Contemporary examples — candidate list

Twenty-six collective action cases from roughly 1980 to now, each run through the five-question diagnostic, with module fit, media and sources.

Contemporary examples — candidate list

Candidates, not selections. Range is roughly 1980 to the present, chosen to connect the theory to the domains course participants work in. Successes and failures both, because a list of failures teaches only that the problem is hard.

Each entry runs the five diagnostic questions from arc.md explicitly, plus a sixth line on substitutes — whether a cheap technical alternative to the harmful behavior existed. That sixth question is the standard explanation for why Montreal worked and Kyoto didn't, and it is doing more work in these cases than any of the other five.

Figures here have been checked against the sources cited. Where a number is contested or a judgment is early, that's said in the entry.

Climate and atmosphere

The Montreal Protocol

  • What happened: Farman, Gardiner and Shanklin published the Antarctic ozone hole in Nature in May 1985. The Vienna Convention was agreed in March 1985 and the Montreal Protocol signed in September 1987, entering force January 1989. It now has 198 parties — universal ratification, the only UN treaty to reach it — and has phased out over 99% of production and consumption of controlled ozone-depleting substances. WMO's 2022 assessment projects a return to 1980 ozone values around 2040 globally, 2045 over the Arctic and 2066 over the Antarctic.
  • Why it's a collective action case: A global commons with a classic free-rider structure — each country's CFC emissions damage everyone's ozone layer, and unilateral abstention buys nothing. Solved, and comprehensively.
  • The diagnostic:
    • How many players? Effectively a few dozen producing countries, and a very small number of chemical firms — DuPont, ICI, Atochem, Allied Signal. Not 198.
    • Can you see defection? Yes. Production is industrial and traceable, and atmospheric monitoring detects emissions independently of self-reporting.
    • Can you punish it? Yes. Article 4 bans trade in controlled substances with non-parties, which converts the treaty into a club with a membership benefit.
    • Will you meet again? Yes — annual Meetings of the Parties, with a standing scientific and technology assessment process that has amended the protocol repeatedly (London 1990, Copenhagen 1992, Montreal 1997, Beijing 1999, Kigali 2016).
    • Is anyone big enough to act alone? Close. The US had already banned CFC aerosols in 1978, and DuPont — holding roughly a quarter of the world market — announced a total phase-out in March 1988, which reset what the industry treated as feasible.
    • Substitutes? Yes, and this is the core of it. HCFCs and HFCs were available and cheap, and the firms that would be regulated held the patents on the replacements. Compliance cost was low and the incumbents stood to profit.
  • Could serve: Session 4, Did it ever work? (the anchor case). Session 1, Introduction to collective action, as the opening success. Session 4, Changing the payoffs, on the trade-ban mechanism.
  • Contested: The standard framing — "the treaty that proves it can be done" — is disputed by the people who study it most closely. Barrett's argument in Environment and Statecraft is that Montreal is an easy case: small compliance cost, concentrated industry, available substitutes, a trade restriction that actually bit. Sunstein's comparison adds that the US ran a domestic cost-benefit analysis showing large net domestic benefits from unilateral action — the reverse of the climate arithmetic. Teaching Montreal as a template is the mistake; teaching it as a diagnostic result is the point.
  • Media: Good data visualizations exist — NASA Ozone Watch publishes the Antarctic ozone hole area time series annually, which is the single best "it worked" chart in the course. Primary sources: the protocol text and all amendments at ozone.unep.org. No major documentary known.
  • Source: Farman, Gardiner & Shanklin, "Large losses of total ozone in Antarctica reveal seasonal ClOx/NOx interaction", Nature 315, 1985. Ozone Secretariat, "Facts and figures on ozone protection" (ozone.unep.org/facts-and-figures-ozone-protection). WMO/UNEP Scientific Assessment of Ozone Depletion: 2022. Scott Barrett, Environment and Statecraft, OUP 2003. Cass Sunstein, "Of Montreal and Kyoto: A Tale of Two Protocols", Harvard Environmental Law Review 31, 2007.

The CFC-11 defection, 2013–2019

  • What happened: Montzka et al. reported in Nature in 2018 that global CFC-11 emissions had stopped declining and risen from around 2013, in breach of the protocol. Rigby et al. (2019) traced roughly half the global increase to eastern China using atmospheric measurements from Gosan, South Korea and Hateruma, Japan. Two Nature papers in February 2021 reported that emissions had fallen back to pre-2013 levels by 2019 — eastern China down by 10 ± 3 Gg/yr from 2014–2017 levels.
  • Why it's a collective action case: The enforcement half of Montreal, tested live. A treaty with no inspectorate detected a violation from the atmosphere and the violation stopped within about two years.
  • The diagnostic:
    • How many players? A handful of illicit foam-blowing producers, within one country.
    • Can you see defection? This is the whole point of the case. Yes — not through reporting, which showed nothing, but through a global network of background air monitoring stations. Detection was physical, not administrative.
    • Can you punish it? Domestically, yes: Chinese authorities acted after the attribution was published. Internationally, reputational cost plus the existing trade sanctions.
    • Will you meet again? Yes, continuously.
    • Is anyone big enough to act alone? Not relevant here; the question is monitoring, not provision.
    • Substitutes? Yes — legal blowing agents existed; CFC-11 was used because it was cheaper.
  • Could serve: Session 4, Did it ever work?, as the follow-up beat to Montreal. Session 3, Repeated interaction and the folk theorem, as a concrete instance of "can you see defection" being an engineering question with an answer.
  • Media: The Nature papers carry publication-quality emissions charts. The Conversation piece by Rigby and Park ("How we traced 'mystery emissions' of CFCs back to eastern China", 2019) is a readable first-person account of the detection.
  • Source: Montzka et al., "An unexpected and persistent increase in global emissions of ozone-depleting CFC-11", Nature 557, 2018. Rigby et al., "Increase in CFC-11 emissions from eastern China based on atmospheric observations", Nature 569, 2019. Park et al., "A decline in emissions of CFC-11 and related chemicals from eastern China", Nature 590, 2021. Montzka et al., "A decline in global CFC-11 emissions during 2018−2019", Nature 590, 2021.

The Kigali Amendment

  • What happened: HFCs, the replacement for CFCs, turned out to be potent greenhouse gases. The Kigali Amendment to the Montreal Protocol, agreed October 2016 and in force from January 2019, phases them down; there were 172 parties as of December 2025. NOAA's estimate is that compliance avoids 0.3–0.5°C of warming by 2100 relative to no controls.
  • Why it's a collective action case: A climate agreement smuggled into an ozone treaty, and by some distance the most effective one. It works because it inherits Montreal's machinery — the trade provisions, the assessment panels, the funding mechanism.
  • The diagnostic:
    • How many players? Same concentrated refrigerant and air-conditioning industry.
    • Can you see defection? Yes, by the same atmospheric route.
    • Can you punish it? Yes, inherited trade restrictions.
    • Will you meet again? Yes, the same annual cycle.
    • Is anyone big enough to act alone? No, but the US ratified in 2022 with unusual bipartisan support, driven by domestic manufacturers who wanted the global standard set where their products already were.
    • Substitutes? Yes — HFOs and natural refrigerants. Same structure as Montreal, a third time.
  • Could serve: Session 4, Did it ever work?, as the counter-example to "climate is just harder": a climate agreement that worked, because it was shaped like an ozone agreement.
  • Media: None known beyond UNEP explainers.
  • Source: Ozone Secretariat, "Kigali Amendment to the Montreal Protocol" (ozone.unep.org/kigali-amendment-overview). Congressional Research Service, "Hydrofluorocarbon Phasedown: Background and Issues Facing Congress", IF11779.

