History and acknowledgements
The people and the lineage of this work are what the paper’s citations cannot carry: lived experience, and the grey literature in which most of it was first written down. We have it because much of the substantive learning came not from academia but from designing and governing production systems in highly automated, highly adversarial, highly politicized digital ecosystems. The story runs through model-based systems engineering, simulation and measurement, decision analysis, economics and governance, and its thread is the ideas: a design space mapped rather than an optimum presumed, trade-offs made visible, and the choice among alternatives left to accountable people. Many people carried those ideas, and they are named here. The firm they worked in is a contract research and engineering firm, founded in 2017 as BlockScience and now the Dynamical Systems Group. We perform requirements, architecture, design, simulation and analysis for clients, all software-defined, studying the spaces of feasible designs of their systems under uncertain, strategic and adversarial users.
Before the firm, the question was already allocation among many decision makers whose choices affected one another. Zargham’s dissertation, Fast, distributed optimization strategies for resource allocation in networks (2014), advised by Ali Jadbabaie and Alejandro Ribeiro at the University of Pennsylvania, took it up as distributed convex optimization; Accelerated dual descent (2014), with Ribeiro, Asuman Ozdaglar and Jadbabaie, was its form for network flows, and with Victor Preciado came geometric programs, posynomial like this paper’s frontier, for allocating protection to networks (2014). Reza Olfati-Saber’s Consensus and cooperation in networked multi-agent systems (2007), published while Zargham was his undergraduate student, and a routing rule of the same years, discounted integral priority (2014), are the foundations conviction voting later rests on. With Preciado and Zixuan Zhang came the 2020 journal version of the 2018 state-space framework.
cadCAD is the work of 19 contributors; its AUTHORS file (2018) credits, among others, Joshua Jodesty, Michael Zargham, Markus Koch, Matthew Barlin, Danilo Bernardineli and Emanuel Lima, and it was introduced in a post (2019).
Conviction voting began as Sensor networks and social choice (2018), a proposal written at Frank Wang’s request, which turned the routing rule into a continuous decision rule: a community’s preferences fused over time as a sensor network fuses its measurements. Prototyped in simulation, built into the Commons Stack with Jeff Emmett and Griff Green (2019) and modelled in cadCAD (2020), it has allocated one community’s funds since 2020 and three others’ after it; a brief history (2020) names the people and the ideas it drew on. It was the substance of computer-aided governance (2019), by Jeff Emmett and Zargham, which named Alvin Roth’s market design and kept humans in charge, informed by models. That is the start of a governance lineage that runs from presuming a unique optimum to mapping a design space, where the choice among competing alternatives is politics, not optimization. Exploring subjectivity in algorithms (2019), written for SourceCred, showed the choices inside PageRank that make a ranking a judgement: the mixing model, the seed and the damping. Algorithms as policy (2020), by Zargham and Kelsie Nabben, took the rules that algorithms set on online platforms as policy-making, with a duty to the public interest. Eric Alston, a longstanding member of the team, argued in Governance as conflict (2022) that disagreement among the members of any organisation is structural, so an institution must support conflict and resolve it rather than model it away. A paper on organisations run partly by code (2022), by Zargham and Nabben, read them as viable systems in Beer’s sense, where the set of rules people can change must be large enough to keep the purpose reachable and no larger, with the community that first ran conviction voting as its case. The two met in What constitutes a constitution? (2023), by Zargham, Alston, Nabben and Ilan Ben-Meir, and in Path dependence (2024), by Alston, Amber Case and Zargham, on the order in which an institution can reach its ideal.
The Balancer whitepaper (2019), by Fernando Martinelli and Nikolai Mushegian, built a market that sets its prices by a rule, holding a weighted product of balances constant, applying the invariant approach of the state-space framework; Martinelli was a BlockScience team member at the time of his first proofs. Configuration spaces (2019), by Zargham, Jamsheed Shorish and Kris Paruch, named the states a mechanism can reach by the invariants it conserves; economic games as estimators (2020) read such mechanisms as signal processors; and a practical theory of fungibility (2021), by Shorish, Matt Stephenson and Zargham, set the context for machine-learning methods and markets without yet giving a coordinate system suited to such markets. What carried over is the practice of reasoning with product invariants.
