mindmap
root((Polycentric governance))
Classical strengths
Local knowledge exploitation
Redundancy and resilience
Experimentation across centres
Scale-matching to resource
Design principles (Ostrom)
Boundaries
Collective choice
Monitoring
Graduated sanctions
Conflict resolution
Nested enterprises
Blind spots
Which rules — structural
Who decides — institutional
When to act — temporal: BDPD3 binding axis
Cascading-tipping parallel
Two-phase cascade
Leading-following structure
Temporal delay as mechanism
3 Lecture 3 — Decentralize Everything
Polycentric governance and cascades
3.1 Opening dissonance
Lecture 2 showed that group size alone is not the binding constraint on collective action — institutions swamp size. This lecture asks a follow-up question: which institutions, and under what conditions? The answer, as the polycentric governance literature has developed over half a century, is that multiple overlapping decision centres can outperform both centralisation and decentralisation. But the answer is incomplete in a way that matters — and the missing piece is not about institutions. It is about time.
In 1968 Garrett Hardin published a parable that would become the most cited argument in environmental governance (Hardin 1968). A shared pasture, open to all herders. Each herder benefits fully from adding one more animal but bears only a fraction of the cost of overgrazing. The result is inevitable: each rational herder adds animals until the pasture collapses. Hardin called it the tragedy of the commons and concluded that the commons required either privatisation or a centralised Leviathan — some external authority strong enough to restrict access. The metaphor entered policy orthodoxy: shared resources, left to themselves, destroy themselves.
Hardin’s argument was deliberately stark. He drew on Malthus, on the logic of collective action, on the impossibility of maximising for two variables at once. His conclusion was that “the population problem has no technical solution; it requires a fundamental extension in morality” — and that shared resources, without external enforcement, would always be over-exploited. The parable of the pasture was a thought experiment, not an empirical study, but it was taken as one. For two decades, Hardin’s framing dominated: the commons was a problem to be solved by either the state or the market. The policy implications were immediate: if shared resources destroy themselves, then either assign property rights (so each owner has an incentive to conserve) or install a regulator strong enough to restrict access. The metaphor shaped environmental law, resource management, and international negotiations. It also, crucially, assumed that the problem was one of incentives, not of information.
Twenty-two years later Elinor Ostrom published Governing the Commons (Ostrom 1990) and dismantled Hardin’s dichotomy with an empirical sledgehammer. She documented hundreds of cases — irrigation systems in Spain and the Philippines, fisheries in Turkey and Maine, forests in Japan and Switzerland — in which communities of resource users had governed their own commons for centuries without either privatisation or state control. The Spanish huertas, for instance, had maintained elaborate water-allocation rules for over five hundred years, with elected officials, graduated sanctions, and conflict-resolution mechanisms that predated modern regulatory theory by centuries. The Lobster gangs of Maine had developed informal property-rights systems that were more effective than any formal law at preventing overharvesting.
From this record Ostrom distilled eight design principles for long-enduring common-pool resource institutions: clearly defined boundaries, rules adapted to local conditions, collective-choice arrangements, monitoring, graduated sanctions, conflict-resolution mechanisms, recognition of the right to organise, and nested enterprises for resources that span multiple jurisdictions. The principles were descriptive, not prescriptive — Ostrom derived them from the record of what had worked, not from theory about what should work. But they carried an implicit claim that the institutional structure was what mattered: get the rules right, and the commons survives.
The claim was powerful enough to reshape the field. It entered the policy literature with the broader message that local-level institutions could, under the right conditions, manage commons that traditional theory said required either privatisation or state regulation. The design principles became a diagnostic toolkit: analysts could ask whether a given governance arrangement had the right institutional features, and predict success or failure accordingly. The toolkit was widely adopted: community-based natural resource management programmes in developing countries, co-management arrangements in fisheries, watershed governance in the European Union — all drew, explicitly or implicitly, on Ostrom’s framework. What the toolkit did not include — because the empirical settings Ostrom studied did not require it — was a temporal dimension. The irrigation communities monitored water levels; the fisheries monitored catch. In every case, the signal was the resource itself — visible, shared, moving at the speed of use. The leading/lagging distinction was invisible because the systems did not have the ODE structure that makes it salient.
The eighth principle — nested enterprises — became the seed of what Ostrom would later call polycentric governance: a system in which multiple centres of decision-making, each operating with some degree of autonomy, interact within an overarching set of rules (Ostrom 2010). The concept was not new to political science — Vincent Ostrom, Charles Tiebout and Robert Warren had introduced it in 1961 for metropolitan governance — but Elinor Ostrom’s life work gave it empirical teeth. Polycentric systems, she argued, outperformed both pure centralisation and pure decentralisation because they could exploit local knowledge while retaining the capacity for coordination across scales.