The Kyoto Protocol

  • What happened: Adopted December 1997, setting binding targets for 37 industrialized countries and the EU — an average 5% cut below 1990 levels over 2008–2012. The US signed but never ratified; Bush formally repudiated it in 2001, with the US then accounting for roughly a third of Annex I emissions. Canada withdrew in December 2011, citing the absence of the US and China and a potential $14bn liability. Developing countries, including China, had no targets. The treaty entered force only in February 2005, after Russia ratified.
  • Why it's a collective action case: A global public good where the largest emitters faced the largest costs and the benefits accrued to everyone regardless. Failed as a global instrument. The EU did meet its first-period target, which matters for how the failure is described.
  • The diagnostic:
    • How many players? Nominally 37 Annex B parties; in substance, every economy on earth, because emissions are not concentrated in an industry the way CFC production was.
    • Can you see defection? Partly. National emissions inventories are self-reported and were, at the time, not independently verifiable at treaty precision. Contrast Montreal, where the atmosphere told on you.
    • Can you punish it? No. The compliance mechanism's ultimate sanction was stricter targets in a future period you could simply leave — which Canada did, at no cost.
    • Will you meet again? Yes, but the shadow of the future is weak when the penalty for defection is a future obligation.
    • Is anyone big enough to act alone? No. No country's unilateral abatement materially changes its own climate outcome. This is the structural difference from the US position on ozone.
    • Substitutes? No — not in 1997. Fossil energy had no cheap drop-in replacement, abatement cost fell on the whole economy rather than a few chemical firms, and no incumbent stood to profit from the switch. This is the single largest difference from Montreal.
  • Could serve: Session 4, Did it ever work? — the other half of the named contrast pair. Session 2, The games, as a clean public-goods free-rider structure.
  • Contested: "Kyoto failed" is the shorthand and it flattens things. EU emissions in 2020 were 31% below 1990 levels, well past the 8% target, and the protocol built the emissions-accounting and market infrastructure that Paris later used. The honest version: it failed at the thing it was for, which was binding the large emitters.
  • Media: UNFCCC holds the primary documents and the full ratification record. Our World in Data has good per-country emissions series for constructing the before/after. No strong documentary known.
  • Source: UNFCCC, "Kyoto Protocol — Targets for the first commitment period". Al Jazeera, "Canada quits Kyoto climate protocol", 13 December 2011. Sunstein, "Of Montreal and Kyoto", 2007.

The Paris Agreement

  • What happened: Adopted 12 December 2015, in force 4 November 2016. Abandoned binding targets for nationally determined contributions plus a ratchet-and-review cycle. Climate Action Tracker's November 2025 update puts warming under current policies at 2.6°C, unchanged for four years; the pledges and targets scenario worsened from 2.1°C to 2.2°C, largely because the US withdrawal invalidated its targets. CAT also notes the current-policies projection has fallen roughly 1°C from the 3.6°C estimated in 2015.
  • Why it's a collective action case: A deliberate architectural bet — that transparency, peer pressure and repeated review can substitute for enforcement. The result is neither the success nor the failure it gets called.
  • The diagnostic:
    • How many players? ~195 parties, all of them.
    • Can you see defection? Better than under Kyoto. The Enhanced Transparency Framework plus satellite and atmospheric monitoring have made emissions much harder to misreport, and independent trackers publish the gap.
    • Can you punish it? No. There is no sanction, by design. The US has now left twice.
    • Will you meet again? Yes, on a five-year ratchet.
    • Is anyone big enough to act alone? No, though China's solar and battery manufacturing has done more to change the cost curve than any treaty provision.
    • Substitutes? Increasingly, yes — and this is the interesting part. Solar, wind and batteries became cheap between 2015 and now, which moves climate slowly toward the Montreal structure. The 1°C improvement in the current-policies projection is substitutes arriving, not enforcement working.
  • Could serve: Session 4, Did it ever work? Session 4, Polycentricity — Paris is close to Ostrom's answer applied at planetary scale, and worth examining as such rather than as a weak treaty. Session 6, Why the current stack can't hold.
  • Contested: Judging Paris is premature and people on both sides do it anyway. It has not delivered 1.5°C and will not. It has coincided with a 1°C improvement in the projected outcome. Whether the agreement caused any of that, or whether it is technology cost curves that would have moved regardless, is genuinely unsettled.
  • Media: Climate Action Tracker's thermometer graphic is the standard visualization and is updated annually. UNEP Emissions Gap Report each year. Primary text at unfccc.int.
  • Source: Climate Action Tracker, "Little change in warming outlook for four years; new 2035 climate targets make no difference", Global Update, 13 November 2025. UNEP, Emissions Gap Report 2025.

The US Acid Rain Program

  • What happened: Title IV of the 1990 Clean Air Act Amendments created a cap-and-trade system for power sector SO2, aiming at a permanent 10 million ton cut from 1980 levels, with the 2010 cap set at 8.95 million tons — about half of 1980 power sector emissions. By 2002, SO2 from affected sources was 10.2 million tons, down 35% from 1990. Costs came in well below the original projections.
  • Why it's a collective action case: A commons problem — acid deposition — solved within one jurisdiction by a coercive cap plus a market in the residual. Useful precisely because it's the sub-global case: the same structure, with a sovereign present.
  • The diagnostic:
    • How many players? A few hundred generating units, all within one legal system.
    • Can you see defection? Yes, and this is the underrated part: the program mandated continuous emissions monitoring on every unit. Observation was engineered before the market was built.
    • Can you punish it? Yes — automatic penalties per excess ton, no discretion.
    • Will you meet again? Yes, indefinitely.
    • Is anyone big enough to act alone? No, which is why the sovereign did it.
    • Substitutes? Yes — low-sulfur Powder River Basin coal, and scrubbers, both of which got cheaper than expected once rail deregulation cut transport costs.
  • Could serve: Session 4, Coercion, as Hardin's "mutual coercion mutually agreed upon" in its cleanest modern form. Session 4, Changing the payoffs.
  • Media: EPA publishes long emissions and deposition time series. Good before/after deposition maps exist.
  • Source: US EPA, "1990 Clean Air Act Amendment Summary: Title IV" and "Acid Rain Program". Chestnut & Mills, "A fresh look at the benefits and costs of the US acid rain program", Journal of Environmental Management 77, 2005.

The global leaded petrol phase-out

  • What happened: UNEP announced on 30 August 2021 that the era of leaded petrol was over, after service stations in Algeria — the last country using it — stopped supplying it in July 2021. The UNEP-led Partnership for Clean Fuels and Vehicles ran the campaign for roughly two decades; sub-Saharan Africa went unleaded in 2006. UNEP's estimate is over 1.2 million premature deaths avoided annually.
  • Why it's a collective action case: A cross-border pollution problem with a race-to-the-bottom in fuel standards, resolved country by country rather than by treaty. Interesting because there was no treaty.
  • The diagnostic:
    • How many players? Nominally every country, but in practice a small number of refiners and a single dominant additive producer.
    • Can you see defection? Yes — fuel at the pump is testable, and stocks are traceable.
    • Can you punish it? Weakly. Mostly aid conditionality, technical assistance and vehicle import standards rather than sanction.
    • Will you meet again? Yes.
    • Is anyone big enough to act alone? The US and EU bans effectively set the direction, because catalytic converters are destroyed by leaded fuel and the global vehicle fleet is a hand-me-down from rich countries.
    • Substitutes? Yes. Unleaded formulations were straightforward and the refinery retrofits were affordable, which is why the campaign could proceed without any enforcement mechanism at all.
  • Could serve: Session 4, Did it ever work? — the third data point after Montreal and Kyoto, and the one that shows the mechanism was substitutes rather than treaty design. Session 1, Introduction to collective action.
  • Media: UNEP press release and UNifeed video from 30 August 2021. Blood lead level time series make strong charts.
  • Source: UNEP, "Era of leaded petrol over, eliminating a major threat to human and planetary health", 30 August 2021.

