In August 2026, Phil Magness published a citation audit of the Stanford Encyclopedia of Philosophy’s entries on capitalism and socialism, coding the cited authors by stance and finding the two bibliographies close to mirror images: each leaning against capitalism and toward socialism. Codings of that kind are always contestable. But the episode raises a question worth more than a scorecard: what would a fair pair of reference entries even look like?
Balancing the citation roster is one answer. These entries try a different one. They enforce symmetry where inference actually happens: the same evaluative outcome space for both systems; the same rule for when an outcome counts as evidence about a system at all; the same distinction between categorizing a formation and attributing an outcome to it; and worked inferences that run an announced five-step evidential test to bounded verdicts – one adverse for a socialist formation (Yugoslav soft budget constraints), one favourable for a capitalist financial architecture (venture capital and innovation), and one favourable for a socialist institutional element (Uruguayan worker-managed firms). Each entry confronts its own boundary cases rather than legislating them away, and each says what further evidence its claims would need before travelling.
The entries
- Capitalism – a structured institutional family organized around recurrent accumulation, with explicit conditions for when a formation qualifies, boundary cases from merchant capitalism to market socialism, and the venture-capital worked inference. (Markdown)
- Socialism – aspirations, projects, and formations kept distinct, a layered rule for what counts as socialist, the socialist calculation debate from Mises and Lange onward, and the paired Yugoslav and Uruguayan worked inferences. (Markdown)
How they were made
Both entries were drafted and revised in collaboration with large language models (OpenAI’s GPT and Anthropic’s Claude), with adversarial machine review from further models, and the process was gated at each step by verification: every bibliographic entry was checked against publisher and journal records (five corrections resulted), the empirical claims of the worked inferences were audited against the cited sources with page-anchored quotations, and negative claims are stated as searches rather than assertions. I am responsible for all claims, arguments, judgments, and errors.
The rules of categorization that the entries rely on have also been formalized in Alloy and machine-checked. Every boundary claim has a verified witness; four definitional checks pass; and one check, written so that it should fail, guards a contradiction that an earlier draft actually contained. A companion page, The categorization rules, machine-checked, presents the results, what such a check does and does not show, and the model source.
Corrections
Corrections and substantive challenges are welcome: brett.reynolds@humber.ca. The entries are versioned; the history lives in the site repository.