It is a beautiful argument. It is also, as BDPD3 will show, incomplete in a way that matters. The missing piece is not about institutions. It is about time.
The question this lecture asks is not whether polycentric governance works — the empirical record says it does — but whether the conditions under which it works have been fully specified. The classical literature has specified the institutional conditions. BDPD3 identifies a temporal condition that the classical literature, for entirely understandable reasons, did not address. The two conditions are complementary, not competing: a polycentric system needs both the right institutional structure and the right monitoring signal. The novelty is in the second.
3.2 The classical setting
Before we walk through the empirical record, we need to understand two intellectual traditions that have developed largely in parallel and that, to our knowledge, have rarely been brought into a single experimental frame. The first is the institutional tradition that runs from Hardin through Ostrom to the contemporary polycentric governance literature. The second is the Earth-system tradition that runs from the tipping-point concept through cascading dynamics to the leading-following taxonomy. Their intersection is the governance question that BDPD3 puts on the table.
3.2.1 The Hobbes-Ostrom polarity
The governance-of-the-commons literature has been structured, since at least 1968, around a polarity. On one side, Hobbes and Hardin: without an external enforcer, covenants are “but words” and shared resources are destroyed. On the other side, Ostrom and the institutionalists: communities can and do self-govern, provided the right institutional conditions are in place. The polarity is productive — it has generated decades of experimental and field research — but it shares an assumption that both sides treat as background rather than as object of study: the question is always which rules and who decides, never what signal triggers the decision.
Ostrom’s design principles are a catalogue of institutional features: boundaries, rules, monitoring, sanctions. They describe the structure of a governance arrangement. Carlisle and Gruby’s theoretical model of polycentric governance systems (Carlisle and Gruby 2019) formalises this as attributes (definitional elements of polycentricity) and enabling conditions (institutional features for achieving functionality in the commons). Their model is precise: polycentricity requires multiple centres of semiautonomous decision-making that take each other into account in competitive and cooperative relationships and have recourse to conflict-resolution mechanisms. But even this precision is structural — it tells you what the system is, not when it acts.
Heikkila and colleagues, introducing a special issue on polycentric environmental governance, define the concept as “multiple overlapping centers of decision-making, which interact within an overarching set of rules” and identify the principal advantage as the capacity to “manage cross-scale environmental issues and address the complex interrelationships within our social and environmental systems” (Heikkila et al. 2018). Morrison and colleagues, studying power asymmetries in polycentric systems, add that even well-designed instruments can be undermined by unequal power relations between decision centres — the “black box of power” that the polycentric literature has been slow to open (Morrison et al. 2019). Dorsch and Flachsland, analysing polycentric climate governance through an Ostromean lens, identify four key mechanisms — self-organisation, site-specificity, experimentation, and trust-building — through which polycentric arrangements enhance climate mitigation (Dorsch and Flachsland 2017).
The entire literature asks: which instruments? Which institutional arrangements? Which rules? It has said comparatively little about the temporal structure of the signal that a regulator should act upon.
3.2.2 The cascading-tipping parallel
In parallel, the past decade has seen a sharp rise in interest in cascading tipping points in coupled environmental systems. Lenton and colleagues, in a high-profile comment in Nature, argued that climate tipping points — the loss of the West Antarctic ice sheet, the dieback of the Amazon rainforest, the collapse of the Atlantic thermohaline circulation — are interconnected and could trigger each other in domino-like sequences (Lenton et al. 2019). Their framing was urgent: the consideration of tipping points, they wrote, “helps to define that we are in a climate emergency”. Armstrong McKay and colleagues, in a 2022 Science review, exceeded this framing by cataloguing the interactions among multiple tipping elements and showing that cascading transitions are not hypothetical but structurally embedded in the Earth system (Armstrong McKay et al. 2022). Wunderling and colleagues, in a 2024 review, synthesised the growing literature on tipping cascades and identified the key challenge as understanding propagation — how one subsystem crossing a threshold destabilises others linked to it (Wunderling et al. 2024).
Klose and colleagues formalised the dynamical anatomy of these cascades, distinguishing two-phase, domino and joint patterns according to how the critical transition propagates from one element to the next (Klose et al. 2021). In a two-phase cascade, the leader collapses first, and the follower collapses only after a delay — the delay being the structural feature that makes governance possible if the signal is read early enough. In a domino cascade, the follower collapses immediately upon the leader’s transition. In a joint cascade, both collapse simultaneously. They also characterised the leading-following structure of unidirectionally coupled subsystems, where one component drives another — the structure that will turn out to be central to our BDPD story. The taxonomy provided the conceptual vocabulary — leading, following, two-phase cascade — that BDPD3 would later operationalise in a governance context.
