Dissensus started this year and is still taking shape. There are two researchers, an operations associate and an affiliate, twenty papers, six open problems, and a company incorporated this July — which means very little here is settled. Someone joining now is not filling a slot in something already built; they get a say in what the lab works on, how it decides a result is finished, and what it becomes over the next few years.
Three rungs, each entered by doing the previous one
What that looks like in practice: research roles, entered by doing research. The ladder below is short on purpose — it starts at a task you could finish in a weekend and ends at a standing affiliation and a title you can put on a CV, and each rung is reached by doing the previous one rather than by applying. There is no funding yet, so none of it is paid, and the limits are set out plainly further down rather than buried. Everything else — which problems matter, which methods the lab trusts, what it refuses to do — is genuinely open, and more so now than it will be later.
I
Contributor
Open
One named task. Five to twenty hours, once.
You take one specified gap from the open projects and close it. Blind-code a table, run one inference pass, rate a sample for severity, sweep a literature claim. The task is defined before you start and finished when the output exists; there is no ongoing obligation in either direction.
What you get
Named acknowledgement in the paper, or co-authorship where the contribution is substantive by the standard on this page
A dated written record of what you did, which is the thing that is actually useful when someone later asks
The output stays yours to talk about, publish alongside, or cite
How to enterPick a task below and say which one. No CV needed.
II
Research Collaborator
Open · 6 projects
One project, carried to a result. Weeks to months.
You own a gap on a named project through to a publishable result rather than to a deliverable. That means the study gets designed, run, written up and defended, and if it fails it gets reported as a failure under the same inference standards a positive result would get.
What you get
Co-authorship, with author order agreed in writing before the work starts rather than after the results are in
Listed by name on the project record here, with what you own stated plainly
Standing to describe the work as done in collaboration with Dissensus
First refusal on the paid version of the role if the project gets funded
How to enterName the project and the gap you would take. Say what you would bring and what you want out of it.
III
Research Affiliate
By agreement
Ongoing, cross-project, by mutual agreement. No hour quota.
The standing version of the above: you work across the programme rather than on one paper, you are consulted on design before things are built, and you are listed as part of the lab. It is not a job and it is not paid. It is entered after you have delivered something, because a title handed out before any work has happened is worth nothing to the holder.
What you get
The title Research Affiliate, Dissensus — usable on a CV, ORCID record, LinkedIn profile, or a paper's affiliation line
A signed confirmation letter on company letterhead stating the role and its dates, which is the artefact institutions and visa processes actually ask for
A dissensus.ai email alias, and listing on the About page
Standing co-authorship eligibility across the programme, and first refusal on paid roles when there is money
An affiliation for grant and fellowship applications where you need a host that is not a university
How to enterNot applied for. It follows from work already done together.
02 · Start here
Tasks, listed at the size they actually are
Every task below is a real gap in a real project, described in the project's own words on the Projects page. The hour figures are the estimates written into those project notes, not sales copy. People say yes to a weekend and no to an open-ended collaboration, so the tasks are listed at the size they actually are — and the fastest way to end up shaping the lab is to close one of them and then disagree with us about what comes next.
EffortA weekend, ×2 coders
Blind-code the focusing-event table with the outcome column stripped
A reportable inter-coder agreement statistic, replacing one author's self-coded consistency check
This is the test the paper itself names as the real one. It has never been run.
Fits: Public-policy or regulatory-studies researcher who works with focusing-event or agenda-setting theoryThe Loaded Wheel →
EffortFifteen minutes
Endorse an arXiv submission in econ.GN or cs.CY
Unblocks an arXiv deposit that is drafted and unsent; the paper currently has SSRN as its only public surface
The smallest useful thing anyone can do for this lab.
Fits: Anyone already holding endorsement rights in econ.GN or cs.CYThe Loaded Wheel →
EffortA few hours
Pressure-test the mandate-boundary claims against the Bank Service Company Act, the DORA critical-third-party regime, and the UK CTP designations
Either confirms or breaks the structural claim that these regimes are service-scoped and never model-scoped
This is the sentence a referee will go after first.
Fits: Financial-regulation lawyer or a supervisor who has worked inside one of these regimesThe Loaded Wheel →
EffortOne inference pass each — cheap in wall-clock, expensive only in access
Run the three queued controls: identical question with varied persona, an equal-cardinality set, a paraphrase set
Separates the persona-specific share of the drift; the identical-question set alone decides whether the headline survives
Data and code are already public, so the claim can be audited before you commit. Expect to edit hardcoded paths and a local inference endpoint.
Design a codebook, train multiple raters, and report a real reliability statistic on a stratified subset of the 215 responses
Converts the result from 'genre changes the vocabulary models emit' to 'genre changes what models do'
The binding gap on this project. The current outcome variable is a fifteen-term keyword matcher measuring safety register, not refusal. Nothing else about the project matters more.
Fits: Measurement or psychometrics researcher who has run annotation properlyGenre mimicry →
EffortFour to six hours
Independent severity rating across the prompt battery
Separates genre from harmfulness — violence prompts and finance prompts differ in more than genre, and that confound is currently unaddressed
Small, and it closes an obvious referee objection.