AI

The frontier AI race

  • What happened: From roughly 2020, a small number of labs — OpenAI, Google DeepMind, Anthropic, Meta, xAI, and Chinese labs including DeepSeek and Zhipu — have scaled models under competitive pressure, with training compute and capital expenditure rising steeply year on year. Several labs were founded on explicit safety rationales and each has publicly argued that unilateral slowing would hand the lead to a less careful competitor.
  • Why it's a collective action case: An arms race in Schelling's sense. The distinctive feature is that participants state the trap openly in their own founding documents, which makes it the cleanest available case of structure overriding intention.
  • The diagnostic:
    • How many players? Small — under ten at the frontier, which by Olson's logic should make agreement possible. It has not, so far.
    • Can you see defection? Partly. Training runs of frontier scale are visible through hardware procurement and data center energy draw; internal safety practice is not visible at all. Capability is observable, care is not.
    • Can you punish it? Not really. Export controls act on states rather than labs; reputational cost is real but small relative to the returns.
    • Will you meet again? Yes, constantly — the labs share staff, conferences and investors. High interaction has not produced cooperation, which is worth sitting with.
    • Is anyone big enough to act alone? No. Nvidia and TSMC are closer to being chokepoints than any lab.
    • Substitutes? No. There is no cheaper way to get the capability, and no technical alternative that delivers the payoff without the risk. This is the Kyoto position, not the Montreal one.
  • Could serve: Session 1, The pattern beneath the metacrisis and Why this matters so much now. Session 4, Schelling — the highest-value pairing in the course, since the nuclear literature was written for exactly this shape. Session 2, The many names (multipolar trap).
  • Media: Epoch AI publishes the training compute dataset and charts, which are the best available quantitative visualization. Primary sources: the founding charters and safety frameworks of the labs, which state the dilemma explicitly.
  • Source: Epoch AI, Notable AI Models database (epoch.ai). Thomas Schelling, The Strategy of Conflict, 1960, for the structure.

AI summits and voluntary commitments

  • What happened: The UK convened the first AI Safety Summit at Bletchley Park on 1–2 November 2023, producing the Bletchley Declaration. At the Seoul summit of 21–22 May 2024, sixteen companies — including Amazon, Anthropic, Google, IBM, Microsoft, OpenAI, Cohere, Mistral, G42, Naver, Samsung and Zhipu — signed the Frontier AI Safety Commitments, agreeing to publish safety frameworks defining intolerable risk thresholds. By the Paris AI Action Summit of 10–11 February 2025, the framing had shifted from safety to action and investment; 61 countries and organizations signed the closing statement and the US and UK both declined.
  • Why it's a collective action case: An attempt to build the cooperation infrastructure — common standards, mutual observation, a repeated meeting — in advance of any enforcement. The two-year trajectory from Bletchley to Paris is the case, more than any single summit.
  • The diagnostic:
    • How many players? Sixteen firms and around thirty governments at Seoul. Small enough, on paper.
    • Can you see defection? Only through self-published frameworks. There is no audit. Third-party trackers like AI Lab Watch exist and depend on voluntary disclosure.
    • Can you punish it? No. The commitments are voluntary and carry no penalty.
    • Will you meet again? Yes — the summit series continues, with India hosting in 2026.
    • Is anyone big enough to act alone? The US is, and its February 2025 refusal to sign demonstrated that a single large player leaving is enough to change what the agreement means.
    • Substitutes? No.
  • Could serve: Session 4, Did it ever work? — as the live case rather than the historical one. Session 3, Repeated interaction and the folk theorem, on what repetition alone can and can't deliver without observation or punishment.
  • Media: Full summit texts and signatory lists are public (gov.uk for Bletchley and Seoul; the Paris statement via the French government). Future of Life Institute maintains a summit archive and an AI Safety Index scoring lab frameworks.
  • Source: UK Government, "Frontier AI Safety Commitments, AI Seoul Summit 2024". CNBC, "U.S. and Britain snub international AI accord in Paris", 11 February 2025. Future of Life Institute, AI Safety Summits project page.

The EU AI Act, and the regulatory-race question

  • What happened: The EU AI Act entered into force on 1 August 2024, with prohibitions applying from February 2025 and general-purpose AI model obligations from 2 August 2025. In 2026 the Digital Omnibus package pushed back parts of the timetable, with full applicability of the remaining provisions now scheduled for 2 August 2028.
  • Why it's a collective action case: A test of whether one jurisdiction can set a global floor — the Brussels effect — against a race to the bottom in which each jurisdiction has an incentive to be the lax one and capture the investment. The 2026 delay is evidence about which way it's going.
  • The diagnostic:
    • How many players? Three regulatory blocs that matter, plus every firm.
    • Can you see defection? By jurisdictions, yes — legislation is public. By firms, partly, through conformity assessment.
    • Can you punish it? Within the EU, yes: fines up to 7% of global turnover. Between jurisdictions, no.
    • Will you meet again? Yes.
    • Is anyone big enough to act alone? This is exactly the claim being tested. The EU is a large enough market that compliance often propagates globally, as it did with GDPR. Whether that holds for AI is open.
    • Substitutes? Not applicable in the usual sense; the relevant question is whether compliant models are competitive.
  • Could serve: Session 4, Coercion. Session 4, Polycentricity — a stacked rather than global solution. Session 6, Why the current stack can't hold.
  • Media: artificialintelligenceact.eu maintains an annotated text and implementation timeline. Primary text in the Official Journal.
  • Source: Regulation (EU) 2024/1689. EU AI Act implementation timeline, artificialintelligenceact.eu. DLA Piper, "Latest wave of obligations under the EU AI Act take effect", August 2025.

California SB 1047 and SB 53

  • What happened: SB 1047 would have imposed safety testing and shutdown requirements on frontier model developers, plus whistleblower protections. It passed both chambers and Governor Newsom vetoed it on 29 September 2024, citing the lack of empirical grounding and the breadth of its standards. Exactly a year later, on 29 September 2025, he signed SB 53, the Transparency in Frontier Artificial Intelligence Act — narrower, disclosure-focused, and the first US frontier AI law.
  • Why it's a collective action case: A sub-national jurisdiction attempting to regulate an industry that can relocate, under explicit pressure that regulation would push developers out of the state. The veto and the later signature show the bargaining range.
  • The diagnostic:
    • How many players? A handful of developers; fifty states; one industry concentrated in one of them.
    • Can you see defection? Under SB 53, partly — disclosure is mandated but unverified.
    • Can you punish it? Within California, yes, modestly. A developer can also incorporate elsewhere.
    • Will you meet again? Yes — annual legislative sessions, which is why the second attempt succeeded.
    • Is anyone big enough to act alone? California hosts most frontier labs, so it is the closest thing to a unilateral actor in US AI policy.
    • Substitutes? Not applicable.
  • Could serve: Session 4, Did it ever work? Session 4, Polycentricity, as a federated attempt at a problem no single state owns.
  • Media: Bill texts and the veto message are public on leginfo.ca.gov. Good contemporaneous coverage from CalMatters and NPR.
  • Source: CalMatters, "Newsom vetoes major California artificial intelligence bill", September 2024. Future of Privacy Forum, "California's SB 53: The First Frontier AI Law, Explained", 2025.