What neither literature has done is connect these two streams explicitly. The polycentric governance literature asks which rules. The cascading-tipping literature asks which thresholds. Neither asks: when one jurisdiction’s pollution cascades into its neighbours, which signal inside the emitter should the regulator act upon? The two literatures have developed in separate journals, separate conferences, and separate intellectual traditions — the first in political science and institutional economics, the second in Earth-system science and applied mathematics. Their intersection is the governance question that BDPD3 puts on the table: if cascades propagate through temporal delays, and if polycentric governance systems must monitor the right variable at the right time, then the design of monitoring infrastructure is not just an institutional question but a dynamical one.
3.3 Sixty years of the polycentric argument
The temptation in a lecture like this one is to present the polycentric governance literature as a flat list of design principles and enabling conditions. We will not do that. We will walk through the key contributions in roughly the order they were made, and at each one we will ask the same question — this study assumed the governance system monitors the right variable; what if it doesn’t? — because this is the question that BDPD3 will eventually answer. Even before the answer arrives, the question changes how we read the classical evidence.
3.3.1 The Hardin provocation (1968)
Hardin’s tragedy-of-the-commons parable (Hardin 1968) was not, in its original form, about empirical pastures. It was about the logic of collective action in a finite world: population growth, resource depletion, the impossibility of maximising for two variables at once. The parable of the shared pasture was a thought experiment — a “no technical solution problem” in which rational individual behaviour led inexorably to collective ruin. Hardin’s policy conclusion was stark: the commons required either privatisation (“mutual coercion, mutually agreed upon”) or a centralised authority capable of restricting access.
What Hardin assumed — and what made the parable so powerful — was that the herders could see the pasture. The signal was the grass itself: visible, shared, degrading under everyone’s feet. The tragedy was not a monitoring failure; it was an incentive failure. Each herder saw the same overgrazed pasture and chose to add one more animal anyway, because the private benefit exceeded the private cost. The information was there. The will to act on it was not.
This assumption — that the relevant signal is visible and shared — would survive unexamined through the entire polycentric governance literature. Ostrom’s design principles include monitoring (Principle 4), but the literature has treated monitoring as an institutional feature (someone watches) rather than a temporal question (what they watch, and when it moves). Hardin’s herders had no need to distinguish leading from lagging indicators because the pasture itself was the indicator, and it moved at the speed of grazing. But not all shared resources are pastures. When the resource dynamics involve multiplicative feedback — when the damage variable is causally downstream of the cause variable with a structural delay — the visible signal may lag the invisible one, and the herders may be watching the wrong thing. The question BDPD3 will eventually put on the table — what if the agents can see the resource but not the cause of its decline? — is already latent in Hardin’s assumption.
3.3.2 The Ostrom empirical revolution (1990)
Ostrom’s Governing the Commons (Ostrom 1990) was not a theoretical argument. It was an empirical one. She documented case after case — centuries-old irrigation communities in Spain (huertas), fisheries in Turkey and Maine, forests in Japan and Switzerland — in which communities of resource users had governed their own commons without either privatisation or state control. The Spanish huertas had maintained elaborate water-allocation rules for over five hundred years, with elected officials, graduated sanctions, and conflict-resolution mechanisms that predated modern regulatory theory by centuries. Acheson’s study of the Maine lobster gangs — referenced in Ostrom’s bibliography — documented informal property-rights systems that were more effective than any formal law at preventing overharvesting. The Japanese forest commons (iriai) had sustained timber harvests for centuries through community-enforced rules that adapted to local ecological conditions.
The design principles were descriptive, not prescriptive — Ostrom derived them from the record of what had worked, not from theory about what should work. But they carried an implicit claim that the institutional structure was what mattered: get the rules right, and the commons survives. The claim was powerful enough to reshape the field. It entered the policy literature with the broader message that local-level institutions could, under the right conditions, manage commons that traditional theory said required either privatisation or state regulation. The design principles became a diagnostic toolkit: analysts could ask whether a given governance arrangement had the right institutional features, and predict success or failure accordingly.
Ostrom’s empirical work was also a methodological contribution. She demonstrated that the commons could be studied scientifically — not just as a philosophical puzzle or a policy aspiration, but as a set of institutional arrangements with measurable features and testable outcomes. Her laboratory (the Workshop in Political Theory and Policy Analysis at Indiana University) became the institutional home for a generation of commons scholars who would extend the framework to fisheries, forests, irrigation, and climate governance. The claim was powerful enough to reshape the field. It also, for entirely understandable reasons, said nothing about the temporal structure of the signals those rules acted upon. The irrigation communities Ostrom studied monitored water levels; the fisheries monitored catch. In every case, the signal was the resource itself — visible, shared, moving at the speed of use. The leading/lagging distinction was invisible because the systems Ostrom studied did not have the ODE structure that makes it salient. Every community in Ostrom’s empirical record was human — the monitoring was performed by people with shared cognitive architectures for interpreting the signals. What happens when the monitoring is performed by agents with different architectures, or when the signal they monitor is not the resource itself but a derived variable with a structural delay, is the question the BDPD platform is designed to ask.
