Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/theislampill/orthemology/llms.txt

Use this file to discover all available pages before exploring further.

Governing a classification system requires more than a list of state-types and a procedure. The system must also declare which higher-order distinctions govern its evidence procedure, its versioning discipline, its closure standard, and its authorization rules. Orthemology formalizes this with two related objects: the metaortheme μ (the repeatable governing configuration and its paired meta-policy) and the metaorthemma μ̄ (the concrete, case-bound configuration token inside a specific episode). These two objects stand in a type–token relation of their own — the metaortheme is the type, the metaorthemma is the token.
The term “metaorthemma” is benchmark-gated candidate vocabulary. The word is a candidate pending the terminology benchmark described in the manuscript’s Section 13.3. The object it refers to — the case-bound configuration token — is adopted and implemented; the word’s utility over ordinary language is undecided. Every coined term in Orthemology is governed by the same rule.

The Metaortheme Type μ

A metaortheme is a metaorthemic configuration μ paired with a separable meta-policy π_μ:
μ = ⟨g, S_μ, select_μ, prov(μ), ver(μ)⟩  +  π_μ
The components are:
mu_id
string
required
The stable identifier for this metaortheme type.
version
string
required
ver(μ) — recorded per episode so audits can scope which placements ran under which edition of the rule.
governed_component
enum
required
g — which part of the handling machinery this distinction governs. Must be one of O, I, E, D, R, V, or W. See the governed-component table below.
states
object
required
S_μ — the set of declared competing higher-order states the governing context may occupy. Must be declared in advance, not read off a single incident. Requires at least two members.
selector
object
required
select_μ — the state-selecting evidence procedure that determines which state in S_μ actually obtains for a given episode. A typed object with kind and description required; a bare string is not a valid selector.
meta_policy
string
π_μ — the conduct rule conditional on the obtaining state, including what the rule prescribes on undetermined selection. Either meta_policy or policy_ref is required; a metaortheme without a paired meta-policy is not a metaortheme.
provenance
object
required
prov(μ) = ⟨authority, warrant, scope⟩ — the rule’s own provenance. A rule of unknown provenance is a stale-evidence problem one level up.

Governed Components

The governed component g must be drawn from the following declared set. Objectives, loss functions, the task itself, and the governance meta-level are excluded.
CodeComponentGoverns
ORepertoireWhich state-types exist and may be posited or retired
IIndividuationWhat counts as the same case; identity keys, versioning, lineage
EEvidenceWhat evidence counts, of which property class, at what grade
DDisclosureWhat may be left open, and how open questions must be declared
RRoutingWhere cases are sent; what happens when the correct route is unavailable
VValidation/closureWhat may be called done, at what standard, when a completion claim must be reopened
WWarrant-classificationWhich warrant states exist (authorized vs established) and what each licenses
Objectives and loss functions are not governed components. In a widely discussed exchange, an instruction of the form “be maximally truth-seeking” was described as if it were a governing distinction. It fails every anti-vacuity condition: it names no governed component, declares no competing higher-order states in advance, and specifies no selecting evidence. It is an objective — and objectives are excluded from g by construction. This is the standing negative example in the formalization.

The Selector: A Typed Object

The selector select_μ is not a bare description — it is a typed object with declared fields:
{
  "selector": {
    "kind": "instrument-reading",
    "description": "Lineage check of ver(m) against the canonical latest version",
    "inputs": ["occurrence.version", "lineage_registry.latest"],
    "partiality": "undetermined when lineage registry is offline; escalate",
    "procedure_ref": "governance/version-check-v2.md"
  }
}
{
  "mu_id": "MU-VERSION-CURRENCY",
  "version": "2.1.0",
  "governed_component": "I",
  "states": {
    "members": ["current", "stale"],
    "partition": true
  },
  "exclusivity": "exclusive",
  "selector": {
    "kind": "query",
    "description": "Compare ver(m) to the canonical latest version in the lineage registry",
    "inputs": ["occurrence.version", "canonical_latest"],
    "partiality": "if registry is unreachable, return undetermined; do not assume current",
    "procedure_ref": "governance/version-currency-check.md"
  },
  "meta_policy": "On state=stale: quarantine occurrence and open a re-derivation residual before any placement is made. On state=undetermined: escalate to governance boundary; do not proceed.",
  "provenance": {
    "authority": "release-governance-board",
    "warrant": "board-resolution-2026-01-15",
    "scope": "all automated build and classification episodes"
  }
}

Anti-Vacuity Conditions

A candidate μ is admitted as a metaortheme only if all four conditions hold:
  1. g is named — the governed component is explicitly declared
  2. S_μ is declared in advance — the competing higher-order states are specified before any episode runs, not read off a single incident
  3. The distinction is consequential — switching the obtaining state changes validated placement, risk, or constraints in at least one episode class
  4. It is not a disguised first-order ortheme or policy parameter — a good policy that consults no in-advance-declared competing higher-order states is just a good policy

The Metaorthemma Token μ̄

The metaorthemma μ̄ is the concrete, case-bound configuration token of a metaortheme type inside a specific episode. It is not the episode; it is not the execution event. It references the episode’s evidence and trace; it never absorbs them.
MetaInst(μ̄, μ)       — μ̄ is a token of metaortheme type μ
Compatible(μ̄, A(e))  — μ̄ binds case-specific values within the declared analysis