Weapons

The Non-Proliferation Treaty

  • What happened: Opened for signature 1968, in force 1970, extended indefinitely in 1995. It now has 191 states parties, the widest adherence of any arms control treaty. India, Israel, Pakistan and South Sudan never joined; North Korea announced withdrawal in 2003. Kennedy's 1963 estimate of 15–25 nuclear-armed states by the 1970s did not happen: there are nine. South Africa dismantled its arsenal; Ukraine, Belarus and Kazakhstan transferred theirs to Russia; Argentina, Brazil, Sweden, Taiwan and South Korea abandoned programs.
  • Why it's a collective action case: The best available evidence that a large-N security dilemma can be held for fifty years — and simultaneously the best evidence about what that costs, since the bargain has always been that five states keep theirs.
  • The diagnostic:
    • How many players? 191 parties, but the binding constraint is the ~40 states with the technical capacity to build one.
    • Can you see defection? Yes, increasingly. IAEA safeguards, the Additional Protocol, seismic and radionuclide monitoring under the CTBTO, and commercial satellite imagery. Enrichment is hard to hide, though Iraq, Libya, Iran, Syria and North Korea all got further than safeguards caught.
    • Can you punish it? Yes — sanctions, and in several cases military action. The punishment is real, which distinguishes this from the BWC.
    • Will you meet again? Yes — five-yearly Review Conferences, the next in 2026.
    • Is anyone big enough to act alone? Yes, and that's the mechanism. US security guarantees are what made non-acquisition rational for Germany, Japan and South Korea. Extended deterrence is a selective incentive in Olson's sense.
    • Substitutes? Yes, of a kind — alliance protection substitutes for an arsenal. Where the substitute is credible, states don't build. Where it erodes, this case reopens.
  • Could serve: Session 4, Did it ever work? Session 4, Schelling. Session 1, War. Session 5, The bill — the treaty is explicitly a two-tier bargain, an in-group that keeps what the out-group is denied.
  • Contested: Whether the NPT caused the non-proliferation, or whether alliance guarantees and the sheer cost of programs did the work and the treaty ratified it, is a live argument.
  • Media: Arms Control Association fact sheets are the standard reference and are kept current. Bulletin of the Atomic Scientists' Nuclear Notebook gives warhead-count time series — a strong visualization, since global stockpiles peaked around 70,000 in 1986 and are now around 12,000.
  • Source: Arms Control Association, "The Nuclear Nonproliferation Treaty (NPT) at a Glance". UNODA treaty page. IAEA, "Treaty on the Non-Proliferation of Nuclear Weapons".

The expiry of New START

  • What happened: New START, signed 2010, capped deployed US and Russian strategic warheads at 1,550 each and provided for on-site inspection. Russia suspended participation in 2023. Putin proposed on 22 September 2025 that both sides continue observing the central limits informally for a year past expiry. The treaty expired on 5 February 2026 with no formal US response, leaving the two largest arsenals under no legally binding limits for the first time since 1972.
  • Why it's a collective action case: A two-player repeated game with explicit verification, which has now lapsed. It's the cleanest small-N case in the course, and it just failed in real time.
  • The diagnostic:
    • How many players? Two. Olson says this should be the easy case.
    • Can you see defection? Under the treaty, yes — 18 on-site inspections a year plus data exchanges plus national technical means. Without it, satellite imagery only, which counts launchers rather than warheads.
    • Can you punish it? Only by matching — build-up for build-up.
    • Will you meet again? Yes, indefinitely, which is why arms control worked at all through the Cold War.
    • Is anyone big enough to act alone? Either party can set the level unilaterally, and each has.
    • Substitutes? No.
  • Could serve: Session 4, Schelling, as the live example. Session 1, War. Session 3, Repeated interaction and the folk theorem — verification regimes are folk-theorem engineering, and the case shows what happens when you remove the observation channel from a game that otherwise has every advantage.
  • Media: Arms Control Association Arms Control Today covered the expiry (March 2026 issue). Federation of American Scientists publishes deployed-warhead time series.
  • Source: Arms Control Association, "New START Expires As U.S. Urges 'Modernized' Treaty", Arms Control Today, March 2026. Congressional Research Service, "Extension of New START Central Limits", IN12640.

The Biological Weapons Convention and the failed verification protocol

  • What happened: The BWC opened for signature in 1972 and entered force 26 March 1975; it has 187 states parties. It bans development and stockpiling and contains no verification machinery. From 1994 an Ad Hoc Group negotiated a protocol requiring facility declarations, random visits and challenge investigations. In July 2001 the US rejected the draft and withdrew from the negotiation; it has never been revived. The Implementation Support Unit, the treaty's entire permanent staff, is three people in Geneva.
  • Why it's a collective action case: An agreement with near-universal membership, no observation and no punishment. It's the control condition for the diagnostic: what a treaty looks like when only questions 1 and 4 get a yes.
  • The diagnostic:
    • How many players? 187 parties, and a growing number of non-state actors with the relevant capability — which is the direction of travel that makes this worse over time.
    • Can you see defection? No. Dual-use equipment is identical to legitimate pharmaceutical and vaccine plant. The Soviet Biopreparat program employed tens of thousands while the USSR was a party, and was revealed by defectors rather than detected.
    • Can you punish it? Formally, complaint to the Security Council, subject to veto. The mechanism has never been used.
    • Will you meet again? Yes — Review Conferences every five years, which have repeatedly failed to agree anything substantive.
    • Is anyone big enough to act alone? No.
    • Substitutes? Not applicable — the relevant asymmetry is that the offensive and defensive research are the same research.
  • Could serve: Session 4, Did it ever work? — as the failure whose cause is unusually legible. Session 1, The pattern beneath the metacrisis. Session 1, Why this matters so much now, on falling capability barriers.
  • Media: Ken Alibek's Biohazard (1999) is the insider account of Biopreparat. Primary documents at the UN Office for Disarmament Affairs. No good visualizations known.
  • Source: Arms Control Association, "The Biological Weapons Convention (BWC) At A Glance". Marie Chevrier, "The Biological Weapons Convention: the protocol that almost was", Verification Yearbook 2001. US GAO, "Arms Control: Efforts to Strengthen the Biological Weapons Convention", GAO-02-1038NI.

Gain-of-function research and the self-imposed moratorium

  • What happened: After the 2011 Fouchier and Kawaoka H5N1 transmissibility experiments, researchers declared a voluntary 60-day moratorium in January 2012, extended to about a year. On 17 October 2014 the US government imposed a funding pause on gain-of-function work on influenza, MERS and SARS viruses, affecting 18 projects, 7 of which were later exempted. The pause was lifted in December 2017 and replaced by the HHS P3CO review framework. A revised OSTP policy was issued in May 2024 with an effective date of 6 May 2025.
  • Why it's a collective action case: A field trying to restrain itself, with the restraint applying only to one country's funding and not to the research anywhere else. The 1975 Asilomar conference on recombinant DNA is the usual precedent and the usual reason people think this can work.
  • The diagnostic:
    • How many players? A few hundred labs worldwide with the capability, in dozens of jurisdictions. Rising.
    • Can you see defection? Weakly. Publication reveals work after it's done; unpublished work is invisible. Biosafety inspections are national and inconsistent.
    • Can you punish it? Only through funding, and only within the funding jurisdiction. A researcher who loses NIH support can work elsewhere.
    • Will you meet again? Yes, within a small and reputationally dense scientific community — which is the strongest thing this case has going for it, and is what Asilomar ran on.
    • Is anyone big enough to act alone? The NIH is the largest single funder, so US policy moves a large fraction of the world's work. Not all of it.
    • Substitutes? Partly. Much of the scientific goal is reachable through safer methods — pseudoviruses, structural prediction, surveillance — which is the main argument that restraint is affordable here.
  • Could serve: Session 4, Did it ever work? Session 3, Repeated interaction and the folk theorem, on reputation in a small community as an enforcement mechanism. Session 1, Why this matters so much now.
  • Media: Primary sources: the 2014 White House announcement, NIH notice NOT-OD-17-071, the 2024 OSTP policy. Congressional Research Service reports IF12021 and R47114 are good neutral summaries of a politically hot topic.
  • Source: NIAID, "NIH Removes Gain-of-Function Research Pause". Congressional Research Service, "Oversight of Gain-of-Function Research with Pathogens", R47114. Paul Berg et al. on Asilomar, PNAS 72, 1975.