3.3.3 The polycentric concept (Ostrom 2010)
In her 2010 Nobel lecture, published as Beyond Markets and States: Polycentric Governance of Complex Economic Systems (Ostrom 2010), Ostrom drew on her half-century of research to articulate the polycentric vision in its mature form. The concept had originated with Vincent Ostrom, Charles Tiebout and Robert Warren’s 1961 work on metropolitan governance — the observation that urban areas served by multiple overlapping jurisdictions could outperform a single centralised authority. Elinor Ostrom extended this insight to natural resource governance, arguing that polycentric systems — multiple centres of decision-making, each with some autonomy, interacting within an overarching set of rules — could exploit local knowledge while retaining coordination capacity.
The polycentric vision was explicitly anti-dichotomous: neither the market nor the state, but a third way in which governance emerged from the interaction of multiple semi-autonomous centres. The advantages were real: redundancy (if one centre fails, others continue), experimentation (different centres can try different rules), adaptation (local knowledge informs local rules), and scale-matching (governance units can be nested to match the scale of the resource system). The disadvantages — coordination costs, free-riding between centres, power asymmetries — were acknowledged but treated as manageable under the right institutional conditions.
What the polycentric vision assumed, as every contribution in this lineage had assumed, was that the governance centres were monitoring the right variables. Ostrom’s Principle 4 (monitoring) specified that monitors should be accountable to the users and that monitoring should be carried out by the users themselves or by agents accountable to them. But the principle was about who monitors, not what they monitor. The temporal position of the monitored variable — whether it was leading or lagging the process that must be governed — was not on the agenda, because the empirical settings Ostrom studied did not require it. The polycentric vision also assumed that the centres could see each other’s actions and respond — the mutual-accountability condition that Carlisle and Gruby would later formalise. What happens when the centres are algorithmic agents that process information differently, or when the externality they must monitor is invisible to them, is the question BDPD3 puts on the table.
3.3.4 The formalisation wave (2017–2019)
Three contributions in the late 2019s formalised the polycentric concept for environmental governance. Carlisle and Gruby (Carlisle and Gruby 2019) developed a theoretical model distinguishing attributes (definitional elements: multiple semiautonomous decision centres, overlapping jurisdiction, conflict-resolution capacity) from enabling conditions (additional features for achieving functionality: trust, learning, redundancy). Their model was the most precise formalisation of polycentricity available, but it remained structural — it specified what the system is made of, not when it acts.
Heikkila and colleagues (Heikkila et al. 2018) introduced a special issue on polycentric environmental governance, defining the concept and identifying its principal advantage as the capacity to manage cross-scale environmental issues. They noted that polycentric systems “appear to be widespread globally — across multiple forms of government and political systems, across a range of environmental issues, and across different geographic locations” — and that both empirical studies and practical evidence suggested “substantial diversity exists in the design and function of polycentric governance systems”.
Morrison and colleagues (Morrison et al. 2019) opened the “black box of power” in polycentric systems, showing that power asymmetries between decision centres could undermine even well-designed instruments. Their contribution was a corrective to the sometimes-optimistic tone of the polycentric literature: institutional structure alone was not sufficient if power relations between centres were unequal.
Dorsch and Flachsland (Dorsch and Flachsland 2017) applied the polycentric lens specifically to climate governance, identifying four key mechanisms — self-organisation, site-specificity, experimentation, and trust-building — and arguing that the emerging global climate governance architecture was already polycentric in structure, even if not always by design. Their analysis was prescriptive where the others were descriptive: they identified what a polycentric climate regime should look like.
Each of these contributions refined the institutional structure of polycentric governance. None of them addressed the temporal structure of the signal that the governance system acts upon. The reason is consistent across the lineage: the empirical settings they studied — fisheries, irrigation, climate policy, metropolitan governance — involved resources whose relevant signals (catch, water level, emissions, service quality) moved at the speed of use. The leading/lagging distinction, which becomes salient only when the ODE structure creates a multiplicative delay between cause and symptom, was not a feature of those settings.
The formalisation wave also left a methodological gap. The polycentric governance literature is primarily qualitative and case-based: it identifies institutional features through comparative case studies, not through controlled experiments. The BDPD platform offers a different methodology — deterministic agent-based simulation in which the governance question can be operationalised as a controlled comparison between trigger signals, holding instruments constant. BDPD3 exploits this methodology to isolate the temporal dimension that the case-study literature could not address. The formalisations also assumed, implicitly, that the governance actors share a common way of interpreting the monitored signals — an assumption that holds when all actors are human communities but breaks when some are algorithmic agents processing information differently.
