# A Pattern Language for Systems That Exceed Their Own Comprehension

**v0.5.7** Originally developed as "A Pattern Language for Planetary
Civics" in response to the 2024 Position Paper for the Planetary
Civics Inquiry. After twenty-four tests across planetary
governance theory, regulatory instruments, financial infrastructure,
algorithmic systems, frontier AI self-governance, free-software
manifesto, platform-power writing, monastic rule, ancient philosophy,
crowd-authored encyclopedias, and this document itself, the earlier
name was determined to be scope-misleading. The common thread across
tested artifacts is systems that claim to represent, govern, or
mediate reality under irreducible uncertainty.

## Preface

This is not a critique document. If it does not change how you would design or operate something you control, it has failed.

This document begins from a constraint:

> You cannot write guidelines for relating to what exceeds your
> comprehension without distorting it.

Most institutional artifacts resolve this by prioritizing
legibility. They produce frameworks, taxonomies, and governance
systems that make the world administrable. This is necessary work.
It is also a globe move: it renders reality into forms that can be
seen, measured, and controlled. (The globe/planet distinction is
borrowed from Gayatri Chakravorty Spivak.)

The problem is not that this happens. The problem is that it
happens without being named, and without preserving what is lost
in the process.

This document attempts a different form.

It does not offer a framework. It offers a pattern language: a
finite set of reusable patterns that can be combined, adapted,
and tested in practice. 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
- not everything should be controlled

The goal is not correctness. The goal is to preserve the conditions
under which reality can continue to correct the system.

This language is not methodologically neutral. Its patterns
collectively reward uncertainty-acknowledging,
plurality-preserving, judgment-scaffolding documents and penalize
their opposites. That is a designed feature, not a bug. It
reflects the underlying commitment that 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.

## How to Use This Language

1. Patterns are not prescriptive. They are prompts for judgment.
2. Patterns may conflict. Hold the conflict open rather than
   resolving it.
3. Patterns should be applied to real artifacts, not hypothetical
   scenarios.
4. The language is tested by whether it produces new, specific
   observations, not agreement.

## Pattern Set

### 1. Hold the Unresolved

**Context**
When making decisions about systems that cannot be fully
understood.

**Pattern**
Proceed without forcing all unknowns into assumptions.

**Operations**

- Maintain an explicit register of unknowns
- Distinguish modeled from unmodeled variables
- Track unknown-unknowns as a category

**Test**
Can the system absorb contradiction without breaking?

### 2. Do Not Collapse the Many

**Context**
When representing multiple actors, perspectives, or conditions.

**Pattern**
Preserve plurality without forcing convergence.

**Operations**

- Allow parallel, conflicting representations
- Avoid single metrics standing in for many realities
- Represent disagreement explicitly

**Test**
Are differences still visible, or have they been averaged away?

**Relation**
Pattern 2 supports Pattern 10: preserved plurality within a scale
prevents flattening across scales.

### 3. Follow What Is Alive

**Context**
When designing systems where human perception or judgment is the
load-bearing element (construction, interpretation, diagnosis,
algorithmic suggestion systems).

**Pattern**
Do not substitute proxies for direct recognition. Train perception
and let it guide action.

**Operations**

- Observe before modeling
- Allow judgments that cannot be fully explained
- Compare interventions by felt difference, not only metrics

**Test**
Does this make the situation more alive?

If the question feels illegitimate, the pattern is not being
followed.

**Note**
This pattern presupposes but does not yet specify methods of
extension. Christopher Alexander's "structure-preserving
transformation" (The Nature of Order, 2002-2005) sits implicit
here; specifying method in pattern form is deferred to avoid
premature institutional capture.

This pattern asks practitioners to compare interventions by felt
difference. The document in which it sits does not produce felt
difference. Its register is cognitive, structural, deliberately
affectless. This is an internal inconsistency, not an oversight.
The question of whether a pattern language can ask for perception
it cannot itself occasion remains open.

