Why the rules look weird

The most important systems are not designed to optimize performance. They are designed to prevent predictable failure. Most rules in those systems describe the normal case. The handful that do not (that interrupt, that constrain, that pre-resolve a specific exploit before it can happen) are the rules carrying the system’s actual integrity. Sports rules are the cleanest place to see this discipline because the system is visible, the failure modes have been observed at scale, and the rule book has had decades of pressure-testing under the most adversarial conditions any rule book ever faces.

Five rules. Five structural lessons.

The Infield Fly Rule

Baseball. With runners on first and second, or bases loaded, and fewer than two outs, the batter hits a fair fly ball that an infielder can catch with ordinary effort. The umpire calls the batter out immediately — whether or not the fielder catches the ball. Almost every new fan finds this rule confusing. It exists because without it, an infielder could intentionally drop the ball, force the runners into uncertainty about whether to advance, and convert one out into a double play the offense had no defense against. The rule does not describe normal play. It pre-resolves a known asymmetry that would otherwise reward an intentional drop.

That move (pre-resolving an exploit before it can be exploited) is the most underused discipline in institutional design. Every governance document I have read at scale describes what people should do. Almost none of them name the specific exploits the structure would otherwise reward, and pre-resolve them. The Infield Fly Rule is what a rule book looks like when someone has actually thought about what the rules are for: not to describe the obvious, but to close the loopholes the obvious would otherwise create.

The Free Guard Zone

Curling. For the first four rocks of every end, no stone in the free guard zone (the area between the hog line and the house, outside the rings) can be removed from play. The rule was added in the 1990s after every team converged on the same strategy: take out, take out, take out. The dominant strategy worked. It also collapsed the game into low-scoring sterility. The rule was added not to tell players what to do, but to prevent rational optimization from destroying the thing the game was built to be.

This is Goodhart’s Law as architecture rather than warning. (Originally Charles Goodhart’s 1975 observation that statistical regularities collapse once they are targeted for policy, popularly reformulated by Marilyn Strathern in 1997 as “when a measure becomes a target, it ceases to be a good measure.”) Every metric eventually gets gamed. Every KPI eventually rewards the behavior that hits the number rather than the behavior the number was meant to encourage. Every growth-at-all-costs strategy eventually hollows the product. The institutional discipline is not to ask whether the metric will be gamed; it will. The discipline is to write the rule that prevents the gaming from destroying the thing the metric was meant to measure. A funder whose impact metric inadvertently rewards risk-averse program design needs the Free Guard Zone equivalent: the rule that prevents the rational pursuit of the metric from killing what the metric was meant to encourage. Almost no funder has that rule written down.

Measurement scientists have a name for the pattern the Free Guard Zone was written to prevent: construct-irrelevant strategies — responses that hit the score without demonstrating the thing the score was built to measure. Samuel Messick built the modern validity framework around exactly this failure mode. The AI benchmark community has been rediscovering it under other names — reward hacking, specification gaming, Goodhart taxonomy — as models learn to score high on evaluations without learning the underlying skill. It is the same problem in a different domain, and it is the reason the weird rule has to be written before the metric is deployed.

Soccer Offside

A player is offside if they are in the opposing team’s half AND closer to the opposing goal line than both the ball and the second-to-last defender when their teammate plays the ball. Generations of new fans have asked: why is this a rule? The answer is that without it, the optimal strategy is to camp a fast striker near the opposing goal and play long balls over the top. The midfield empties. The game dissolves into two penalty boxes and a void in between. The structure of the game disappears in pursuit of the most efficient way to score.

Offside preserves the distributed structure of the game. It forces play through the midfield. It rewards the coordinated movement that is the game. The institutional parallel is direct: organizations that do not have rules preventing the cherry-pick (the function that bypasses the structure because it can, the team that goes around the process because it is faster) eventually discover that the structure has dissolved. The rule that prevents cherry-picking is not a rule about how to play. It is a rule about preserving the playing field. Every cross-functional process I have watched fail, has failed because the rule that protected the structure was not written. The cherry-pickers won, and the field collapsed.

The Advantage Rule

Soccer and rugby. When a foul occurs, the referee can choose not to enforce it if enforcing it would penalize the team that was fouled — if, for example, the fouled team is in the middle of a promising attacking move that a free kick would interrupt. The rule exists because rigid enforcement of process compliance can defeat the purpose of the process. The fouled team is the team the rule was written to protect; enforcing the foul would punish them. The referee has the authority to say: not now.