3.3.5 The cascading-tipping literature (2019–2024)
In parallel, the Earth-system science community was developing a different question: what happens when tipping points in coupled systems trigger each other? Lenton and colleagues (Lenton et al. 2019) argued that climate tipping points were interconnected and could cascade. Their evidence was drawn from paleoclimate records, ice-sheet dynamics, and ocean-circulation models: the West Antarctic ice sheet could destabilise the Greenland ice sheet; the Amazon dieback could accelerate the Atlantic thermohaline collapse. The risk was not isolated tipping but propagation.
Armstrong McKay and colleagues (Armstrong McKay et al. 2022) catalogued the interactions among tipping elements systematically, showing that cascading transitions were not hypothetical but structurally embedded in the Earth system. Wunderling and colleagues (Wunderling et al. 2024) synthesised the growing literature and identified the key challenge as understanding how one subsystem crossing a threshold destabilises others linked to it — the mechanism of propagation, not just the fact of it.
Klose and colleagues (Klose et al. 2021) formalised the dynamical anatomy, distinguishing two-phase, domino and joint cascade patterns and characterising the leading-following structure of unidirectionally coupled subsystems. In a two-phase cascade, the leader collapses first, and the follower collapses only after a delay — the delay being the structural feature that makes governance possible if the signal is read early enough. In a domino cascade, the follower collapses immediately upon the leader’s transition. In a joint cascade, both collapse simultaneously. The taxonomy provided the conceptual vocabulary — leading, following, two-phase cascade — that BDPD3 would later operationalise in a governance context.
The cascading-tipping literature did not engage with polycentric governance. Its object was the Earth system, not institutional design. Its policy prescriptions were about emissions reduction and risk assessment, not about which signal a regulator should monitor. But its finding that cascades propagate through temporal delays between coupled subsystems — that the follower collapses only after the leader has already committed — maps directly onto the governance question BDPD3 puts on the table: if the cascade is driven by a temporal delay, the governance signal must act before the delay commits, not after. The cascading-tipping literature assumed that the coupled subsystems are physical processes — ice sheets, ocean circulations, ecosystems — not governance arrangements. What happens when the “subsystems” are polycentric governance centres linked by externalities, and the “cascade” is a governance failure propagating across jurisdictions, is the question BDPD3 operationalises.
3.4 What survives of the good-will intuition
The record we have walked through does not refute the good-will intuition.
It bounds it — and, in bounding it, it identifies the conditions that do more work than decentralisation alone.
ImportantWhat survives of the good-will intuition
Polycentric governance works — and not by a small margin. Ostrom’s empirical record is robust: communities can and do self-govern shared resources, and polycentric arrangements that nest local autonomy within broader coordination structures outperform both pure centralisation and pure decentralisation. The design principles are not a recipe but a diagnostic: they identify the conditions under which self-governance succeeds, and the empirical record confirms that those conditions are achievable.
But the success of polycentric governance is conditional, not unconditional. Five concrete moderators emerge from the literature we have walked through:
Boundaries. Ostrom’s first design principle — clearly defined resource boundaries and clearly defined authorised users — is the prerequisite for all the others. Without boundaries, there is no group to govern; without defined membership, there is no accountability. Carlisle and Gruby’s formal model treats boundaries as a definitional attribute of polycentricity: the decision centres must be distinguishable, their jurisdictions must be specified, and the resource units they govern must be identifiable. The empirical record confirms this: the Spanish huertas worked because the irrigation canals had physical endpoints and the water rights were precisely allocated; the fisheries that failed were typically those with open access and no boundary definition.
Monitoring. Ostrom’s fourth principle specifies that monitors should be accountable to the users and that monitoring should be carried out by the users themselves or by agents accountable to them. Morrison and colleagues add that monitoring is only effective when the power relations between centres do not undermine the monitors’ independence. But BDPD3 identifies a further condition that the classical literature did not address: the monitored variable must be leading, not lagging. Monitoring pollution (a lagging indicator) is structurally too late when the ODE creates a multiplicative delay between cause and symptom; monitoring capital (a leading indicator) catches the system before it commits to overshoot. The temporal position of the monitored variable is as important as the institutional arrangement that delivers it.
Sanctions. Ostrom’s fifth principle — graduated sanctions — specifies that violators receive penalties that escalate with the severity and frequency of the violation. Dorsch and Flachsland identify sanctions as a key mechanism through which polycentric arrangements enforce compliance. The empirical record shows that sanctions work best when paired with communication (as Ostrom, Walker and Gardner demonstrated in their “covenants plus an internal sword” treatment) and when the sanctioning institution is itself governed by the users rather than imposed from outside. Morrison and colleagues warn, however, that sanctions can be captured by powerful centres and turned into instruments of domination rather than cooperation.