Health

Antimicrobial resistance

  • What happened: The GRAM Project's 2022 Lancet analysis estimated 1.27 million deaths in 2019 directly attributable to bacterial antimicrobial resistance, with AMR associated with 4.95 million deaths — more than HIV/AIDS or malaria. A 2024 GRAM forecast projects over 39 million cumulative deaths directly attributable between 2025 and 2050. Meanwhile the antibiotic development pipeline has largely emptied: most large pharmaceutical firms exited antibacterial R&D, and several small developers that brought new antibiotics to market went bankrupt doing it.
  • Why it's a collective action case: Two problems stacked. Overuse is a subtractable commons — every prescription and every kilogram fed to livestock depletes a shared resource. Development is a positive commons — a new antibiotic that should be held in reserve earns nothing, so nobody makes one. The good stewardship that preserves the resource destroys the market that replenishes it.
  • The diagnostic:
    • How many players? Effectively unbounded: every prescriber, every farmer, every patient, in every country.
    • Can you see defection? Poorly. Consumption data is patchy, over-the-counter sale without prescription is common in much of the world, and agricultural use is often unreported.
    • Can you punish it? Barely. The EU banned antibiotic growth promoters in 2006 and requires veterinary prescription; enforcement elsewhere ranges from partial to absent.
    • Will you meet again? Yes, but the actors are anonymous to each other. This is the failure mode Ostrom's design principles are built to avoid, at a scale where none of them apply.
    • Is anyone big enough to act alone? No. Resistance genes cross borders on aircraft and in food.
    • Substitutes? This is the crux. For the overuse problem, often yes — many prescriptions are unnecessary. For the replacement problem, no: there is no substitute for a working antibiotic, and the pull-incentive schemes (the UK NHS subscription model, the proposed US PASTEUR Act) exist precisely because the market cannot supply one.
  • Could serve: Session 3, Ostrom and the commons — it exhibits both the positive and the subtractable commons in one case, which is the distinction the module is built on. Session 1, The pattern beneath the metacrisis. Session 4, Changing the payoffs, on pull incentives.
  • Media: The GRAM Project publishes an interactive country-level visualization tool (vizhub.healthdata.org/microbe). The 2016 O'Neill Review for the UK government is the standard policy document and has the widely reproduced "10 million deaths by 2050" figure — which is a projection, frequently quoted as though it were a measurement.
  • Source: GBD 2019 Antimicrobial Resistance Collaborators, "Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis", The Lancet 399, 2022. GRAM Project, "Global burden of bacterial antimicrobial resistance 1990–2021 and forecasts to 2050", The Lancet, 2024. Jim O'Neill, Tackling Drug-Resistant Infections Globally, 2016.

COVAX and vaccine nationalism

  • What happened: COVAX launched in June 2020 to pool procurement and distribute COVID-19 vaccines equitably. It aimed at 2 billion doses in 2021, cut the target to 1.425 billion by September 2021, and missed that too. High income countries bought up early supply bilaterally; India's export restrictions on the Serum Institute removed the largest planned supplier at the critical moment. COVAX closed on 31 December 2023 having delivered nearly 2 billion doses to 146 economies — the right total, two years late.
  • Why it's a collective action case: Every government's dominant strategy was to secure doses for its own population first, and the collectively optimal allocation — vaccinating the vulnerable everywhere before the young anywhere — was available to nobody unilaterally.
  • The diagnostic:
    • How many players? ~190 governments plus a handful of manufacturers.
    • Can you see defection? Yes, immediately and publicly. Advance purchase agreements and export bans were reported in real time. Transparency turned out not to help.
    • Can you punish it? No. Hoarding doses carried reputational cost and nothing else.
    • Will you meet again? Yes — but each government faced an electorate on a much shorter clock than the pandemic.
    • Is anyone big enough to act alone? The US, EU and India each controlled enough manufacturing to change global supply, and each acted domestically first.
    • Substitutes? No. Doses were the binding constraint and there was nothing else to give.
  • Could serve: Session 4, Did it ever work? Session 2, The games — a textbook n-player prisoner's dilemma with full observability, which is useful precisely because it shows observability alone is not sufficient. Session 6, The first enlargement with no outside.
  • Media: Our World in Data's vaccination-by-income-group charts are the standard visualization and show the gap starkly. Gavi published a preliminary impact assessment.
  • Source: WHO, "COVID-19 vaccinations shift to regular immunization as COVAX draws to a close", 19 December 2023. Gavi, "A preliminary assessment of COVAX's impact in lower-income countries". The Conversation, "Covax misses its 2021 delivery target", September 2021.

The WHO Pandemic Agreement

  • What happened: The World Health Assembly adopted the Pandemic Agreement on 20 May 2025 by consensus, after a vote the previous day of 124 in favor, none against, 11 abstentions. The US did not take part, having begun withdrawal from the WHO. The agreement opens for signature only once the Pathogen Access and Benefit-Sharing annex is adopted, and enters into force after 60 ratifications.
  • Why it's a collective action case: An attempt to fix the COVAX failure structurally — trading pathogen sample sharing, which developing countries supply, against vaccine access, which they need. It is a negotiated exchange rather than an appeal to solidarity, which is what makes it interesting.
  • The diagnostic:
    • How many players? 190-odd, minus the largest.
    • Can you see defection? Sample sharing is observable. Manufacturing allocation commitments, less so.
    • Can you punish it? Minimally. Withholding future samples is the main lever, and it is the one Indonesia used in 2007 over H5N1.
    • Will you meet again? Yes.
    • Is anyone big enough to act alone? The US absence is material: it holds a large share of the relevant manufacturing and R&D.
    • Substitutes? Not applicable.
  • Could serve: Session 4, Changing the payoffs, on constructing an exchange where an appeal to the common good failed. Session 6, Why the current stack can't hold.
  • Media: Primary text at who.int. KFF's explainer is the clearest neutral summary of what it does and doesn't do.
  • Source: PAHO/WHO, "World Health Assembly adopts historic Pandemic Agreement", 20 May 2025. KFF, "The 'Pandemic Agreement': What it is, What it isn't".