This is the rule the AB audience lives. Foundations enforce reporting requirements that consume the program officer’s week and prevent the program officer from doing the work the reporting was meant to evidence. Healthcare networks enforce documentation rituals that pull clinicians away from the clinical work the documentation was meant to track. Schools enforce compliance audits that consume the principal’s bandwidth and prevent the principal from supporting the teachers the audit was meant to evaluate. The Advantage Rule names the discipline that resolves this: process compliance is not the same as serving the mission. The referee’s authority to wave off the foul, when enforcing it would defeat the foul’s purpose, is what keeps the institution serving its mission rather than its bureaucracy. Most institutions do not have this rule written down. The senior people who run the institution apply it informally (the program officer who quietly skips a step that would derail a grantee, the dean who lets a faculty member miss a deadline that would damage the research), and the system depends on their discretion to function. The rule that names that discretion explicitly, and gives the operator the formal authority to exercise it without breaking the rules, is one of the most useful rules an institution can have, and almost no institution has it.

The Baton Exchange Zone

Track and field, the 4×100 relay. Runners must complete the baton handoff within a 20-meter exchange zone. The fastest team does not win. The team with the cleanest handoffs wins. The interface (the place where one runner meets the next) is where the race is decided. A team with four world-record-holding individual sprinters and a sloppy handoff loses to a team with four ordinary runners and a clean one. The rule structures the interface and makes the interface itself the locus of accountability.

Every organizational failure I have watched up close has happened at an interface. The handoff between admissions and financial aid. The handoff between the clinician and the population-health analyst. The handoff between the program team and the impact-reporting team. The handoff between the data engineer and the governance committee. The work inside each function is rarely the problem. The handoff between them almost always is. The Baton Exchange Zone is the rule that says: name the interface, structure the interface, make the interface itself the place where accountability lives. Most institutional process maps show the boxes. They do not show the handoffs. The rule book that shows the handoffs (and gives the handoff its own structured zone with its own rules) is the one that wins races.

Five structural lessons

Five rules. Five structural lessons. None of them describes normal play.

The rules that matter most in any decision system are not the ones that tell people what to do. They are the ones that prevent optimization from destroying the thing the system was built to preserve. The Infield Fly Rule pre-resolves an exploit. The Free Guard Zone prevents convergent over-optimization. Offside preserves distributed structure. The Advantage Rule defends the mission against process. The Baton Exchange Zone structures the interface.

Behavioral rules — the clarification

A clarification before that lands as the wrong claim. Behavioral rules (the ones that tell people what to do inside their function) are not optional. A relay team that cannot run does not win because the handoff was clean. A funder whose program officers cannot write coherent goals does not get rescued by a Free Guard Zone equivalent. Behavioral rules are the baseline; they have to be in place and executed for the structural rules to matter at all. What this essay names is the layer above that baseline: the rules that govern what happens between functions, the rules that pre-resolve the exploits behavioral compliance alone cannot prevent. Most institutions have the behavioral layer; what they typically lack is the structural layer. The argument is for both, with attention to the half that is usually missing.

The rule book that works

Every institution has these rules implicit somewhere. Most do not have them written. The board policies describe how to make decisions; rarely do they pre-resolve the exploits the structure would otherwise reward. The strategy documents describe how to grow; rarely do they name the rule that prevents the growth metric from hollowing the product. The org charts describe the functions; rarely do they show the interfaces between functions, where the actual accountability has to live. The compliance frameworks describe the requirements; rarely do they include the operator’s authority to wave them off when they would defeat their own purpose.

When a leader asks me what they should be writing down that their predecessors did not, this is the answer. Not more policies describing the normal case. The handful of weird-looking rules that pre-resolve the exploits, prevent the optimization collapse, preserve the distributed structure, protect the mission from the process, and make the interfaces themselves accountable. Those rules will not look like a strategy document. They will look like a baseball umpire calling a batter out for a ball that was never caught. That is what working governance looks like up close.

There is a corollary worth naming. G. K. Chesterton put it as a heuristic a century ago: never remove a fence until you know why it was put there. The rule that looks weird is probably load-bearing. If you cannot say what predictable failure it was written to prevent, the safest assumption is that someone before you could — and that removing it will reveal, at cost, the exploit the rule was there to close.

The most important systems are not designed to optimize performance. They are designed to prevent predictable failure. The rules that do that work look weird at first reading, and they are doing the job.

Written July 2026 for the Analytic Bytes Library. The five rules are drawn from baseball, curling, soccer, rugby, and track and field; the institutional parallels reflect patterns observed across higher education, foundations, healthcare, and mission-driven organizations. Goodhart’s Law is included as the closest available warning about the pattern the Free Guard Zone was written to prevent.