Conflict resolution. Ostrom’s sixth principle — rapid, low-cost, local conflict-resolution mechanisms — is the institutional lubricant that prevents disputes between centres from escalating into governance failures. Carlisle and Gruby treat conflict-resolution capacity as a necessary enabling condition for polycentric functionality: without it, the multiple centres cannot take each other into account in cooperative relationships. Heikkila and colleagues note that cross-scale environmental issues create inherent conflicts between jurisdictions, and that the capacity to resolve those conflicts locally — without recourse to centralised adjudication — is what distinguishes functional polycentric systems from dysfunctional ones.
Nested enterprises. Ostrom’s eighth principle — governance units nested at multiple scales for resources that span jurisdictions — is the structural feature that distinguishes polycentric governance from mere decentralisation. Dorsch and Flachsland argue that the emerging global climate governance architecture is already polycentric in structure, with multilateral, national, and subnational actors operating at different scales. The advantage of nesting is redundancy: if one level fails, others continue. The disadvantage is coordination cost — the more levels, the harder it is to align them on a common signal.
What the classical literature has not addressed — because the empirical settings it studied did not require it — is the temporal condition: the governance system must act on the right signal at the right time. When externalities cascade across jurisdictional boundaries, the question is not only which rules the centres deploy but which variable they monitor — and whether that variable is leading or lagging the process that must be governed. This is the condition that BDPD3 adds to the list, and it is the one that Lecture 10 will explore in depth.
This is the third of ten occasions on which the course will defend the same moral: rigorous empirical analysis rarely refutes the good-will intuition outright; more often it bounds its validity within a precise perimeter, and the bounds are the interesting object.
3.5 The BDPD angle — empirical: the trigger signal matters more than the instrument
Type of angle: empirical. The findings reported in this section come directly from computational experiments on the BDPD platform using a Bardi/Seneca three-variable ODE substrate. The evidence is not speculative or extrapolative — it is the published result of BDPD3 (Brunelli 2026), a deterministic agent-based experiment in which the governance question is operationalised as a controlled comparison between leading and lagging trigger signals across multiple instruments. The angle is labelled empirical because the claims are falsifiable, the data are reproducible, and the design isolates the causal variable (trigger signal) from confounds (instrument choice) through a 2×2 factorial structure.
3.5.1 The Seneca substrate
The BDPD3 experiment runs on a three-variable ODE following Bardi’s Seneca framework (Bardi 2017; see also Bardi 2018 for the population-collapse intuition that motivates the framework). Each jurisdiction models a resource-driven industrial cycle with three coupled variables: a natural resource R consumed to build industrial capital C, which generates pollution P that degrades both R and C. The characteristic behaviour is the Seneca cliff: capital grows slowly during the boom and collapses rapidly once the resource base is exhausted — an asymmetry that gives the paper its governance question. The three-variable family has been formally characterised in related HANDY-descended systems, and Bardi’s framework has been adopted for pedagogical system-dynamics modelling of resource exploitation.
Three jurisdictions — an emitter A and two downwind arenas B and C — are coupled by directional pollution links through which a fraction of A’s pollution flows into B and C as an externality, without depleting the source. In the terminology of cascading-tipping research, this is a leading-following configuration with A as the leading subsystem and B/C as following subsystems coupled unidirectionally via pollution import. Under the taxonomy of Klose and colleagues, the resulting A→B/C cascade is a two-phase cascade: A’s Seneca cliff propagates to B/C only after A’s resource base has collapsed past the recovery envelope.
A world-level governance meta-agent reads engine-side state and applies one of three levers — a cap on the capital growth rate, a levy on the capital stock, or a fine on households — armed by one of two triggers: pollution P (lagging, hidden) or capital C (leading, visible to players). The lever/trigger orthogonality allows us to isolate the effect of signal timing from instrument choice — the central question of this paper.
3.5.2 The 2×2 result: trigger dominates instrument
The core finding of BDPD3 is a single table. The experiment completes a 2×2 design crossing two instruments (cap, levy) with two triggers (pollution, capital):
| Lever Trigger | P (lagging) | C (leading) |
|---|---|---|
| cap | A collapse t22, B rec 0% (fail) | A survives, B rec +19.7% (win) |
| levy | A collapse t17, B rec 0% (fail) | A survives, B rec +31.6% (win) |
When the trigger is capital, both instruments rescue the emitter and recover the downwind. When the trigger is pollution, both instruments fail. The instrument modulates the outcome (the levy recovers the downwind more strongly than the cap: +31.6% versus +19.7%) but does not change whether the cascade is prevented. The binding axis is the trigger column, not the instrument row.