Fits: Anyone comfortable rating harm severity consistently against a written scaleGenre mimicry →
EffortCompute-bound rather than time-bound
Run the agent against an established security benchmark suite and report a success rate with intervals
Replaces one demonstration run — an anecdote — with a measured success rate
The central claim has never been tested. Compute is the real constraint here.
Fits: Someone who evaluates agentic systems empirically; equally, anyone who can host the evaluationAutonomous red team →
EffortAbout an hour
Re-deposit the current 22-page revision so the DOI resolves to the paper that exists
Anyone citing the DOI today cites the earlier single-author version under the old title
Not a collaboration. It is on this list so the list is complete.
Fits: Lab-internal admin — listed because it is blocking and honest to admitAutonomous red team →
EffortWeeks
Assemble the country-by-domain-by-year panel and estimate the first hypothesis with fixed effects and cluster-robust inference
The hypotheses are currently untested — there is no panel dataset, no regression code and no results, only a research design
On offer: a fully specified design, a machine-checked formal core, and co-authorship on work that has not yet been done. Report it if it fails.
Fits: Comparative political scientist or political economist hands-on with V-Dem, ACLED or GDELT event data and union-density seriesConsensual Sovereignty →
EffortOne to two weekends
Independently code the suffrage and labour cases against the published ordinal protocol
A reliability statistic instead of an assurance — every ordinal coding is currently by one person
The pilot study the paper names as the precondition for a full panel has not been run.
Fits: Historian or historical-institutionalist comfortable with a written coding protocolConsensual Sovereignty →
EffortAn afternoon
Verified literature sweep across the nationalisation-of-AGI debate against the outline's two novelty claims
Either sharpens the contribution or ends the project early, which is a good outcome either way
Grunt work nobody has done. It matters as much as anything else on this project.
EffortOngoing — this is a co-authorship, not a task
Argue the state-ownership position at full strength, as a co-author rather than as a foil
The only argument map that exists was built to win a debate for the private position; one author who has already argued one side cannot supply the other
There is no draft yet. Zero words of manuscript prose exist. Joining now means shaping what the paper argues.
Up front, because finding it out later wastes more of your time than reading it now. Some of these are constraints that change with funding; the rest are choices, and they are marked.
No salary, stipend, hourly rate or honorarium, for any role on this page. This is a funding problem rather than a policy, and the first two tiers carry first refusal on the paid version of the role if and when that changes.
No compute grant, data budget, travel money, publication-fee cover or equipment. Where a task needs inference budget, that is named as part of the task, because the person who has it is the person who can do it.
No unpaid operations, communications, administration, fundraising or board work. Those are jobs, and the lab does not have the money to hire for them, so it does not advertise them as volunteering.
No directorship, company officer position or Companies House filing for any collaborator or advisor. Dissensus Ltd has one director and that is a legal structure, not a reward — having a say in the research does not and should not require taking on a statutory role.
No visa sponsorship, and no claim that an affiliation here substitutes for an institutional post. If a paid position or an institutional host is what you need, this is the wrong lab, and that is a reasonable conclusion to reach.
04 · Organisations
Applied work and partnership
Commercial engagements are the one place money changes hands, and they are deliberately separate from everything above: a client owns the deliverable, whereas funding the programme produces public goods. For friction analysis, systemic-risk indices, formal verification in Lean 4, or adversarial evaluation of multi-agent systems, email research@dissensus.ai with the problem as you currently understand it, or book fifteen minutes. If we are not the right lab, we will say so.
Partnership does not have to mean money. If your group, lab, reading circle, student society or journal is working on questions that overlap ours — coordination and its overhead, alignment under competing objectives, legitimacy and consent, systemic risk, or the methodology of checking any of it — the useful move is usually the 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. Dissensus already runs one such arrangement with the Sineidocracia movement, where the lab supplies technical infrastructure and gets a genuinely different intellectual tradition to argue with. We are looking for more of those, including with groups whose conclusions we expect to disagree with — that is closer to the point than agreement is.
The lab is also looking for external oversight: people or bodies willing to audit its methods, sit in an advisory capacity, or check results it has a stake in believing. That is not a formality. A lab which 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 its own internal scrutiny for months. Current partnerships and acknowledgements are listed on the About page.
Not every useful conversation should have to start with an email and a proposal. The lab runs an open Discord where the papers get argued about before they are finished, drafts go up for people to pick holes in, and open problems sit in a channel rather than in someone’s head. Joining commits you to nothing and there is no application — lurking is a legitimate use of it, and several of the tasks above were first raised there by people who had no formal role at all.
Say which task or which project, what you would bring, and what you want out of it. Concrete beats enthusiastic. If nothing on this page fits, write anyway — the listed tasks are the gaps the lab happens to have noticed, not the boundary of what it will work on. Anything touching complex adaptive systems under competing objectives is worth a message: multi-agent alignment, market microstructure, formal verification, political economy, or a field none of us has thought to look at yet. So is any capacity, from one afternoon to a standing collaboration. A first reply usually takes weeks rather than days — the lab is one researcher and a research assistant, both with papers of their own in flight — and a reminder after a month is a courtesy rather than a nuisance.