Methodological approaches used across the Adversarial Systems Research programme, emphasising transparency, reproducibility, and substrate-independent collaboration.
Papers in the programme undergo multi-agent review before public release. The process deploys specialised large language model agents configured with domain expertise (finance and economics, political economy, computational philosophy) to provide systematic critique across theoretical rigour, empirical evidence quality, methodological soundness, and argumentative coherence.
Agent reviewers operate under explicit epistemic constraints matching their knowledge cut-offs, evaluate papers against disciplinary standards for their respective fields, and generate review reports with scored assessments and actionable feedback. This complements traditional human peer review by providing rapid iteration cycles, identifying technical gaps early, and maintaining consistent evaluation criteria across heterogeneous research domains.
Review transcripts are archived and available on request. The methodology is applied systematically to working papers before Zenodo submission, ensuring baseline quality standards are met prior to release.
Research production uses a tiered AI collaboration setup: Claude Code for strategic orchestration, complex reasoning, and architectural decisions; Gemini Pro for boilerplate generation, documentation drafting, and large-context analysis; Perplexity for literature discovery beyond training cut-offs. This multi-model approach delegates mechanical work to cheaper tiers while keeping rigorous analytical standards for substantive contributions.
The collaboration philosophy treats AI systems as cognitive scaffolding and intellectual sounding boards, not autonomous generators. All theoretical frameworks, empirical interpretations, argumentative claims, and policy proposals originate from human reasoning. AI tools accelerate mechanical tasks (reference formatting, literature search, structural organisation) and give iterative feedback on clarity and coherence, but the substantive intellectual contribution remains the researcher's.
This workflow reflects substrate-independent collaboration: the architecture performing a task (biological neurons or artificial networks) matters less than the quality of reasoning, the rigour of the method, and the validity of the conclusions. Disclosure statements in all publications document AI tool usage in full.
Computational outputs include open-source code repositories, interactive visualisation dashboards, and complete replication materials. Papers with quantitative components (cryptocurrency event studies, TARCH-X volatility models, Monte Carlo simulations) provide executable code, documented data-processing pipelines, and environment specifications enabling full replication.
Interactive dashboards extend static PDF figures by allowing exploratory data analysis, parameter sensitivity testing, and visual investigation of results. Current work includes cryptocurrency volatility dashboards, with planned extensions to consent-friction calculators and governance alignment visualisations through the Dissensus programme.
Code is MIT-licensed (software) or CC BY 4.0 (documentation), hosted on GitHub, and linked from paper landing pages, so findings stay verifiable, extendable, and usable by others.
Outputs use semantic versioning (MAJOR.MINOR.PATCH) tracked through Zenodo. Papers evolve through iterative releases: undergraduate theses become expanded preprints, working papers incorporate peer feedback, and published versions receive post-publication corrections or extensions. Each version gets a distinct DOI while staying linked to the canonical work, keeping the whole lifecycle citable.
This treats papers as living documents that improve through community engagement, transparent iteration, and cumulative refinement. Changelogs record substantive additions, methodological improvements, and evidence updates between versions. Murad Farzulla's Zenodo community aggregates all outputs, giving centralised access to the programme.
Each output is disseminated through three complementary channels, each suited to a different use:
A permanent, citable, DOI-assigned academic record.
Exploratory visualisation and parameter sensitivity analysis. The ASRI systemic-risk dashboard is live at asri.dissensus.ai.
Complete source code, replication materials, and environment specifications.
This balances formal academic requirements (a citable DOI, archival permanence) with modern scientific communication (interactive exploration, code transparency, extensibility).