Metaorthemma Schema Fields

token_id
string
required
The token’s own identity and lineage. Each metaorthemma is individually identified.
of_type
object
required
MetaInst(μ̄, μ) — exactly one metaortheme type reference (mu_id + mu_version). Single-typing is the current rule; a case governed by several types uses several separate tokens.
analysis
object
required
Compatible(μ̄, A(e)) — the analysis this token is compatible with (analysis_id + version). The token binds case-specific values within the declared analysis; it never overrides A without a new analysis version.
anchor
object
required
The target occurrence (κ, v)identity_key + version. The token is occurrence-anchored.
governed_component
enum
required
Must match the governed component of the parent metaortheme type.
binding
object
required
The case-specific binding map B — reference frame, tolerance value, fixture, success surface, instrument, calibration, etc. Must be non-empty: a metaorthemma exists only where material case-specific binding exists (the zero-burden rule).
binder
object
required
Who or what made the binding — actor, binding_warrant (w_bind), and binding_time. Kept distinct from the designated executor.
validity
object
required
The temporal window: effective_from, expiry, superseded_by. A binding with no declared start is undated governance; effective_from or a declared validity_unbounded_reason is required.
scope
object
required
The claims in 𝒬(e) that depend on this token. Must carry the dependent claims, an explicit reason why no claim depends on this token, or a declared external scope model.
{
  "token_id": "MUBAR-ep-0042-version-currency",
  "of_type": {
    "mu_id": "MU-VERSION-CURRENCY",
    "mu_version": "2.1.0"
  },
  "analysis": {
    "analysis_id": "build-compliance-analysis",
    "version": "3.0.0"
  },
  "anchor": {
    "identity_key": "repo/src/main.py",
    "version": "commit-abc123"
  },
  "governed_component": "I",
  "binding": {
    "reference_version": "commit-abc123",
    "canonical_latest": "commit-abc123",
    "tolerance": "exact-match"
  },
  "scope": {
    "claims": ["claim-provenance-currency", "claim-artifact-identity"]
  },
  "binder": {
    "actor": "release-pipeline-v4",
    "binding_warrant": "governance-auto-bind-rule-2026",
    "binding_time": "2026-07-20T14:32:00Z"
  },
  "designated_executor": "ci-scanner-agent-12",
  "validity": {
    "effective_from": "2026-07-20T14:32:00Z",
    "expiry": null,
    "superseded_by": null
  }
}

The Four Layers

The split-model formalization keeps four layers explicitly separate:
1

Type μ — the repeatable governing configuration

The metaortheme type: which component it governs, which states it declares, which evidence selects among them. Adequacy verdict: V3a (GOV_CONFIG_ADEQUATE).
2

Policy π_μ — the repeatable conduct rule

The meta-policy paired with the type: what the system must do conditional on the obtaining state. Adequacy verdict: V3b (GOV_POLICY_ADEQUATE).
3

Token μ̄ — the case-bound configuration binding

The metaorthemma: the concrete token that applies the type’s binding to a specific occurrence under a specific analysis, with declared scope, calibration, and binder warrant. Adequacy verdict: V3c (GOV_TOKEN_ADEQUATE).
4

Execution — the application event

The episode trace recording how the executor applied the token. The application event ApplyEvent(μ̄, e) = ⟨μ̄, Trace_e|_μ̄⟩ is derived from the episode; it is never the metaorthemma itself. Fidelity verdict: V3d (EXECUTION_FAITHFUL).
The adequacy questions at each layer are independent. A correct standard (V3a pass) can carry an ill-formed policy (V3b fail). A correct standard and sound policy can have a defective case-binding (V3c fail) — wrong reference plane, wrong-role tolerance, expired calibration, wrong fixture or success surface. And a correctly bound token can be faithfully executed by an executor who still makes an error at the procedure level. Each layer routes to a different remedy.

V3c: GOV_TOKEN_ADEQUATE

The verdict V3c (GOV_TOKEN_ADEQUATE) judges every concrete metaorthemma applied in an episode:
V3c(e) ⟺ ∧_{μ̄ ∈ MetaTok(e)} TokenAdequate(μ̄, e)
where TokenAdequate(μ̄, e) holds iff:
ConjunctWhat it checks
MetaInst(μ̄, μ)The token is an instance of a declared metaortheme type
Compatible(μ̄, A(e))The token is compatible with the episode’s declared analysis
Anchored(μ̄, κ(e), v(e))The token is anchored to the correct occurrence identity and version
ScopeCorrect(μ̄, 𝒬(e))The token is scoped to exactly the claims that depend on it
Current(μ̄, t(e))The token’s validity window covers the episode time
Provenanced(μ̄)The token has declared and traceable provenance
AuthorizedBinding(μ̄)The binding was made under authority
Per-token statuses are preserved alongside the episode-level conjunction. An auditor must be able to see which governing application failed — for example, μ̄_2: stale calibration; μ̄_3: wrong claim scope. Zero-burden rule: when MetaTok(e) = ∅ (no material case-specific binding exists), V3c ∉ ReqPath(e) and the status is recorded not-applicable with a reason. The absence of metaorthemmata is an explicit recorded fact, not silence.

Plurality, Conflict, and Precedence

A typical episode runs under several metaorthemes at once:
μ⃗(e) = ({μ₁, …, μ_k}, ≼)
Configurations governing disjoint components compose freely — their constraints conjoin. Two metaorthemes governing the same component can conflict: when their meta-policies prescribe incompatible constraints under the states that actually obtain, resolution is only via the declared strict partial order . An unresolved conflict is an ANDON condition — stop or escalate. Silent override is itself a metaorthemic error.

Build docs developers (and LLMs) love