About
Twenty models of strategic behaviour, each stated formally, implemented as a working figure, and set against the empirical work that tested it.
What this is
Game theory is usually taught through models that were never measured. This collection takes the opposite selection rule: every paper here describes a model that somebody went out and tested, against lizards, bacteria, sparrows, dung flies, fig wasps, road networks, oil leases or laboratory subjects.
Each paper states the model formally, implements it as a figure you can drive, reports what the model predicts and who established it, gives the empirical work by name and date, and then says where the model is weak. That last section is not a formality. Most of these models are weak somewhere, and a collection that only reported the wins would be worth less than one that did not exist.
The figures are simulations rather than animations. Every control changes a real parameter and every reported number is computed from the model during the session, so the equilibrium values quoted in the readouts can be compared directly against what the running model produces.
Who made it
Built by Leif Heaney. It is a companion to the philosophy game theory hub, which covers games with a philosophical literature behind them rather than an empirical one. The two collections are deliberately separate: no model appears in both, and this site does not assume you have read the other.
Corrections are welcome and are the most useful thing anyone can send. The source, including the generator, the audit suite and the verification logs, is at github.com/leifheaney5/measured-games. Contact and issues.
How it is built and checked
The pages are generated from a set of build scripts rather than written by hand, so a change to the shared layout or a correction to a shared reference propagates everywhere at once. The output is static HTML with no runtime dependencies.
Nine automated check groups run over the whole site: that the published output matches a fresh build, that internal links and assets resolve, that social metadata and structured data are present and valid, that the on-page verification statements match the claims log, that no CSS declaration is being silently discarded by the browser, that layout holds at three widths with no overflow or collapsed blocks, that axe-core reports no WCAG 2.1 AA violations, that every control and hover target works, and that each simulator still reproduces the theoretical result its page claims.
That last group is the one worth naming. It asserts the science, not the rendering: that Braess returns 80 minutes open and 65 closed, that rent dissipation matches V(N-1)/N, that foragers converge on the predicted equilibrium proportion. If a model is broken, the build fails.
What has been checked
All references have been verified against Crossref: every DOI resolves and points at the work it claims to. That check found five errors, three of them DOIs wrong by a single digit that resolved to different articles in the same journal while the printed citation looked correct.
The prose is a separate matter and is only partly checked. 9 of the 20 papers have had their claims read against the sources they describe, and eight of those 9 required a correction. Every paper states its own status next to its references, and the contents page explains the distinction. Papers that have not been audited say so.
The errors found so far share a shape: none were invented citations, and all were plausible elaborations of a real result. A number made more impressive, a finding stated in the direction that feels intuitive rather than the one reported, a live debate presented as settled. None of that is detectable by reading a page and finding it convincing.
Licence
The prose, figures and generated pages are released under Creative Commons Attribution 4.0. You may share and adapt them, including commercially, with credit.
The source code is released under the MIT licence. Spectral, Archivo and IBM Plex Mono are used under the SIL Open Font Licence. The full licence text covers all three.
Neither licence extends to the works cited in the references, which belong to their publishers. Each figure offers its underlying series as a CSV download, and that data is covered by the same Attribution licence as the rest of the site.
Colophon
Set in Spectral for reading, Archivo for headings and labels, and IBM Plex Mono for the formal models and data. One muted print colour per discipline. The mark is the letter M drawn as a plotted series, with data points at the vertices and the centre valley as the equilibrium.
Figures are drawn on HTML canvas with no plotting library. Charts respond to hover with a crosshair and are reachable from the keyboard: focus a chart and the arrow keys read values along the horizontal axis aloud. Under reduced-motion preferences nothing starts on its own and running simulations step at a quarter rate.
Served as static files from a Caddy container. No analytics, no cookies, no trackers. The only third-party request any page makes is for the fonts.