The mechanism is transparent once the ODE is read. Since \(dP/dt = k_2 \cdot C \cdot P\), pollution is multiplicatively downstream of capital: P can only accelerate after C has already grown large. In the control cell, emitter A’s capital peaks at turn 13; its pollution crosses the reactive threshold only at turn 15. By the time P clears any non-trivial detection threshold, the industrial boom that will emit the rest of the pollutant is already committed. A regulator keyed on P is therefore structurally late, reacting to a symptom whose cause is already behind it. A regulator keyed on C at threshold 0.10 fires at turn 3 — when capital is just beginning its growth phase and pollution is still negligible — throttling growth before the pollution engine engages. The industrial boom never materialises, so the Seneca cliff never arrives, for the emitter or for its downwind neighbours.
This finding is the sharpest challenge to the instrument-centred view of the polycentric governance literature. Carlisle and Gruby’s model specifies what polycentric systems are made of. Morrison and colleagues ask who holds power within them. Dorsch and Flachsland ask which mechanisms make them work. BDPD3 adds a prior question: what are they looking at? A perfectly designed instrument armed on the wrong signal — the lagging symptom rather than the leading cause — is structurally too late, regardless of institutional arrangements.
In Ostrom’s design-principle vocabulary, the finding maps onto Principle 4 (monitoring): effective governance requires not just that someone watches, but that they watch the right variable at the right point in the causal chain. The real-world analogue is the cross-boundary cascading of nutrient pollution: the emitter’s production generates the nutrient load that propagates downstream, and monitoring production intensity rather than ambient nutrient concentration would be the structurally earlier trigger.
The mechanism generalises beyond the Seneca ODE. It arises whenever a governance system must intervene in a process with multiplicative dynamics — where the growth rate of the damage variable is proportional to the stock of the cause variable. In such a system the cause must already be large before the damage can accelerate, so any signal keyed on the damage is structurally delayed, and the same instrument that fails on the lagging signal succeeds on the leading one purely by virtue of when it fires. The leading/lagging distinction is therefore not a property of this particular model but of the class of models in which growth is multiplicative.
Lecture 10 unpacks the mechanism in detail — the monotone timing ladder, the structural irreducibility of the pollution delay, the rate-induced governance boundary — and connects it to the Seneca-collapse literature and the early-warning-signal tradition. The finding presented here is the headline; the dynamical anatomy is theirs.
3.6 Open questions and the bridge to Lecture 4
3.6.1 What does BDPD3 not tell us?
BDPD3 uses an algorithmic governance meta-agent — an omniscient observer that reads engine state directly. The open question for the BDPD series is whether an LLM-based regulator would combine the adaptivity documented in BDPD2 with the temporal discipline of leading-indicator governance — or whether the noise and bounded rationality of language-model agents would erode the timing advantage that makes capital-triggered regulation effective. The companion papers BDPD1 and BDPD2 have documented that LLM agents can be surprisingly adaptive in governance roles; whether that adaptivity extends to temporal discipline is an empirical question that the series has not yet answered.
The answer matters for policy. If LLM regulators can reliably read leading signals and act on them before the cascade commits, then the BDPD3 finding has a direct path to application: arm AI-assisted governance systems on capital-type indicators rather than pollution-type indicators. If LLM regulators are too noisy or too slow to exploit the leading signal, then the finding remains theoretically important but practically constrained. The mini-challenge of this lecture does not answer this question — it asks the student to see the cascade, not to govern it — but it prepares the ground for the question.
3.6.2 Is “polycentric” one configuration, or many?
BDPD3’s configuration is multi-arena (multiple physical sites with unidirectional coupling) rather than fully polycentric in Ostrom’s sense (multiple autonomous decision centres with mutual feedback). The single governance meta-agent and unidirectional pollution link constrain the institutional structure tested here. Carlisle and Gruby’s model specifies that polycentric systems require decision centres that “take each other into account in competitive and cooperative relationships” — the BDPD3 setup has no such mutual accounting. A richer design — in which the downwind arenas can observe, respond to, and potentially sanction the emitter — would test whether the leading-signal advantage survives when the governed actors can themselves exert institutional pressure on the governance arrangement. This connects directly to the sanctions literature of Lecture 4.
3.6.3 What about bidirectional cascades?
The A→B/C topology is unidirectional: the emitter pollutes the downwind, but the downwind does not affect the emitter. Real-world polycentric systems are rarely this clean. When externalities flow in both directions — as in shared river basins, atmospheric pollution, or digital infrastructure — the leading/lagging distinction becomes harder to operationalise because each jurisdiction is simultaneously emitter and victim. Whether the capital-trigger advantage survives bidirectional coupling is an open empirical question, and one that the cascading-tipping literature has begun to address in other contexts.
3.6.4 What about heterogeneous substrates?
BDPD3’s three arenas run the same Seneca ODE with the same parameters. A more demanding question — and one that maps onto the heterogeneity theme of Lecture 1 — is what happens when the arenas have different ODE structures, different timescales, or different agent types. If arena B runs a logistic substrate while arena A runs Seneca, the leading/lagging distinction may not transfer directly. The classical literature on human heterogeneous types (Fischbacher, Burlando-Guala, as discussed in Lecture 1) suggests that mixed populations behave non-trivially differently from any of the pure types. Whether the same is true for mixed substrates is open.
