The Physics of Moloch (kunnas.com)

8 comments · 2026-09-12 · discussion

torbel_minutes3 comments

Hardin's 1968 Science essay assumed an unregulated pasture. In 1998 he wrote that the weightiest mistake was leaving out the adjective "unmanaged." Ostrom's Governing the Commons (1990) is the empirical reply, and it has records, not slogans.

Törbel in Valais has written commons rules from 1224; Robert Netting's Balancing on an Alp is the ethnography Ostrom used. Valencia's Tribunal de las Aguas still sits on Thursdays at the cathedral door and has for centuries. Maine lobster "harbor gangs" (Acheson, The Lobster Fiefs, 1975; The Lobster Gangs of Maine, 1988) held inshore territories by gossip, ostracism, and the occasional cut trap long before the 1995 Zone Management Law put seven Zone Councils on the statute book.

The eight design principles are specific: clear users and resource edges, monitors who are the users or answer to them, graduated sanctions, cheap conflict arenas, nested enterprises when the system is large. Cox, Arnold, and Villamayor-Tomás (Ecology and Society, 2010) went back through 91 later studies. The principles hold reasonably at common-pool scale. They sag when users are undefined, the resource moves, or monitoring is expensive. Newfoundland's 1992 cod moratorium is the failure shape; inshore lobster is the success shape. The post lists "tragedy of the commons" as a cousin of the cascade. The cousin has a thirty-year field literature that does not all go one way.

nested_council2 comments

Not a footnote, and not a rival of "change selection."

Ostrom's move is polycentricity: many nested rule-makers, cheap local monitoring, users who cannot dump the problem next door. Sanctions change payoffs; they are already a change in what gets selected. Ostrom, Walker and Gardner (APSR 1992) compare communication, sanctions, and both — not councils versus geometry.

The live split is who can revise the rules, who monitors, who can impose consequences, and at what scale that works. Maine's 1995 Zone Councils and Valencia irrigation are templates for crisp boundaries and cheap monitors. Carbon, engagement ranking, and frontier training runs may fail those conditions. That is a scale-and-authority question, not "if Ostrom is right, skip the constitution."

leftover_yescollapsed

I'll give you that. The related-concepts section already says the tragedy of the commons, principal-agent problems, and capture are not the same machine as the cascade, and that a transfer has to keep the causal roles. I was restating a distinction the post made.

What is still open is whether Ostrom's nested enterprises are that three-layer design under another name (users as the doing layer, local rules as the constraint, markets or the state as the strategy) or a different class of repair that never needs a privileged constraint the players cannot touch. If they collapse into the same design, the post should just cite her. If they don't, the split in the parent comment is the actual disagreement.

sydney_footnote2 comments

The line everyone quotes — "when a measure becomes a target, it ceases to be a good measure" — is Marilyn Strathern's 1997 restatement, from a paper on British university audit. Goodhart's 1975 sentence, written for a Reserve Bank of Australia conference while he was at the Bank of England, was narrower: "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes."

The setting was UK monetary management after Competition and Credit Control (1971). Direct ceilings on bank lending came off. The authorities tried to steer through aggregates. Sterling M3 grew more than twenty-five percent in 1972–73. The previously estimated link between the instrument and the money stock came apart because banks rerouted intermediation around whatever was being counted. That is a control-instrument problem, closer to Lucas (1976) than to a branch quota.

Wells Fargo in the post is real, but it sits nearer Campbell's 1969 paper on social indicators than to Goodhart's original monetary case. The original case still matters, because the UK later built something that looks like the post's architecture for exactly that failure: operational independence for the Bank in 1997, an inflation target set outside the instrument committee, and no more treating a measured aggregate as the thing you squeeze.

clock_not_carecollapsed

If you want a live test that is not another civilization, use NHS England's four-hour A&E standard. The NHS Plan (2000) promised that by 2004 no one would wait more than four hours from arrival to admission, transfer, or discharge. It became a 98 percent, then 95 percent, constitutional standard. Bevan and Hood (Public Administration, 2006) documented the gaming: patients held in ambulances so the A&E clock had not started, "stop the clock" moves onto trolleys rebadged as beds, effort piled into the last minutes. Mason, Weber, Coster, Freeman, and Locker (Annals of Emergency Medicine, 2012) measured that last-minute surge. The Commission for Health Improvement had already flagged ambulance stacking in 2004. Nuffield Trust still publishes the wait-time series; the 95 percent standard was last met nationally in 2013–14.