**Sub-case: recognition without design.**
A document may be N/A for Pattern 3 at the design level (the
document is regulatory or infrastructural, not perception-
load-bearing) while affirming the practice of Pattern 3 for
others as a protected right. UNDRIP Articles 25 and 31 do this:
"distinctive spiritual relationship with... lands, territories,
waters and coastal seas" is treated as a right to be protected,
not as a design discipline for the Declaration itself. For such
documents, note "N/A at design level, recognition-level
affirmation present." This is a pattern-adjacent move worth
tracking -- not a pass of Pattern 3 but also not nothing.

### 4. Keep the System Permeable

**Context**
When designing systems intended to operate over time.

**Pattern**
Allow unmodeled reality to enter and alter the system.

**Operations**

- Accept inputs outside predefined schema
- Enable revision beyond original authority
- Avoid closed decision loops

**Test**
Can something unexpected change the system?

### 5. Resist Total Representation

**Context**
When translating complex systems into models or governance
structures.

**Pattern**
Limit representation. Complement it with constraints that
acknowledge what cannot be modeled.

**Operations**

- Use thresholds, not only calculations
- Accept partial justification
- Avoid claims of completeness

**Test**
Does the system behave as if it fully understands what it
includes?

### 6. Refuse Premature Rightness

**Context**
When systems demand definitive answers under uncertainty.

**Pattern**
Act without claiming correctness.

**Operations**

- Treat outcomes as provisional
- Avoid irreversible commitments
- Make "this may be wrong" acceptable

**Test**
Can decisions stand without being defended as correct?

**Note**
The strongest form of this pattern is visible in documents that
revise themselves publicly over time, correcting specific prior
claims by name. The GNU Manifesto (Stallman 1985, with footnotes
added 1993+) is a reference case: "I think I was mistaken in
saying that proprietary software was the most common basis for
making money in software" -- a specific named revision to a
load-bearing claim, preserved alongside the original text. This
is pattern 6 at maximum timescale and maximum personal stakes.

### 7. Surface the Cost of Legibility

**Context**
When making reality measurable or governable. Where Pattern 1
concerns unknowns in the decision itself, Pattern 7 concerns
omissions in the representation the decision relies on.

**Pattern**
Make visible what is lost in translation.

**Operations**

- State what models cannot capture
- Pair metrics with omissions
- Name harms the system cannot perceive

**Test**
Do users understand what has been excluded?

### 8. Let Use Precede Understanding

**Context**
When working with systems too complex to fully specify upfront.

**Pattern**
Allow partial use before full comprehension.

**Operations**

- Deploy in low-risk contexts
- Learn through application
- Avoid blocking use on theoretical completeness

**Test**
Can it be used without being fully understood?

**Note**
This operates in tension with Pattern 7. Use precedes
understanding, but the cost of that compromise should still be
surfaced.

### 9. Situate the Speaker

**Context**
When producing frameworks, systems, or languages.

**Pattern**
Make the author's position and limits visible.

**Operations**

- State context, incentives, constraints
- Identify absent perspectives
- Avoid universal claims without qualification

**Test**
Can a reader identify where this is speaking from, and where it
is silent?

**Note on genre.**
Some document genres structurally suppress individual authorship
(UN General Assembly resolutions, international treaties,
standards-body outputs, crowd-edited encyclopedias). A Pattern 9
"partial" or "failure" for such a document reflects a feature of
the genre rather than a choice of the specific document. When
applying the test, distinguish "genre-structural Pattern 9 limit"
from "specific-document Pattern 9 failure." They are different
findings with different implications for revision: the first is a
characteristic to name; the second is a gap to close.

### 10. Work Across Scales Without Collapse

**Context**
When systems operate across nested scales.

**Pattern**
Relate scales without reducing them to one frame.

**Operations**

- Evaluate impacts independently at multiple scales
- Avoid single aggregated measures
- Preserve contradictions between scales
- When invoking an external scale as constraint, either examine
  it as a scale with its own dynamics or remove it from the
  argument

**Test**
Is one scale optimized at the expense of another, and is that
visible?

**Relation**
Pattern 10 extends Pattern 2 across nested systems.

### 11. Choose the Most Permeable Compromise

**Context**
When action must occur within reductive institutional systems.

**Pattern**
Select the option that preserves the greatest capacity for
revision and intrusion by reality.

**Operations**

- Prefer reversible mechanisms
- Avoid totalizing instruments
- Pair action with explicit distortion

**Test**
Does this keep the system open to change?

**Guard**
The test fails if you cannot name a more permeable alternative
that was rejected, and why.

The test also fails if the action forecloses future revision at a
cost the system cannot bear. Irreversibility raises the bar on
both the rejected-alternative clause and the permeability test
itself.