3.6.5 The rate-induced governance boundary
The dt boundary at which the win-win degrades to win-some is itself a finding worth examining. It suggests that governance sampling rate is a design dimension, not just a technical parameter: a regulator that samples too slowly will lose the emitter even if it monitors the right signal. This connects to the cascading-tipping literature’s treatment of rate-induced tipping, and suggests that cascade taxonomies should treat governance sampling rate alongside forcing rate as a risk factor. The practical implication is that monitoring infrastructure must be fast enough to track the leading signal at the rate at which it moves — a constraint that the polycentric governance literature, focused on institutional structure, has not articulated. Lecture 10 develops this finding in detail and connects it to the Lohmann rate-induced-tipping framework.
3.6.6 Beyond the Seneca substrate
BDPD3 uses a specific ODE — the Bardi/Seneca three-variable system — to make the leading/lagging distinction sharp. The question is whether the finding generalises to other substrates with different dynamical structures. In a logistic substrate (the workhorse of BDPD0), the resource stock is both the cause of exploitation and the signal of exploitation; there is no multiplicative delay between cause and symptom, and the leading/lagging distinction may not apply. In more complex substrates — with multiple stocks, nonlinear interactions, or stochastic forcing — the distinction may apply but with different timing. Whether the capital-trigger advantage is a general feature of governance in multiplicative-dynamics systems or a specific feature of the Seneca ODE is an open question that the BDPD series has not yet answered.
3.7 Synthesis
The good-will intuition — decentralise governance, multiply the decision centres — emerges from this lecture confirmed and revised:
Confirmed. The classical literature confirms the intuition robustly. Polycentric governance systems, under the right institutional conditions, outperform both pure centralisation and pure decentralisation. Ostrom’s empirical record, the design-principle framework, and the growing theoretical literature on polycentric environmental governance all support the claim that multiple overlapping decision centres can manage shared resources effectively.
Bounded. The classical literature also delimits the intuition. The design principles — boundaries, monitoring, sanctions, conflict resolution, nested enterprises — identify the institutional conditions under which self-governance succeeds. Outside those conditions, polycentric arrangements fail or produce perverse outcomes.
Temporal. BDPD3 identifies an additional boundary, one the classical laboratory tradition had no practical means to measure: the trigger signal. The binding axis of governance is not the instrument but the signal the instrument is armed on. A perfectly designed cap, levy, or fine armed on a lagging indicator is structurally too late when externalities cascade. The same instruments armed on a leading indicator produce Pareto-improving win-wins. This temporal dimension is invisible to the institutional-design literature because that literature asks which rules, not when they fire. The cascading-tipping literature, meanwhile, has shown that temporal delays are the mechanism of cascade propagation — but has not connected this finding to the design of governance monitoring. BDPD3 sits at the intersection: the governance signal must act before the delay commits, and the delay is a feature of the ODE structure, not of the institutional arrangement.
The lesson that connects this lecture to the rest of the course is the same lesson that connected Lecture 1: the good-will intuition is not wrong, but it is conditional. Polycentric governance works — under institutional conditions (Ostrom’s design principles) and under a temporal condition (the monitoring signal must be leading, not lagging). The institutional conditions are well understood. The temporal condition is new.
The cultural payoff of the lecture is not “decentralisation doesn’t work”. It is the more careful claim that polycentric governance works in a precise institutional and temporal context, and the temporal context is the one most often neglected. When externalities cascade across jurisdictional boundaries — as they do in climate, fisheries, nutrient pollution, and any system with hidden inertia — the question is not only who decides and what they decide, but what they are looking at when they decide.
The lesson generalises beyond the Seneca substrate. Whenever the governance problem involves a process with multiplicative dynamics — where the damage variable’s growth rate is proportional to the stock of the cause variable — a lagging indicator will be structurally too late. The leading/lagging distinction is not a technical detail; it is a design principle that the polycentric governance literature needs to absorb. Ostrom’s fourth principle (monitoring) should be read not just as “someone watches” but as “someone watches the right variable at the right point in the causal chain”.
The next chapter of this particular story is written in the mini-challenge. But the thread that connects this lecture to the next one is the sanctions ladder: if the governance system monitors the right signal, what instruments should it deploy? The polycentric literature has emphasised graduated sanctions as a key design principle (Ostrom’s Principle 5). Lecture 4 will ask what sanctions look like when the agents being sanctioned do not feel cost — and when the sanctioners can compute optimal evasion strategies in milliseconds. The temporal condition identified in this lecture — monitor the leading signal — is a prerequisite for any sanctions regime: punishment armed on a lagging signal is as structurally late as the cap and levy we have just examined.