Jamsheed Shorish co-wrote Generalized Dynamical Systems (2022) and its video series; he and Kris Paruch wrote the parameter selection article (2021). Andrew “Octopus” Penland led workshops on MAIDs, GDS and cadCAD under a contract with BlockScience (2024). Rohan Mehta wrote gds-core and gds-axelrod (2026), and gds-core credits Sean McOwen’s MSML (2023). Zeke Medley wrote Block diagrams of dynamical systems (2024), a write-up and a Julia executor, in a summer internship on block diagrams as a compositional algebra and the formal methods associated with them; it credits Carolina Riascos-Álvarez’s consensus-network work in the write-up, and Danilo Bernardineli reviewed it. That work on block diagrams and MSML was subsumed when the firm adopted the OMG standard SysML v2 and the OpenMBEE implementations of it, among them OpenSysML, in which this paper’s cubesat model is written. Andrew Clark wrote with Alexander Mihailov and Zargham a community-currency model (2023), developed with the Red Cross and Grassroots Economics, which applied related methods in modeling, simulation and analysis.
Market design lessons for safety-first institutional engineering (2023), by Jamsheed Shorish with Zargham and Ilan Ben-Meir, argued that an institution secures its safety requirements before it optimises for performance, and weighs qualitative goals against quantitative ones, drawing on Roth’s market design. Knowledge networks and the politics of protocols (2024), by Zargham, Ben-Meir and Kelsie Nabben, took tools that let people run and govern their own instances as a new commons whose design is never neutral, since it structures the action space. Ilan Ben-Meir’s An algorithm is a camera (2023) made the point that a model is a framed view of a reality that exceeds it, not a view from nowhere. Engineering legitimacy (2024), by Ben-Meir and Zargham, took legitimacy as a sensed quality and modelled liberty, equality and agency as trade-off curves between competing readings of each, so that an institution can be tuned to stay legitimate to its stakeholders; it stood beside a position paper (2024) for a client’s governance on building an accountable, cost-conscious network of expert service providers, co-developed with the public and domain experts. Min-maxing complexity (2024), by Bernardineli, Ben-Meir, Zargham and Jakob Hackel, set three principles for choosing a mechanism, variety, legibility and safety, as complex as the behaviour it must encode and as simple as its stakeholders can read, and applied them to the trade study above. Protocols and institutions (2025), by Zargham and Ben-Meir, defined institutions by their stability in Lyapunov’s sense and set protocols as the engineered part of them, formalising an argument first made in From Nash to Lyapunov (2018). The politics of progress (2025) read two popular visions of technology and society and found that neither acknowledged trade-offs: abundance needs institutions that make them visible and negotiate shared constraints. With Kelsie Nabben and Ellie Rennie, whose ethnography of the CredSperiment (2023) had followed a community paying its contributors by an algorithm’s scores, that lineage continued in computer-aided ethnography (2023) and Building the Loop (2025), on artificial organisational intelligence.
The ADCS lifecycle demo (2026) takes one geosynchronous satellite’s attitude control and one requirement through evidence of several kinds, symbolic, simulated and model-checked, to a person’s judgement on the record: a tool says pass, a person says satisfied. It was presented at the JAXA MBSE/MCE Workshop in June 2026 and to the Open Source Architects community in October 2026. Formalizing document assurance (INCOSE, 2026) set the posture this project takes toward AI assistance: verification by machine against a specification, validation of fitness by a named person, and assurance only where both close. Derivations, not just simulations (SciPy, 2026) showed derivations scripted in SymPy, with the identities as test assertions run in continuous integration; the lemma scripts of this repository follow that pattern. Governed compute, end to end (2026), led by Sayer Tindall, took algorithms as policy into the firm’s infrastructure, with every decision surface explicit so that its governance is informed and accountable: a run opens on authorization, advances only when it conforms and is fit for its purpose, and its evidence is a byproduct of the work.
Zargham serves on the NumFOCUS Advisory Council and is a board member and research director of the Metagovernance Project.
Jakob Hackel and Carolina Riascos-Álvarez worked on client projects we cannot name, as did others we thank but cannot describe here. Some client work is public and used these methods: for an algorithmic debt market, the design record of its feedback controller, written with its developers before launch (2021), then Zargham’s talks on the same system after years in production, at PyData NYC (2023) and at a workshop in Edinburgh (ICMS), October 2024, where models served the configuration management of the fielded system; for the governance of a file storage and retrieval network, an independent review of two competing proposals and a study of its consensus pledge (2023); and for a client’s engineering team, a document-based trade study of two candidate protocols against stated technical and economic criteria (2023).
The learning came from systems that were running, automated, under attack and contested, where no single objective exists and disagreement is structural. That is why this work maps a frontier instead of solving for an optimum: it states what is possible and what trades against what, in a form a machine can check, and leaves what is preferable to the people accountable for it. This work is critical for humans to participate meaningfully in decision-making in an era when machines mediate more decisions, and faster ones: Meltdown (2025) says why the bottleneck has moved from generation to judgment.