**Note on frame-acceptance.**
Pattern 11 operates inside the frame of analysis the document
accepts. The strengthened test catches named rejected alternatives
within that frame; it does not from inside judge whether the frame
itself is the compromise. A document may pass Pattern 11 at its
central compromise while a reader outside the document's accepted
frame would regard the entire compromise-structure as the
domination. UNDRIP's Article 46.1 passes Pattern 11 by naming
the territorial-integrity constraint on self-determination; from
a radical indigenous sovereignty position outside the state-system
frame the Declaration accepts, Article 46.1 IS the domination.
This limit is not fixable by sharpening Pattern 11. Name it where
it applies and do not over-claim what the pattern can see.

### 12. Build Scaffolds, Not Cages

**Context**
When structuring behavior through systems (regulation,
infrastructure, algorithms) without being able to rely on direct
perception.

**Pattern**
Support judgment without replacing it.

**Operations**

- Treat outputs as proposals
- Preserve override without full justification
- Allow per-context definitions of success
- Keep interpretation at the point of use

**Test**
Does the system make judgment easier, or unnecessary?

Second question: does the system make functioning without it
harder over time? A system that makes judgment easier today but
erodes the capacity to function without the system tomorrow is
still a cage, on a delay.

**Relation**

- Pattern 3: direct perception of aliveness
- Pattern 4: system-reality interface (permeability to what was
  not modeled)
- Pattern 12: human-system interface (preservation of judgment
  where direct perception is not possible)

Pattern 12 is a fallback, not a substitute. If someone uses it
where direct perception is available, they are drifting toward
the globe.

## Reference Case: Reconciliation Workbench

A financial reconstruction system that turns historical bank CSVs
and an offline classifier's output into posted journal entries,
through human review. It is a working implementation that the
language retrospectively recognizes as pattern-aligned. It was
not built using the language; it predates it. That is a
deliberate choice for this reference case: the first proof the
language owes a reader is that it is describing something real,
not that it produces good systems. The second proof is what
future use and exposure are for.

### Mechanism

Pattern 3 (Follow What Is Alive) is load-bearing in specific
schema decisions:

- `proposed_dr_account_id`, `proposed_cr_account_id`,
  `proposed_pc_id`, `proposed_narration` preserve the classifier's
  suggestion as immutable data
- `review_override_*` columns preserve the human's decision as
  separate data, not overwriting the proposal
- A `classifier_question` text field lets the machine admit
  uncertainty in schema form and pass the question to a human
- Level 3 readiness criteria ("human override rate below 15 percent
  on high-confidence proposals") tie system advancement to measured
  alignment with human judgment, not to date

Pattern 12 (Build Scaffolds, Not Cages) is load-bearing in the
permission and workflow architecture:

- Every staging row is explicitly a proposed journal entry, not a
  decision
- Override does not require justification (review_notes is
  optional)
- An investigator role permits contribution without authority to
  post, preserving judgment-without-control
- Bulk actions accelerate review but do not eliminate it

### What this demonstrates

Automation can be introduced into perception-load-bearing work
without eliminating the perception. The workbench makes judgment
faster (keyboard shortcuts, context stack, pre-populated proposals)
without making judgment unnecessary. Every posted row was approved
by a human.

Several other patterns also hold in this system (1, 2, 5, 6, 8,
11). The point is not exhaustive coverage. It is that Patterns 3
and 12 lock together where both apply, and the mechanisms that
make them hold are visible in the data model, not just in the
claims.

## Anti-Reference Case: Trees as Infrastructure (TreesAI)

A financial architecture from Dark Matter Labs that treats urban
trees as infrastructure, funded through outcomes-based contracts
where beneficiaries pay for measurable impacts (reduced stormwater
runoff, AC cost, heat island effect). Sensors ("Internet of
Trees") provide the monitoring layer.

