About
A small group with a long thesis.
Dissensus is an independent complexity-science research group in London. We study complex adaptive systems — and specifically what happens in them when the parts want different things — through formal, computational and empirical work.
What the group is for
The subject is complex adaptive systems: many interacting agents, no central controller, aggregate behaviour that none of the parts intends. Markets, institutions, minds and machine collectives all qualify, which is why the papers look scattered until you notice they ask one question in four places.
The question is what happens when the parts want different things. Acting together requires delegation, delegation produces friction, and that friction is measurable — in how well interests align, how much is at stake, and how badly intent degrades in transmission. The aim is not to remove it. Friction is the cost of existing among others who want other things; the aim is to measure it well enough to say something falsifiable.
The group is independent by construction: no parent institution, no outside owner, no funder with editorial reach. Where it affiliates, it affiliates on terms that are not ownership — described under partnerships. Everything it produces is open — papers, code, datasets and tools, archived with DOIs so the work can be checked rather than taken on trust.
One programme, four fronts
The research is a single programme rather than a set of side interests. A paper on cryptocurrency volatility and a paper on machine ethics look unrelated until you notice they are both probing how a complex adaptive system behaves when its parts want different things. A separate public home for the programme is being reconsidered; until it exists, everything is published here.
The programme runs on four fronts, each a different place to catch the same phenomenon in the wild. Publications in all four are listed under Research; the ones still open to collaborators are on Projects.
- 01 Computational finance
- Market microstructure, volatility asymmetry, systemic risk, event-study methodology.
- 02 Political economy
- Legitimacy quantification, institutional friction, governance and consent mechanics.
- 03 Cognition & AI alignment
- Multi-agent coordination, machine ethics, training dynamics as development.
- 04 Formal methods
- The friction formalism itself — carried into Lean 4, which is how we found the closed form that did not survive checking.
Five rules, and what breaking them would look like
These are commitments rather than aspirations, which means each one is specific enough to fail. They are written down here so that if the group stops honouring one, the gap is visible to anyone who reads this page.
- 01
- Everything is free to read, run and citePapers, datasets and software go out with DOIs — CC-BY for text, MIT or Apache for code. Nothing the group produces sits behind a paywall or a request form. Breaking it looks like: an output described in public that you cannot actually get to.
- 02
- Every number traces back to code you can runReported results come from committed code with fixed seeds, and the datasets and pipelines are archived alongside them. Breaking it looks like: a figure whose provenance stops at “available on request”.
- 03
- We attack our own work before anyone else canEvery claim is stress-tested internally first, and collaborators are invited to break whatever survives. The goal is not to be right, it is to be less wrong sooner. Breaking it looks like: a confident headline with no stated failure mode.
- 04
- No funder, client or institution can change a resultNobody outside the group owns it, employs it or edits it. Commercial work and programme funding sit behind separate doors, and neither buys a conclusion. Breaking it looks like: a finding that happens to suit whoever paid for it.
- 05
- Work that fails gets published as having failedA null is reported as a null under the same inference standards a positive result would face — the accepted Digital Finance paper reports one about its own headline. Breaking it looks like: a project that quietly disappears instead of saying what happened.
Who does the work
Murad Farzulla
“We are more beholden to custom than reason; it is thus reason’s task to interrogate custom.”
I run Dissensus. My training is in finance — MSc Finance Analytics at King’s College London, after a First-Class BSc in Accounting & Finance at SOAS — and the research goes wherever the friction thesis leads: computational finance, political economy, AI alignment, formal methods. I develop the Axiom of Consent and Replicator-Optimization Mechanism frameworks the lab is organised around, and I hold the unfashionable view that a theory that cannot be computed, tested, or proved is not finished yet.
The full résumé lives at farzulla.org; the papers themselves are here, under Research.
Andrew Maksakov
Andrew keeps the lab’s empirical machinery running — distributed systems, GPU compute, and simulation pipelines — and is co-author on the ASRI systemic-risk index and the CBDC privacy-architecture work.
Tofik Israfilov
Tofik joined the lab in September 2026 and works on the foreign exchange strand of the programme, beginning with the alpha-asymmetry paper, where he is a co-author on the forthcoming revision.
Helena Dye
Helena runs the parts of the lab that are not research and would otherwise not happen — scheduling, correspondence, records, and the administrative side of incorporation and funding applications. The role is part-time and sits alongside a full-time job elsewhere, which is the honest shape of operations at a lab this size.
Felipe Pachano Azuaje, PhD
Felipe is an affiliated researcher contributing methodological perspectives across the programme — the standing outside view every small lab needs, without which a one-thesis lab starts agreeing with itself.
The affiliation runs through the standing Sineidocracia partnership described below.
Who the lab works with
Three kinds of thing sit here: networks the lab belongs to, groups it builds with, and people who have given it something it could not otherwise afford. None is a client relationship — those are further down, under applied work.
