. The runs that taught us something were the ones near the edge of what the hardware could hold. ## How it shaped the second paper The sequential study exists because of a gap the first campaign made obvious. Every one of those 45 experiments was, in the end, a single-shot measurement of a mechanism. None of them told us what happens when the belief-update service is called repeatedly and its output feeds the next prior. So the follow-on asks exactly that, across a sequential Tiger POMDP horizon, with 8-step and 12-step primary campaigns and 20-step and 32-step controls. The controls are there specifically to find the failure, not to confirm the success. ## The thing I would tell someone starting Design the campaign so that a null result is publishable. If your experiment plan only produces a paper when the method works, you have built an incentive to stop looking at the point where it stops working. The 45-experiment structure was expensive in queue time and it meant that whichever way the result went, we had something to report. That is worth paying for. Both papers are public preprints on arXiv. Neither has been peer-reviewed, and neither claims wall-clock advantage over strong classical methods. --- # The gap NODERIQ closes - Canonical URL: https://bayrameker.com/blog/the-gap-noderiq-closes/ - Published: 2026-07-03 - Topic: Systems - Summary: Autonomous systems are good at producing recommendations and bad at carrying the evidence for them. That gap is where the work is. - Image: https://bayrameker.com/media/product/noderiq.webp — NODERIQ programme, distributed mission intelligence across uncrewed platforms There is a specific failure that shows up in almost every autonomous system I have looked at, and it is not the one people expect. The models are usually fine. The sensors are usually fine. What breaks is the handover: the moment a machine produces a recommendation and a person has to decide whether to act on it. At that moment the person needs three things. What did the system observe. How confident is it, and on what basis. What would happen if it is wrong. In most deployed systems, at least one of those is unavailable, because the pipeline flattened it somewhere upstream. A probability distribution became a label. A confidence became a threshold. A chain of inference became a single number on a dashboard. NODERIQ exists because of that flattening. ## Recommendation is not authority The design starts from a separation that sounds obvious and is routinely violated in practice. A system may recommend. A person authorises. Those are different operations with different evidentiary requirements, and collapsing them is how organisations end up with autonomy they cannot defend afterwards. Keeping them separate has consequences all the way down the stack. If a human is going to authorise, the system has to preserve what the human needs in order to authorise responsibly. That means uncertainty has to survive the journey from sensor to screen. It means the evidence has to be small enough to cross a constrained link, because a contested environment does not give you bandwidth on request. It means ambiguity has to route somewhere, to someone accountable, with the supporting material attached rather than described. So the operating loop reads: sense, qualify, share, coordinate, verify, escalate. Qualify is the step most architectures skip. It is the one that represents uncertainty rather than resolving it prematurely. ## Where the core lives The core is classical and edge-operable. That is a deliberate constraint rather than a limitation I am apologising for. Edge-operable means the system remains useful when the link degrades, which is the condition it will actually operate in. Classical means the primary path uses computation that exists, is understood, and can be reasoned about under failure. Anything more exotic has to earn its place against that baseline. This is where NODERIQ connects to the quantum research. The QANTIS work asks whether a quantum belief-update primitive can be trusted inside a classical decision loop. That is not a question about quantum supremacy. It is a question about whether a specific bounded component adds decision value once accuracy, latency, reliability and cost are all accounted for. Four evidence gates decide it: **G0, define.** The workload has to be explicit, measurable, latency-tolerant and safe to run outside local control. If it is not, nothing else matters. **G1, simulate.** The alternative method has to produce valid results against a strong classical reference. Not a weak one chosen to flatter it. **G2, preserve.** Hardware execution has to preserve information, constraints and uncertainty. A method that gives a faster answer by discarding the uncertainty has not solved the problem, it has moved it. **G3, qualify.** The complete path has to add decision value after everything is counted. A method that fails a gate does not advance. That is the entire mechanism, and it is deliberately unglamorous. ## Two surfaces, one core The same core serves two operating contexts. NODERIQ Defense addresses multidomain autonomy for uncrewed platforms in contested environments, and went to NATO DIANA's challenge track in June. NODERIQ-CRN covers civil resilient operations: infrastructure inspection, emergency logistics, field service, connected mobility. The engineering is largely shared because the underlying problem is shared. What differs is the operating boundary. In defence the boundary is contest. In civil resilience it is safety and continuity. Both need a system that can say what it saw, how sure it was, and who authorised what followed. ## What this is not NODERIQ is an applied research and productisation programme. It is not a released product, a certified capability or a deployed system, and I am careful about that wording because the distinction matters in this domain more than most. Claiming operational validation you do not have is not merely embarrassing. In a defence context it is dangerous. The honest position is that the architecture is worked out, the evidence gates are defined, the civil and defence surfaces are described, and the classical core is where the effort currently sits. The quantum pathway remains conditional. It will stay conditional until it passes G3, and it may never pass. That last sentence is the one I would most like people to take seriously. A research programme that cannot describe the conditions under which it would abandon its most interesting idea is not a research programme. ## Machine-readable routes - Concise LLM index: https://bayrameker.com/llms.txt - Complete corpus: https://bayrameker.com/llms-full.txt - XML sitemap: https://bayrameker.com/sitemap.xml - RSS feed: https://bayrameker.com/rss.xml - Robots policy: https://bayrameker.com/robots.txt Last updated: 3 September 2026