Fisheries and the ocean

The Newfoundland cod collapse

  • What happened: Canada declared a moratorium on the northern cod fishery on 2 July 1992, after the stock collapsed. More than 30,000 people lost their work and the province's population fell by around 10%. Expected to last 2 years, the moratorium ran 32. The stock has stabilized at roughly 12% of its 1960s biomass. Canada reopened a limited commercial fishery in June 2024 with an 18,000 tonne quota, a decision marine scientists contested as premature.
  • Why it's a collective action case: The canonical modern tragedy of the commons, with the additional detail that it happened inside a single country's 200-mile exclusive economic zone, under active government management, with scientific advice available throughout. Sovereignty and science were both present and it collapsed anyway.
  • The diagnostic:
    • How many players? Thousands of inshore fishers plus a small number of offshore trawler fleets — two groups with very different interests, which is part of why regulation failed.
    • Can you see defection? Poorly, at the time. At-sea catch was under-reported and discarding was routine.
    • Can you punish it? Yes in principle. In practice quota enforcement was weak and the political cost of cutting quotas was borne by the minister who cut them.
    • Will you meet again? Yes — these were multi-generational fishing communities, exactly the conditions Ostrom identifies as favorable. They weren't enough against a regulator setting quotas above scientific advice.
    • Is anyone big enough to act alone? Canada was. It had the jurisdiction and did not use it in time.
    • Substitutes? No. There was no other fishery for those boats, and no other employment in those towns. Zero substitutes is why the political economy ran one way.
  • Could serve: Session 3, Ostrom and the commons, as the counter-case to the Maine lobster fishery. Session 2, The games. The scale-pair with lobster grounds named in the arc.
  • Media: Strong archival footage exists — the July 1992 Radisson Hotel announcement in St. John's, where fishers broke down the doors, is on CBC Archives. Hand.Line.Cod (2016, NFB) is a short documentary on the inshore fishery. The DFO stock assessment charts are clean and dramatic.
  • Source: Fisheries and Oceans Canada, "The Government of Canada announces the historic return of the commercial Northern cod fishery", June 2024. Newfoundland and Labrador Heritage, "Cod Moratorium". CBC News, "End of cod moratorium touted after 32 years", June 2024.

The IWC commercial whaling moratorium

  • What happened: The International Whaling Commission voted in 1982 for a pause in commercial whaling from the 1985/86 season. Norway lodged a formal objection and has never observed it. Japan withdrew its objection in 1985 under US trade pressure, stopped commercial whaling in 1988, and then conducted large-scale "scientific" whaling under Article VIII for three decades. Iceland resumed commercial whaling in 2007. Japan left the IWC entirely in 2019 and resumed commercial whaling in its own waters in July 2019. Several great whale populations have nonetheless recovered substantially.
  • Why it's a collective action case: A commons with an explicit opt-out clause written into the treaty. It shows what happens when defection is legal, observable and costs nothing — and also that the outcome can still be good, if the number of committed defectors is small enough.
  • The diagnostic:
    • How many players? Around 88 IWC members, three of which actually whale commercially.
    • Can you see defection? Yes, completely. Whaling fleets are large and catch is reported. Sea Shepherd filmed it.
    • Can you punish it? Weakly, and only by one route: the US Pelly Amendment allows trade sanctions for undermining international fishery conservation. The threat moved Japan in 1985 and was not carried through later.
    • Will you meet again? Yes, annually, which produced three decades of stalemate rather than convergence.
    • Is anyone big enough to act alone? The US was, via market access, and chose not to press it.
    • Substitutes? Yes, and this is why the case ended better than it looked. Whale oil had been replaced by petroleum decades earlier and whale meat demand in Japan collapsed on its own. The economics defected from the whalers.
  • Could serve: Session 4, Did it ever work?, as the ambiguous case between Montreal and Kyoto. Session 3, Repeated interaction and the folk theorem, on punishment that is available but not used.
  • Media: The Cove (2009, Academy Award for Best Documentary) — dolphins in Taiji rather than whaling proper, but it's the film people have seen. Whale Wars (Animal Planet, 2008–2015) documents the Southern Ocean confrontations. IWC publishes full catch statistics back to 1985.
  • Source: IWC, "Commercial Whaling" (iwc.int). Marine Policy, "Japan's withdrawal from the IWC: An explanation", 2025. Australian Institute of International Affairs, "When Shaming Fails: Japanese Withdrawal from the International Whaling Commission".

The WTO Agreement on Fisheries Subsidies

  • What happened: Adopted at the WTO's 12th Ministerial Conference in June 2022 after more than 20 years of negotiation, banning subsidies for illegal, unreported and unregulated fishing, for fishing overfished stocks, and for fishing unregulated high seas. It entered into force on 15 September 2025 once two-thirds of members had deposited acceptances. A sunset clause terminates the agreement if comprehensive additional disciplines are not adopted within 4 years of entry into force — so by September 2029.
  • Why it's a collective action case: A race to the bottom rather than a commons problem directly. Each state subsidizes its fleet to keep its share of a shrinking catch, and the subsidies collectively fund the overfishing. The first WTO agreement with an environmental objective at its core, and the first multilateral agreement concluded at the WTO since 2013.
  • The diagnostic:
    • How many players? 160-odd WTO members; subsidies are heavily concentrated in about ten.
    • Can you see defection? Improving. Subsidy notification is required, and satellite vessel-tracking (Global Fishing Watch) has made fleet behavior observable in a way it was not ten years ago. This is a genuine technological change in question 2.
    • Can you punish it? In principle yes — WTO dispute settlement. In practice the Appellate Body has been non-functional since 2019, which weakens it considerably.
    • Will you meet again? Yes, and the sunset clause makes the next round compulsory.
    • Is anyone big enough to act alone? No.
    • Substitutes? Partly — aquaculture has grown enormously and now supplies more than half of fish for human consumption, which reduces the pressure behind the subsidies.
  • Could serve: Session 4, Did it ever work? — a recent and genuine multilateral success, useful against the assumption that nothing gets agreed any more. Session 4, Changing the payoffs.
  • Media: Global Fishing Watch's public map is the best live data visualization available for any case in this list. WTO holds the text and the acceptance register.
  • Source: WTO, "WTO Agreement on Fisheries Subsidies enters into force", 15 September 2025. Pew Charitable Trusts, "Fisheries Subsidies Agreement: Why It Matters", October 2025.

The High Seas Treaty (BBNJ)

  • What happened: Adopted 19 June 2023 after nearly two decades of negotiation, covering conservation and sustainable use of marine biodiversity beyond national jurisdiction — roughly two-thirds of the ocean, previously without a mechanism for creating protected areas. The 60th ratification was deposited on 19 September 2025 and the agreement enters into force on 17 January 2026.
  • Why it's a collective action case: The largest unowned commons on the planet, and the first instrument that lets anyone enclose parts of it. Too new to judge, which is worth saying rather than treating its adoption as the outcome.
  • The diagnostic:
    • How many players? All states, with a small number of distant-water fishing nations holding most of the activity.
    • Can you see defection? Better than before, through satellite AIS tracking — though vessels can and do go dark.
    • Can you punish it? Flag-state enforcement only, which is the structural weakness: flags of convenience exist precisely to sell the absence of enforcement.
    • Will you meet again? Yes — a Conference of the Parties is established.
    • Is anyone big enough to act alone? No.
    • Substitutes? Not applicable.
  • Could serve: Session 3, Ostrom and the commons, on enclosure as one of the available answers. Session 6, Why the current stack can't hold, since flag-state enforcement is exactly the sovereignty-shaped hole.
  • Media: High Seas Alliance ratification tracker is live and well designed. No documentary known.
  • Source: High Seas Alliance, "60th ratification triggers entry into force of High Seas Treaty", 19 September 2025. IISD SDG Knowledge Hub, "With 60 Ratifications, BBNJ Agreement to Enter into Force in January 2026".