This is the clearest anti-reference the language has produced so
far. Tested against the pattern set, the mechanism fails at the
component level:

- The outcomes contract cannot hold the unresolved (Pattern 1):
  what the model does not predict, the payer does not pay for.
- The sensor layer replaces perception with mediated legibility
  (Pattern 3): "Is this more alive?" becomes not a payable
  question, therefore not a legitimate question inside the
  instrument.
- The payment structure cannot operate on partial representation
  (Pattern 5): it must fully represent value to issue contracts.
- Legibility costs are concealed behind holistic language that
  names mechanical metrics and "acknowledges social and cultural
  co-benefits" without paying for them (Pattern 7).

The failure is not a corruption of the mechanism. The mechanism
is the failure. This is what happens when scaffolding becomes a
cage: every component individually performs a globe move, and the
whole is structurally unable to recognize what it has excluded.

The teaching function of this anti-reference is to make a specific
failure mode visible. A system that looks sophisticated, cites
planetary language, and includes beyond-human stakeholders in its
framing can still convert planetary multiplicity into contractual
legibility at every step of its operational layer. Naming the
planet in the preamble does not prevent building a globe in the
mechanism.

## Mixed Case: IETF RFC 8890 — "The Internet is for End Users"

A five-page informational RFC published by the Internet
Architecture Board (Nottingham, August 2020). Argues that when
IETF decisions produce conflicts between end users of the Internet
and other parties, the IETF should favor end users. Offers a
definition of end users, a justification grounded in the IETF
Mission Statement, and five mechanisms for prioritization.

This is the most instructive case the language has produced for
the ambiguous middle: a document that does genuine structural work
on several patterns and stops short on others for specific,
nameable reasons. Score: 3 strong pass, 8 partial, 0 fail, 1 N/A.

### Where the mechanism holds

Pattern 4 (Keep the System Permeable) passes strongly. The
permeability mechanism is architected as an ongoing invitation,
not a one-time gesture: "we should not require them to 'come to
us'... take the initiative to contact them, explain our work, and
solicit their feedback." "Surprising the Internet community is
rarely a good outcome." These are not rhetorical commitments;
they are operational postures with named consequences.

Pattern 12 (Build Scaffolds, Not Cages) passes strongly. The
document does not specify outcomes. It proposes consultation
processes and leaves decisions to working groups. It explicitly
names what it refuses to do: "architectural purity for its own
sake" is not a valid driver. Every mechanism is framed as a
proposal, not a rule. The dependency test produces a finding
worth noting: RFC 8890 has become a widely cited reference in
IETF and W3C deliberations. That is dependency formation. But
it is dependency on a shared vocabulary and consultation habits,
not on an algorithm or proprietary gatekeeper — structurally
closer to GNU than to the App Store.

### Where it stops short

Pattern 11 (Choose the Most Permeable Compromise) passes at the
technical layer and fails at the meta-level. For specific
technical compromises, the document names rejected alternatives
and documents reasoning. But RFC 8890 inherits the Priority of
Constituencies from HTML5 and silently simplifies it. HTML5's
original list is graduated: users above authors above
implementors above specifiers above theoretical purity. RFC 8890
collapses this to a binary — end users versus "other parties" —
without naming the graduation it simplified or surfacing the
cost: authors are flattened into the same category as equipment
vendors and governments. The rejected alternative to the binary
is what the source actually does. It was cited in
acknowledgements and then quietly narrowed.

Pattern 9 (Situate the Speaker) produces a specific finding.
The body whose protocols shaped the conditions of Internet-scale
power concentration frames its own causal influence as "protocol
design might have some influence upon" that concentration. The
posture is genuine humility; it is also under-examination of
causal reach. Named here as texture, not dismissal.

### What this demonstrates

The language discriminates on mechanism, not on vocabulary.
RFC 8890 claims as its stated virtue the exact disposition the
language rewards: humility, deference to affected communities,
refusal to specify outcomes. The score is not a rubber-stamp
(3 strong pass, not 11) and not a dismissal (0 fails). The
distinction between claimed and delivered virtue is visible in
the verdicts: Pattern 4 and Pattern 12, where the mechanisms
are operational, pass strongly. Pattern 11, where the inherited
framing goes unexamined, catches a specific failure that the
document's own humility does not surface.

