Why Platform Rules Are Impossible to Agree On

Content moderation on large platforms is a notoriously thorny problem. But David Turner's simple web game, No Vehicles in the Park, demonstrates that even a trivial rule—"no vehicles in the park"—cannot be interpreted consistently by people. The game asks a short series of yes/no questions about whether various objects count as vehicles subject to the rule.

One of the most striking results is that people are wildly overconfident about their own interpretations. A highly upvoted comment on Hacker News claimed the answers were "crystal-clear" and that the exercise showed the opposite of what the author intended—that moderation rules are clear. Another commenter agreed, citing a "clear majority in the answers." But the actual data contradicts this confidence.

The Data Shows Fragmentation, Not Consensus

After completing the survey, users see a per-question bar chart of yes/no responses. It looks "wiggly," not uniform. The more telling analysis comes from the raw survey data, which David Turner shared upon request. When the data was pulled—shortly after the site was posted to Hacker News, meaning the respondent pool was a relatively homogeneous HN audience—the results were striking:

  • 36,902 people expressed opinions on what constitutes a vehicle in the park.
  • They produced 9,432 distinct opinion combinations.
  • The most popular combination of answers was held by just 11.7% of users.
  • The next two most popular positions—held by 8.5% and 6.5% respectively—differ from the top position only on whether a non-functioning WWII-era tank used as a memorial violates the rule.
  • The 4th through 7th most popular positions each had roughly 1% agreement, and every other position had less than 1%.
  • The median user agrees with only 0.16% of other users, meaning the average opinion was shared by about 3.9 people (roughly 0.01% of respondents).

To visualize this, ordering the positions from most to least popular and plotting on a log scale shows a rapid drop-off: a few points with noticeable support, followed by a long tail of near-unique opinions. There is no single interpretation, or even a small set of nearby interpretations, that garners widespread agreement. Michael Chermside articulated this well in a response to the "crystal-clear" commenter: the right question is not whether the rule is clear to you, but whether "different people produce the same answers." The smooth curve of responses shows they do not, even for this simple rule.

A Predictable Failure in Real-World Contexts

This overconfidence is prevalent. Most people assume that others share their "obvious" interpretations. This effect intensifies with emotionally charged issues. Consider the question of whether a specific athlete plays dirty. Many basketball fans would agree on a player like Draymond Green, yet a sizable group—Warriors fans—will find ways to justify every one of his plays. When a fanbase is emotionally invested, even questions with seemingly objective answers, such as whether stomping on an opponent's chest is acceptable, become fiercely contested.

Moving from sports to politics and identity intensifies the problem further. In her 1976 essay "Trashing," feminist Jo Freeman documented how the women's movement would "cancel" members for minute differences in opinion—a phenomenon that persists. Natalie Wynn's experience, detailed in a 2021 video, is a more recent parallel: she faced calls to be deplatformed, doxed, and assaulted for differences in position that were, to outsiders, barely distinguishable from those of her detractors. To describe the precise nature of that disagreement would take thousands of words, whereas a left-wing vs. right-wing political difference could be summarized in a sentence.

These "small" differences are what make charged issue spaces fractally contentious. People with near-identical views can hold the most intense animosity toward each other. For a platform, this means any set of moderation policies will fail to satisfy a broad user base. Users fundamentally disagree on what should be allowed, and the problem scales with the platform's size. Attempts to solve it with simplistic approaches invariably fail:

  • No moderation doesn't work for large platforms; they would quickly be overrun by spam bots, engagement rings, and scam content.
  • Allowing all humor leads to forums flooded with the lowest-common-denominator memes and repetitive jokes, which many users dislike.
  • Federation (as with the fediverse) merely shifts the disagreement to the server level. Users on different servers block different content, creating fragmentation where no single server offers access to everything a user wants. Even solving a simpler problem like consistent content display across different implementations (e.g., Mastodon and Honk) has proven impossible.

The lessons from "No Vehicles in the Park" are not abstract. The inability to agree on the application of a simple rule in a contrived, low-stakes scenario foretells the impossibility of universal agreement on real-world issues where personal values and emotions are deeply involved. A minimal set of "obvious" answers may exist in theory, but actual users find even that minimal set elusive. The only thing people reliably agree on is that everyone else should agree with them.