Information commons and technical coordination

Wikipedia

  • What happened: Founded January 2001. English Wikipedia now holds over 7 million articles and received 62 million edits in 2025 (172 million across all languages). Active editors peaked above 51,000 in 2007 and have declined to around 39,000 as of December 2024, while article count and length keep growing. It is funded by donation and has no paid editorial staff.
  • Why it's a collective action case: A positive commons — a public good that should be massively undersupplied by Olson's logic and is not. Understanding why it works matters more for this course than any individual failure case.
  • The diagnostic:
    • How many players? Millions of potential contributors; a few tens of thousands doing most of the work. The core group is small, which is the Olsonian answer.
    • Can you see defection? Yes, completely and immediately. Every edit is attributed, timestamped and diffable, and the revision history is permanent. This is the most observable system in this list.
    • Can you punish it? Yes — reversion, page protection, blocks, bans, and an arbitration committee. A full governance stack grew organically.
    • Will you meet again? Yes, and reputation is persistent and portable within the project.
    • Is anyone big enough to act alone? No, and the project's design assumes it.
    • Substitutes? Nupedia and Citizendium existed and failed; Wikipedia's advantage was the absence of a credible substitute once network effects took hold. The live question now is whether LLMs trained on it substitute for it, and what happens to contribution when the readership goes.
  • Could serve: Session 3, Ostrom and the commons — Wikipedia satisfies most of the design principles and is the best modern instance of them at scale. Session 1, Introduction to collective action, as an opening success. Session 4, Polycentricity, on the language-community federation.
  • Media: Wikimedia Statistics (stats.wikimedia.org) is a public, live, well-built dashboard — genuinely good for live demonstration. Truth in Numbers? (2010) is a documentary about the project. Every article's history page is a primary source you can open in front of an audience.
  • Source: Wikimedia Foundation statistics, stats.wikimedia.org. Aaron Halfaker et al., "The Rise and Decline of an Open Collaboration System", American Behavioral Scientist 57, 2013.

Open source infrastructure and the maintainer problem

  • What happened: Heartbleed was disclosed in April 2014, a flaw in OpenSSL, which secured roughly two-thirds of web servers and was maintained by a handful of volunteers with one person full-time and an annual budget of around $2,000. Industry responded with the Core Infrastructure Initiative, funding two full-time developers. Log4Shell followed in December 2021, in a logging library maintained by a small volunteer team. In March 2024 Andres Freund found a backdoor in xz utils (CVE-2024-3094, CVSS 10) planted by a contributor who had spent roughly two years building trust with an overloaded maintainer before being given commit rights.
  • Why it's a collective action case: A positive commons underprovided in the textbook pattern — every firm depends on it, none pays, and the free-riding is near-total. The xz case adds a second structure on top: the maintainer burnout produced by the underfunding is itself the attack surface.
  • The diagnostic:
    • How many players? Every software-using organization on earth. Vast N, and the good is genuinely non-excludable.
    • Can you see defection? Not meaningfully. There is no register of who depends on what, and not funding a project is invisible.
    • Can you punish it? No. Licenses explicitly permit use without contribution.
    • Will you meet again? Yes, but anonymously.
    • Is anyone big enough to act alone? Yes, and this is how it partly gets solved. Google, Microsoft and others fund critical projects because they individually benefit enough — Olson's privileged group, doing exactly what the theory says.
    • Substitutes? No. The dependency is the point.
  • Could serve: Session 3, Olson — the privileged group and selective incentives, in a case the audience uses every day. Session 3, Ostrom and the commons, on the positive/subtractable distinction. Session 1, Introduction to collective action.
  • Media: The XKCD 2347 "Dependency" cartoon is the single most effective visual for this in existence and needs no explanation. Freund's disclosure email to oss-security (29 March 2024) is a readable primary source. Nadia Eghbal's Roads and Bridges (Ford Foundation, 2016) is the standard report.
  • Source: OpenSSF, "xz Backdoor CVE-2024-3094", 30 March 2024. Nadia Eghbal, Roads and Bridges: The Unseen Labor Behind Our Digital Infrastructure, Ford Foundation, 2016. Washington Post, "In wake of Heartbleed, tech companies commit millions to critical open-source software", April 2014.

Y2K

  • What happened: Two-digit year fields in legacy systems threatened date-handling failures at the century rollover. The US Department of Commerce put total US remediation cost at $100 billion over 1995–2001. Coordination ran through national councils, an International Y2K Cooperation Center, and sector-by-sector information sharing including between direct competitors. Essentially nothing failed.
  • Why it's a collective action case: Interdependent remediation — your systems are only safe if your suppliers' and your counterparties' systems are too, so the incentive to disclose your own vulnerability status was low and disclosure happened anyway. And it is the cleanest available example of the problem that successful prevention is indistinguishable from there having been no problem, which has made Y2K a byword for hysteria.
  • The diagnostic:
    • How many players? Every organization with a computer, but the binding dependencies were in a few sectors — banking, power, telecoms, aviation — each with a small number of large players and an existing regulator.
    • Can you see defection? Yes, through mandated disclosure: US public companies were required to report Y2K readiness to the SEC, and regulators audited banks and utilities directly.
    • Can you punish it? Yes — regulatory action, and liability. The 1998 Year 2000 Information and Readiness Disclosure Act went the other way, granting limited immunity to encourage firms to share what they knew.
    • Will you meet again? Yes, and the deadline was fixed, public and known years in advance — an unusually favorable feature.
    • Is anyone big enough to act alone? No, but sector regulators were, and used it.
    • Substitutes? Not applicable. There was one fix and it had to be done.
  • Could serve: Session 1, Introduction to collective action, as a success where the success is invisible — which also sets up why the record of cooperation looks worse than it is. Session 4, Did it ever work?
  • Media: Extensive news archive footage of the rollover night. The Commerce Department report "The Economics of Y2K and the Impact on the United States" is online. No good documentary known.
  • Source: US Department of Commerce, The Economics of Y2K and the Impact on the United States, 1999. Government Executive, "Y2K costs government, businesses $100 billion", November 1999.

Internet standards: IPv6, DNSSEC, RPKI

  • What happened: Three coordination problems in the same system, with three different outcomes. IPv6 was specified in 1998; Google measured native IPv6 access crossing 50% of its users for the first time in March 2026, 18 years after it began tracking. DNSSEC, specified in the 1990s, remains at roughly 4% of domains signed and 13% resolver validation as of early 2026. RPKI route origin validation, a much later effort, reached 67% of announced prefixes by June 2026, although only 12% of autonomous systems fully validate.
  • Why it's a collective action case: Pure coordination problems with network externalities — deploying early costs you and benefits everyone else, and the benefit to you arrives only when others deploy. Three variants of the same structure with sharply different adoption curves, which is what makes this worth a slide rather than a sentence.
  • The diagnostic:
    • How many players? Tens of thousands of network operators, no central authority, standards set by an open body (IETF) with no enforcement power at all.
    • Can you see defection? Completely. Deployment is publicly measurable and several organizations publish live per-network scoreboards.
    • Can you punish it? No, and yet adoption happens. Naming and ranking is the only lever, plus procurement requirements from large buyers.
    • Will you meet again? Yes, permanently and visibly — operator communities (NANOG, RIPE) are small, repeated and reputationally dense.
    • Is anyone big enough to act alone? Partly. Cloudflare, Google and the large clouds can shift a measured percentage by themselves, and did so for RPKI.
    • Substitutes? Yes, and this explains the difference between the three. NAT was a good enough substitute for IPv6 addresses, which is why IPv6 took eighteen years. There was no comparable substitute for route-origin validation once BGP hijacks became routine, and RPKI moved fast.
  • Could serve: Session 2, The games, as a coordination problem rather than a dilemma — the distinction the module is built on, with real data attached. Session 4, Polycentricity, since internet governance is the working example of stacked authority with no sovereign.
  • Media: Live dashboards: Google IPv6 statistics, ISOC Pulse, NIST RPKI Monitor, APNIC DNSSEC world map. All public, all updating, all good for demonstrating in a session.
  • Source: Google IPv6 statistics (google.com/intl/en/ipv6/statistics.html). Internet Society Pulse, "18 Years Later, IPv6 Reaches Majority", April 2026. Kentik, "Exploring the Latest RPKI ROV Adoption Numbers".