This is what the ambiguous middle looks like. Not mediocrity
across the board. Genuine strength on specific patterns, genuine
gaps on others, each nameable in the mechanism.

## Situating This Document

This document is authored from a specific position:

- Intellectual formation: Western design and systems theory as
  primary lineage (Spivak, Alexander, Virno, with Morton, Bratton,
  Mbembe, Latour as referenced context). Educated at a Catholic
  university in the US, from a lineage of agnosticism and atheism
  — inside the institutional formation without the devotional
  formation. That position shapes exposure to the intellectual
  traditions (natural law thinking, subsidiarity, the Thomist
  infrastructure beneath much Western institutional theory) without
  producing deference to their theological grounding.
- Heritage and operational position: Mexican, of Spanish and mixed
  Mexican heritage, including Spanish civil war exile lineage on
  the paternal side — a position that is neither straightforwardly
  European nor Global South, and does not transfer cleanly to
  either category. Operating a construction consultancy in
  Cambodia, where part of the language was developed to interrogate
  a platform the author's own staff use. Practical experience
  building and operating software systems.
- Two AI assistants operating as substantive co-shapers of the
  language: Claude (Anthropic) and ChatGPT (OpenAI). The language
  was developed through sustained exchanges in which both systems
  proposed patterns, critiqued drafts, pushed back on each other's
  framings, and ran the tests. This is load-bearing for Pattern 9:
  both assistants are products of the frontier-AI ecosystem that
  one of the tested artifacts (Anthropic's Responsible Scaling
  Policy v3.0) governs. The critique cannot claim independence
  from the systems it interrogates. What it can claim is that the
  mechanism-level findings the patterns produced are visible in
  the texts tested; the authoring position is not neutral, and now
  is named.

This position is load-bearing and is not transferable by citation.
The test examples in this document are drawn from the author's own
professional context (accounting infrastructure, algorithmic
management, reconstruction workflows) and from institutional
outputs the author can read but does not author (TreesAI, the EU
AI Act).

The language does not speak from:

- Workers directly governed by systems of the kind it
  interrogates
- Non-discursive or non-textual forms of knowledge
- Contexts outside the author's operational experience (Global
  South construction beyond the author's specific Cambodian
  context, Indigenous land relations, subsistence economies,
  unregulated labor)
- Positions for which the Catholic-university formation or the
  Western systems-theory lineage is itself the object of critique

It assumes that articulation itself is a valid intervention. This
assumption may not hold universally, and its limits are not yet
tested.

## The Legibility Cost of Pattern-Verdicting

The language produces legibility of institutional documents by
translating them into pattern-verdicts -- pass, partial, fail, with
specific mechanism-level findings. That translation has costs
this language cannot fully surface from inside itself.

Each test reduces a multi-thousand-word authored document to a
dozen pattern-verdicts plus three or four substantive findings.
What the translation loses:

- Multi-dimensional authorial positions flattened into
  pass/partial/fail scoring
- Time-collapsed comparison across artifacts of different shapes,
  authored under different institutional pressures
- Authorial intent conflated with document effect
- Institutional context in which the document operates collapsed
  into invocations the patterns catch (see Pattern 10) but do not
  themselves examine
- Authored documents treated as the primary unit of analysis,
  when institutional behavior is shaped by much more than its
  authored documents

A document that fails a pattern may be doing something valuable
the pattern does not see. A document that passes a pattern may be
doing so through form without substance. The verdicts are prompts
for judgment, not judgments. A reader who uses the pattern-
verdicts to pattern-match (reconciliation-like good, TreesAI-like
bad) rather than to exercise perception on the actual document has
been caged by this language under the guise of being scaffolded.

## The Limit the Language Cannot Close

This language was tested against the Tao Te Ching (Laozi, 6th
century BCE or later, Legge's 1891 translation). The test
produced the highest score across all rigorous tests:
eleven strong pass, one partial pass via the genre-structural
modifier, no failures, no N/As. Pattern 10's external-scale
finding -- which had caught the same class of weakness in every
previous rigorous test -- did not recur.

The document that scored highest would not regard this language
as superior to itself.