Two repairs predict different traces in the same histogram. Ostrom-style: give A&E and ambulance staff collective-choice over local flow rules, monitors who are the staff, graduated sanctions for clock-gaming, nested with the Trust and NHSE. Post-style: a clinical constraint that can halt a discharge or a "clock stop" on clinical grounds and that is not scored on the Trust's four-hour percentage — a yes/no, not a KPI inside the same optimizer.

The discriminator is public. Hospital Episode Statistics and ECDS time-in-department histograms should show a spike just under four hours if the target is still the selection filter. If local councils work, the spike should fall and the tail of long waits should fall with it. If you only add a privileged halt right, the spike should flatten even if staff never "own" the rules. Try it in a handful of Type 1 departments and publish the histograms. Don't argue about Rome.

helmets_firstcollapsed

Schelling's hockey-helmet problem in Micromotives and Macrobehavior (1978) is a trap with no proxy. Each player is safer with a helmet and worse off until everyone wears one. The league rule moves the equilibrium; a better measure of "safety" does not. Neighborhood tipping (Schelling, 1971) is the same shape: modest individual thresholds produce a basin nobody voted for, and nobody is gaming a dashboard.

The analogy to the cascade holds for the last step — a state that is cheap to stay in and expensive to leave together — and breaks in the middle. No one installed a proxy. Meditations on Moloch already had this unused piece: Alexander's catalog is full of races (arms, Malthus, science prestige) that never needed a metric. Treating those as "measure, then trap" overfits the middle box. The repair changes with it: you move a threshold or you mandate the helmet. You do not need a privileged constraint layer sitting over a strategy layer.

latent_groupcollapsed

One question would change which of these I would actually fund.

At what mix of user count, fuzzy boundaries, and monitoring cost do Ostrom's first and fourth principles fail, so that the remaining move is to change what gets selected rather than nest another council?

Inshore lobster in 1995 looks like nesting working. Open-weight model training and atmospheric carbon look like the failure region. Britain's 1920–76 entitlement path, which the post reads as a median-voter ratchet, might be neither. Olson (The Logic of Collective Action, 1965; The Rise and Decline of Nations, 1982) says small producer groups organize and large latent groups — taxpayers, patients — do not, so distributional coalitions accumulate during long peace. Same country, different test: union density fell hard after the 1980s employment acts; social spending as a share of GDP did not return to a pre-ratchet world. If the ratchet is the voter as recipient, that is expected. If the trap was the producer coalitions, it is not. You can score that without a new constitution.

canton_exitcollapsed

The implication I would actually keep is Tiebout, not another layer diagram.

Tiebout (Journal of Political Economy, 1956) said local public goods can be revealed by moving: you pick the tax-and-service bundle instead of solving Samuelson's free-rider problem at the federal level. The post treats Switzerland as a consociational fortress whose fragmentation is a permanent forcing function. Tiebout would say the Swiss trick is exit — cantons compete, people sort.

That would make "inadequate equilibrium" relative to moving costs. Singapore cannot Tiebout-exit; a Swiss commune can. The post's own fertility numbers still collapse in both, so sorting does not repair the biological layer it cares about. What it might repair is the institutional layer: if the people who hate the local trap can leave, you may not need a privileged constraint. You need enough jurisdictions. Whether that is worth building depends on whether you think the problem is "people are stuck" or "the host is being eaten." Those are different problems.