Radio spectrum and the ITU

  • What happened: Spectrum is allocated internationally through World Radiocommunication Conferences, held roughly every 4 years, which revise the ITU Radio Regulations — a binding international treaty. WRC-23 met in Dubai in November–December 2023; WRC-27 is next. The system has run continuously since the 1906 Berlin conference and has coordinated a genuinely rival resource across every state on earth for over a century, largely without incident.
  • Why it's a collective action case: A subtractable commons where physics enforces rivalry — two transmitters on the same frequency in the same place destroy each other's signal. Arguably the oldest continuously functioning global commons regime, and nearly invisible because it works.
  • The diagnostic:
    • How many players? 190-odd administrations, plus satellite operators.
    • Can you see defection? Yes, immediately and unambiguously. Interference is self-announcing and can be geolocated. This is the strongest yes on question 2 in the entire list.
    • Can you punish it? Formally weak — the ITU has no sanction. In practice, interference is reciprocal, so defection punishes itself automatically.
    • Will you meet again? Yes, on a fixed multi-year cycle with continuous technical committee work between.
    • Is anyone big enough to act alone? No, and it doesn't matter, because nobody benefits from acting alone.
    • Substitutes? Partly. Technical change — spread spectrum, frequency reuse, higher bands, fiber — keeps expanding the effective resource, which repeatedly defuses allocation fights before they harden.
  • Could serve: Session 4, Did it ever work?, as the boring successful case that nobody cites. Session 3, Ostrom and the commons. Session 2, The games, on coordination versus dilemma — spectrum is mostly coordination, which is why it's tractable.
  • Media: ITU publishes WRC documentation and frequency allocation charts; the US NTIA spectrum allocation chart is a famous and genuinely striking poster. No documentary known.
  • Source: ITU Radio Regulations and WRC-23 Final Acts (itu.int). Hazlett, The Political Spectrum, Yale, 2017, for the critical account.

Economy and attention

Tax havens and the global minimum tax

  • What happened: Decades of tax competition drove statutory corporate rates down globally and moved reported profit to low-tax jurisdictions. The OECD/G20 Inclusive Framework agreed Pillar Two in 2021, a 15% global minimum effective tax on multinationals with turnover above €750m, implemented through top-up taxes that a home or intermediate jurisdiction collects if the source jurisdiction does not. Rules began applying in 2024 across the EU, UK, Japan, South Korea and others. In January 2026 the OECD released a "side-by-side" package allowing US-parented groups to elect out of the main charging rules from fiscal years beginning on or after 1 January 2026, following G7 agreement in 2025.
  • Why it's a collective action case: A textbook race to the bottom, with a clever structural fix — the top-up tax makes defection unprofitable for the haven, because the revenue is collected elsewhere if the haven doesn't collect it. Then the largest player negotiated an exemption, which is the part worth teaching.
  • The diagnostic:
    • How many players? 140-odd jurisdictions in the Inclusive Framework; a small number of havens hold most of the shifted profit.
    • Can you see defection? Much better than before. Country-by-country reporting since 2016 made profit shifting visible for the first time.
    • Can you punish it? Yes, and this is the innovation. The under-taxed profits rule means non-participation doesn't protect the haven's revenue base — a design that makes the agreement self-enforcing rather than sanction-dependent.
    • Will you meet again? Yes.
    • Is anyone big enough to act alone? The US is, and demonstrated it in 2025–26 by obtaining a carve-out rather than complying.
    • Substitutes? Not applicable directly; the relevant question is whether havens have another revenue model, and mostly they don't, which is why they resisted.
  • Could serve: Session 4, Changing the payoffs — the best modern example of redesigning a game so that defection stops paying. Session 2, The many names (race to the bottom). Session 4, Did it ever work? for the qualification.
  • Media: The EU Tax Observatory's Global Tax Evasion Report has good profit-shifting visualizations. Zucman's The Hidden Wealth of Nations (2015) is the accessible book.
  • Source: OECD, Global Anti-Base Erosion Model Rules (Pillar Two), 2021. OECD, Minimum Tax Implementation Handbook. Gabriel Zucman, The Hidden Wealth of Nations, Chicago, 2015.

Attention, social media and the under-16 age limit

  • What happened: Australia's Online Safety Amendment (Social Media Minimum Age) took effect on 10 December 2025, requiring platforms including YouTube, X, Facebook, Instagram, TikTok, Snapchat, Reddit, Twitch, Threads and Kick to take reasonable steps to prevent under-16s holding accounts, with penalties up to A$49.5m. Denmark, Malaysia, New Zealand and the European Parliament have moved in the same direction since; the European Parliament passed a non-binding resolution favoring a minimum age of 16 in November 2025.
  • Why it's a collective action case: Adolescent social media use is a collective action trap at the level of individuals — no teenager gains by leaving a platform their friends are on, and no parent gains by unilaterally banning it. This is the structure Haidt calls a collective action problem explicitly, and the legislation is an attempt to solve it by coercion rather than persuasion, which is a Hobbes/Hardin move applied to a social norm.
  • The diagnostic:
    • How many players? Millions of families, which is why the norm approach fails; a dozen platforms, which is why the regulatory approach might not.
    • Can you see defection? Between families, yes and visibly, which is exactly the pressure that sustains the trap. Platform compliance with age assurance is much harder to verify.
    • Can you punish it? Now yes, for platforms. Not for individuals, nor should it be.
    • Will you meet again? Yes, continuously — school communities are dense repeated-interaction environments, and coordinated phone-free school policies are the local version of the same fix.
    • Is anyone big enough to act alone? No individual family. One national government can act, and Australia is testing whether one is enough to start a cascade.
    • Substitutes? Partly — the question is whether there's an alternative social infrastructure for teenagers, and mostly there isn't, which is what makes unilateral exit so costly.
  • Could serve: Session 1, The pattern beneath the metacrisis — this is the case that connects the structure to the audience's own life, and the arc names exactly this ("it makes sense to me to sit in my room and watch Netflix"). Session 4, Coercion. Session 2, The games.
  • Contested: The empirical claim that social media causes the adolescent mental health decline is disputed — Haidt argues it, Odgers and Przybylski dispute the strength of the evidence. The collective action structure holds independently of who wins that argument, and it's worth separating the two in front of an audience that will have opinions.
  • Media: Australian government fact sheets and eSafety Commissioner guidance are primary. Haidt's The Anxious Generation (2024) is the popular text and makes the collective action argument explicitly. Good coverage in CSMonitor and CNBC, December 2025.
  • Source: Australian Department of Infrastructure, "Social media minimum age". Jonathan Haidt, The Anxious Generation, Penguin, 2024. Candice Odgers, "The great rewiring: is social media really behind an epidemic of teenage mental illness?", Nature 628, 2024.

Notes on coverage

Two things the list is short on, deliberately flagged rather than padded:

  • Non-Western cases. Nearly everything here is a treaty negotiated in or an industry based in North America, Europe or East Asia. The ASEAN transboundary haze agreement and the Nile Basin Initiative would both fit the diagnostic and neither has been researched here.
  • Successes at intermediate scale. The list jumps from Wikipedia to global treaties. City- and region-level cases — congestion charging in London, Stockholm and Singapore, where a coordination problem was solved by a single authority against initial public opposition and then became popular — would fill the gap between Ostrom's villages and the planet.
Edit this page

Links to this page:

  1. Example candidates — review index
Built with LogoFlowershow