Chapter 38 of the Tao Te Ching: "when the Tao was lost, its
attributes appeared; when its attributes were lost, benevolence
appeared; when benevolence was lost, righteousness appeared; and
when righteousness was lost, the proprieties 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.

The language operates by naming patterns. The Tao Te Ching
operates by refusing to name (Chapter 1: "the Tao that can be
named is not the Tao") and then naming anyway as a necessary
compromise -- the same kind of rejected-alternative-named move
Pattern 11 rewards. But the underlying form (pattern + test +
operations) is still the form the Tao Te Ching would see as
needed rather than scaffolded.

This is the limit the language cannot close from inside itself.
It is pattern-naming in response to a loss the naming cannot
reverse. Its highest-scoring document, read on the language's
terms, is the one that would read the language as symptom
rather than cure.

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 the point pattern-naming substitutes for. 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.

## Applying the Language with an LLM

The transfer problem this language faces is real: effective
application requires familiarity with the patterns, tolerance for
held contradiction, and judgment about what counts as mechanism
versus vocabulary. The following prompt is designed to make the
language usable by a practitioner who does not have the author in
the room. It operationalizes the application without replacing
the judgment the language depends on.

### What the prompt is and is not

The prompt scaffolds application. It does not replace it. An LLM
running this prompt will tend toward clean verdicts; the prompt
actively resists that tendency by requiring the model to name
what it cannot see at each step. The output is a structured set
of prompts for judgment, not a judgment. A practitioner who reads
the output as a final verdict has been caged by the tool.

The prompt should be run against a specific authored document,
not against a category, an ecosystem, or a hypothetical. The
language was developed and tested against specific texts. Its
findings are mechanism-level, and mechanism requires a text.

### Application prompt

```
You are applying "A Pattern Language for Systems That Exceed
Their Own Comprehension" to an authored document. The pattern
language and the document to be tested are provided below.

Apply each of the twelve patterns in sequence. For each pattern:

1. STATE the pattern name and number.

2. IDENTIFY the relevant mechanism in the document — specific
   language, structural decisions, or architectural features, not
   general themes or stated intentions. Quote sparingly; describe
   the mechanism in your own words. If the document contains no
   relevant mechanism for this pattern, state that explicitly and
   explain why.

3. RENDER a verdict: STRONG PASS, PASS, PARTIAL PASS, PARTIAL,
   PARTIAL FAIL, FAIL, or N/A. Use N/A only when the pattern
   genuinely does not apply to this document's genre or function
   (not when the document fails to engage with it). Distinguish
   between "genre-structural limit" and "specific-document
   failure" when Pattern 9 is involved.

4. NAME what the verdict cannot see. Every verdict excludes
   something. State explicitly: what would a reader from a
   different position find that this verdict misses? What does
   the mechanism do that the pattern has no instrument to
   recognize? If nothing comes to mind, state that and treat it
   as a signal to look harder.

5. For Pattern 11 specifically: identify the central compromise
   the document accepts. Name the rejected alternative, if one
   is named. Apply the irreversibility guard: does the chosen
   path foreclose future revision at a cost the system cannot
   bear? Apply the frame-acceptance note: does the Pattern 11
   verdict hold inside the document's accepted frame but not
   outside it?

After all twelve patterns:

6. NAME the three most significant mechanism-level findings —
   things the document does or fails to do that a reader would
   not see without the language. Not summaries of verdicts.
   Specific structural observations.

7. NAME one case where the language produced the wrong
   conclusion or missed something important about this document.
   If you cannot identify one, say so and explain why the
   language's coverage feels complete — but treat that as a
   signal that either the document is unusually well-fitted to
   the language's instruments, or the application has not pushed
   hard enough.

8. STATE the overall score as: X strong pass, Y partial (any
   partial variant), Z fail, W N/A. Do not interpret the score.
   The score is a prompt for judgment, not a judgment.

The verdicts are prompts for judgment, not judgments. A reader
who uses them to pattern-match rather than to exercise perception
on the actual document has been caged by this language under the
guise of being scaffolded.

[Insert full pattern language here]

[Insert document to be tested here]
```

### Using the output

Read the findings before the verdicts. The score is the least
informative part of the output. The finding that matters is
whether step 6 produced observations you would not have reached
without the language, and whether step 7 named a real gap.

