What Happened When I Built the Pattern Language

A follow-up to "Notes on the Planetary Civics Inquiry," published April 16, 2026.

On April 16th I published an essay arguing that a pattern language form might be closer to what planetary civics requires than a governance framework can be. The argument was theoretical. I had not built the language.

Then I built it. Twelve patterns, each with a context, a pattern statement, operations, and a test. I called it "A Pattern Language for Planetary Civics."

Then I tested it. Not against hypothetical scenarios but against authored documents: the Planetary Civics Inquiry position paper itself, a Dark Matter Labs financial architecture for urban trees, the EU AI Act, Anthropic's Responsible Scaling Policy, the GNU Manifesto, the UN Declaration on the Rights of Indigenous Peoples, the Rule of St. Benedict, the Tao Te Ching, Wikipedia's core content policies, Fidel Castro's "History Will Absolve Me," Steve Jobs' Stanford commencement address, three EZLN documents spanning eleven years of the same movement, Elinor Ostrom's design principles for common-pool resource governance, and several systems I had built or was building myself.

Twenty-four tests, two on documents I cannot publish. Four things happened that I did not expect.

The language renamed itself

After ten tests across planetary governance theory, regulatory instruments, financial infrastructure, algorithmic systems, frontier AI self-governance, and a free-software manifesto, the name "Planetary Civics" was wrong. The common thread across the tested artifacts was not planetary governance. It was systems that claim to represent, govern, or mediate reality under irreducible uncertainty. The EU AI Act has nothing to do with planetary civics. Anthropic's scaling policy has nothing to do with planetary civics. But both perform the same structural moves the patterns detect.

The language is now called "A Pattern Language for Systems That Exceed Their Own Comprehension." The rename was not a branding decision. It was a finding. The patterns were telling me what they were about, and it was not what I thought.

The strongest anti-reference was not a bad document

The original essay critiqued the Planetary Civics position paper for dropping Spivak's globe/planet tension when it moved into governance proposals. TreesAI is a different Dark Matter Labs project. The language tested it rigorously, and it scored lowest of any document tested: eight fails, two mixed. But what made TreesAI the clearest anti-reference was not that it is poorly made. It is sophisticated. It cites planetary language. It includes beyond-human stakeholders in its framing. It is authored by serious people working on real problems.

The failure is that every component of the operational mechanism individually performs a globe move. The outcomes contract cannot hold what the model does not predict. The sensor layer replaces perception with mediated legibility. The payment structure requires total representation to issue contracts. Holistic framing language at the top, mechanical metrics at the bottom, and no structural path between them.

The teaching function is specific: a system that names the planet in its preamble can still build a globe in its mechanism. The language reads mechanism, not vocabulary.

The highest-scoring document would not regard this language as useful

The Tao Te Ching scored eleven strong passes and one partial. Highest of any document tested. The partial was Pattern 9 (Situate the Speaker), caught by a genre-structural modifier the language developed for documents that suppress individual authorship by design.

The document that scored highest would not regard a pattern language as superior to itself. Chapter 38: "when the Tao was lost, its attributes appeared." Named categories emerge precisely when the underlying reality they point at has already weakened. By that logic, writing down patterns to teach perception is itself evidence that perception has weakened to the point where it must be named to be retained.

This is not a paradox to resolve. It is a condition to carry. The language is useful where pattern-recognition has already weakened enough that pattern-naming helps. That is most institutional documents under conditions of scale and pressure. The language is not useful to someone whose perception has not weakened to that point. The Tao Te Ching tests that person's perception directly. This language makes that person's perception testable.

Both are valid moves at different points of weakening. The language's value is specific to a range of conditions. Outside that range, it is not a worse tool. It is the wrong tool.

The language discriminates within a movement across time

Three EZLN documents tested in the same run -- the First Declaration of the Lacandon Jungle (1994), the Sixth Declaration of the Lacandon Jungle (2005), and the design document for the Good Government Juntas (2003) -- produced scores of 0/7/4/1, 4/7/0/1, and 6/6/0/0. Same movement, same speaker-position, eleven years apart.

The First Declaration is the movement's most authoritarian text: a declaration of war with six military orders, a single collective subject, no rejected alternatives named. The Good Government design is the movement's most structurally rigorous: "mandar obedeciendo" (governing by obeying) as architectural principle, rotation built into the design, a redistribution mechanism, and an oversight body explicitly designed to catch the governance system's own failure modes. The Sixth Declaration describes the same governance architecture in rhetorical-political register and scores between the two.

The finding is methodological: the language produces different results from the same source depending on document type. Design documents produce richer mechanism-level findings than manifestos. This is not a judgment on the movement. It is a finding about what the language can and cannot see -- and about where mechanism lives in different genres.

The comparison that mattered most: Ostrom's eight design principles for common-pool resource governance scored identically to the IETF's RFC 8890 (3/8/0/1). Two documents from different traditions, different domains, different centuries, facing the same problem of governing shared resources at scale -- the same profile. The language does not know their politics. It reads their mechanisms.

What the language does

The twelve patterns share a common commitment: systems exceeding their own comprehension should be built and interrogated with that exceedance named, rather than concealed behind the confidence of the documents that govern them.

The patterns are not rules. They are hypotheses about how to act in systems where not everything can be known, not everything can be represented, and not everything should be controlled. Each pattern has a test. The tests produce specific, mechanism-level observations about authored documents. The observations are prompts for judgment, not judgments.

Twelve patterns is a small language. Alexander's original contained 253. The difference is not ambition; it is evidence. A pattern language grows when tests at specific scales demonstrate that the existing patterns underfit: they cannot catch what needs catching without a new pattern. Twenty-four tests have not yet produced that evidence. Several tests produced findings that sharpened existing patterns (Pattern 11 gained an irreversibility clause, Pattern 12 gained a dependency test, Pattern 3 gained a sub-case, Pattern 9 gained a genre-structural modifier). None required a thirteenth pattern. The language stays small until the work demands otherwise. If it grows without evidence, it becomes the framework it was written to avoid.

The language includes a reference case: a financial reconstruction system I built that turns historical bank records and an AI classifier's output into posted journal entries through human review. The system predates the language. It was not built using the patterns. The language retrospectively recognizes it as pattern-aligned. That is a deliberate choice for the first reference case: the first proof the language owes a reader is that it describes something real, not that it produces good systems.

What this is not

The language is not a governance framework, a compliance tool, a theory of change, or a complete system. It interrogates individual artifacts. It does not tell anyone how to shift institutions at scale. If it becomes any of these, it has failed.

The closing of the original essay hoped that the Planetary Civics Inquiry would sustain Spivak's challenge longer than its opening paper did and consider whether Alexander's later work offers the nearest methodology. The language I built tests that hope against real documents. The results suggest the methodology is closer to what is needed than I expected, and further from finished than I hoped.

---

The language document

The full pattern language is published as a working document. It includes twelve patterns, a reference case, an anti-reference case, a mixed case, and a section situating the author's position and limits.

Download as PDF (v0.5.7)

Download as Markdown -- the Markdown version is provided so you can feed it to an LLM alongside a document of your choice. The language includes an application prompt at the end that operationalizes the twelve patterns for testing. Give the LLM both texts and let it apply the patterns. The interesting part is not the verdicts but whether the findings match, surprise, or miss what you already know about the document.

Test archive -- twenty-four test runs (two on internal or private documents shown as stubs), plus summary table and cross-test findings.