International Association for Safe and Ethical AI (IASEAI)
Dissensus was accepted as an IASEAI affiliate in August 2026, listed among organisations that include MIRI, METR, FAR.AI, MATS Research, the Cooperative AI Foundation, Harvard’s Berkman Klein Center, Oxford’s Institute for Ethics in AI, and Berkeley’s Center for Long-Term Cybersecurity.
It is worth being precise about what that is, because affiliate lists are easy to over-read. Affiliation means the lab broadly supports the association’s mission. It is not a grant, not an endorsement of any Dissensus result, and not a set of positions the lab is obliged to hold. IASEAI’s own terms are explicit on this in both directions: affiliates are free to take different positions and pursue different tactics, and no affiliate is listed as endorsing anything unless it opts in.
What it provides is a standing room of groups working on the same failure modes from other directions. For a lab whose stated worry is agreeing with itself, that is the whole of the appeal.
The Sineidocracia movement
A standing partnership around the movement’s manifesto project. Dissensus supplies the technical infrastructure — repository, typesetting, release process — and Murad Farzulla sits on its editorial committee as a full member with a vote, writing two of the chapters: what an assisted-deliberation infrastructure would have to guarantee, and how one would actually be built.
The movement owns its own work and can take it elsewhere at any time. That condition is written down rather than assumed, because a small lab supplying another group’s infrastructure is exactly the arrangement that quietly becomes ownership if nobody says otherwise.
The partnership is interesting rather than merely cordial, and it is worth being plain about why. The movement’s intellectual tradition is not ours, its conclusions are not ones this lab would have reached on its own, and its manifesto makes claims about legitimacy that overlap our work without agreeing with it. That is the value. A one-thesis lab with no external friction starts agreeing with itself inside a year.
Its editorial committee: Felipe Pachano Azuaje, PhD (coordinator) · Heidi Contreras · Romina Roca · Murad Farzulla, with Alejandro Landaeta and Rafael Espín as observers. The movement’s own site is forthcoming and will be the authoritative place for its people and its text; this is a description of the partnership, not a roster we maintain.
Apodex
Dissensus holds $500 in API credits under the Apodex Frontier Program, granted in August 2026 and running to November. It is the lab’s only external compute support to date, it was given with no conditions attached to what we conclude, and it is named here because unfunded labs should say where their compute comes from. Our thanks to them.
And we are looking for more
Partnership does not have to mean money, and it does not have to mean agreement. If your lab, institute, reading circle, student society or journal works on adjacent questions — coordination and its overhead, alignment under competing objectives, legitimacy and consent, systemic risk, or the methodology of checking any of it — the useful first move is usually a small one: a joint reading, a shared dataset, a reciprocal review where each side tries to break the other’s result, a co-authored piece, or simply naming each other as places to send people.
The lab is also actively looking for external oversight: people and bodies willing to audit its methods, sit on an advisory capacity, or check results it has a stake in believing. A lab that only ever hears from people who agree with it is not being reviewed, and this one has already had to retract a result that survived internal scrutiny for months. Groups whose conclusions we expect to disagree with are more useful to us than groups whose conclusions we share.
The two places money changes hands
Two separate doors, deliberately. A client owns a deliverable; a funder produces public goods. Neither buys a result, and neither buys editorial reach.
- 01 Commercial
- Organisations retain the group for applied analysisFriction analysis — mapping where competing interests generate structural conflict in a market, protocol or organisation, and decomposing it into alignment, stake and entropy. Also bespoke ASRI-style systemic-risk indices, Lean 4 formalisation with assumptions tracked in an explicit ledger, adversarial evaluation of multi-agent systems, and GARCH-family volatility modelling built on the open-source estimators we maintain. Engagements run as advisory, as scoped research with a written deliverable, or as custom tooling shipped with source and tests — a few at a time, so the research does not stall.
- 02 Funding
- Grants and philanthropy fund the programme itselfThe bottlenecks are concrete: compute for multi-agent and privacy experiments at real scale, research time to carry results through to publication, and the unglamorous distance between a working prototype and a replicable finding. What funding buys stays open — archived datasets, machine-checked Lean formalisations, and the index running daily at asri.dissensus.ai. Grants work at project or programme level with milestones agreed up front; academic partnership runs as joint proposals, co-authorship and shared infrastructure. Smaller ongoing support goes through Patreon.
Independent by construction
The legal shape of the lab
Dissensus is the research brand of Dissensus Ltd, a company registered in England and Wales (no. 17309927). The independence is the point: the company answers to no institution, no outside owner, and no funder. There is one director, and that is a legal structure rather than a reward — no collaborator or advisor is asked to take a company office. The rules the lab holds itself to are written down above, in public, where they can embarrass us if we break them.
We respond to substantive messages, and “substantive” is doing real work in that sentence. A first reply usually takes weeks rather than days. For applied work or funding, the section above is the fuller picture; to work on the research, start at Join.
Phone: 020 3807 1624 · Email: research@dissensus.ai