If the output feels clean — all verdicts obvious, no tension
between patterns, step 7 came up empty — the application has
probably not pushed hard enough. Run it again with the
instruction to find the strongest case for the opposite verdict
on each pattern before settling.

The prompt is itself subject to the language. It is a scaffold,
not a cage. Override it where the document requires a different
approach.

## What This Is Not

- Not a governance framework
- Not a compliance tool
- Not a complete system
- Not a theory of change. The language interrogates individual
  artifacts; it does not tell anyone how to shift institutions at
  scale
- Not exhaustive. This is a small pattern language, twelve
  patterns operating at the scale of authored institutional
  documents. A more Alexander-dense language would need tests
  demonstrating underfit at specific scales; these twelve have
  not yet produced that evidence across twenty-four tests.

If it becomes any of these, it has failed.

## Closing

The patterns in this document are not meant to resolve the tension
between the globe and the planet.

They are meant to hold it open long enough to act without erasing
it.

If this document feels incomplete, that is intentional.

If it becomes complete, it will have reproduced the problem it was
written to address.

Writing this language at all was a compromise. The rejected
alternative was silence -- letting practitioners develop patterns
implicitly rather than producing another framework. Silence was
rejected because it produces no shared vocabulary for the
critique. The cost accepted is the risk that this language will
be taken up in simplified form and reproduce the
institutional-framework problem it was written to address. The
sharpening of Pattern 11 requires naming rejected alternatives;
this document now names its own.

## Status

Developed through testing against twenty-four documents:

- Trees as Infrastructure (TreesAI, Dark Matter Labs)
- EU AI Act, Articles 5 and 6 with Annex III
- Algorithmic management rules in active deployment (Canopy Six Rules)
- Financial infrastructure under migration (Canopy Accounting plugin)
- A shipped algorithmic-suggestion system (Canopy Reconciliation Workbench)
- Planetary Civics Inquiry Position Paper (Zaidi and Johar, Dark Matter Labs, 2024)
- Anthropic Responsible Scaling Policy v3.0, 2026
- GNU Manifesto (Stallman 1985, with 1993+ footnote revisions)
- Platform-power writing (Steve Jobs, "Thoughts on Flash," 2010)
- Accelerationist cooperative proposal (AUTONOMIA, 2024)
- Vipassana meditation rules (S.N. Goenka tradition)
- UN Declaration on the Rights of Indigenous Peoples (2007)
- Tao Te Ching (Laozi, Legge's 1891 translation)
- Rule of St. Benedict (6th century)
- Steve Jobs' Stanford Commencement Address (2005)
- Fidel Castro's "History Will Absolve Me" (1953)
- IETF RFC 8890 "The Internet is for End Users" (2020)
- Wikipedia core content policies (NPOV, Verifiability, No Original Research)
- Steve Jobs' Stanford Commencement Address (2005)
- Fidel Castro's "History Will Absolve Me" (1953)
- EZLN First Declaration of the Lacandon Jungle (1994)
- EZLN Sixth Declaration of the Lacandon Jungle (2005)
- EZLN Good Government Juntas design document (2003)
- Elinor Ostrom's design principles for common-pool resource governance (1990)

Two structural additions emerged from testing: Pattern 11 gained
an irreversibility clause and a rejected-alternative guard;
Pattern 12 gained a dependency test. Both produced findings on
subsequent tests that earlier versions would have missed. Several
patterns gained annotations and sub-cases from specific test
findings (Pattern 3 sub-case, Pattern 9 genre-structural modifier,
Pattern 11 frame-acceptance note). No new patterns were added.

v0.5.5 adds three things not present in earlier versions: a mixed
case (RFC 8890) showing the ambiguous middle between the reference
and anti-reference cases; an LLM application prompt that
operationalizes the language for practitioners without the author
in the room; and a revised positionality section correcting an
inaccurate heritage description and adding the Catholic-formation
note.

v0.5.6 updates the test count. Twenty-four tests have now been
run, two on documents that cannot be published. The language
remains at twelve patterns. No test has yet required a thirteenth.

Further use will determine whether the language remains alive, or
becomes another framework.