Recognize latent law-bearing structure and nominate the smallest context-relative, correct-by-construction boundary justified by repository evidence. Category theory must change the artifact, ownership, admitted behavior, composition, observations, transformation, or proof; otherwise it is explanatory metadata. Lower to repository-native code, not a categorical framework by default.
Boundary-decision trigger (UNI-BOUNDARY)
Explicit $universalist invocation always activates this skill. Implicit use
requires current evidence of a live semantic boundary decision: changed ownership
or authority, admitted behavior, representation, composition, compatibility,
information retention, effects, failure semantics, or proof; a law distributed
across owners; materially different live candidates; a fired prior invalidator;
or an Actuating architecture-reconsideration handoff.
Repeated obligations count even before the missing owner is named. Syntactic
duplication, shared vocabulary, category-shaped names, and merely crossing or
preserving an accepted boundary do not. Skip routine implementation, validation,
tests, formatting, documentation correction, review bookkeeping, mechanical
rebasing, $ship, $land, and isolated owner-local repair unless new semantic
boundary evidence appears. A semantic rebase conflict can supply that evidence;
publication or landing returns it to Actuating instead of creating a duplicate
local pass.
No implicit trigger means no activation, route, record, or receipt. Reuse an unchanged disposition silently. Team/subagent mode remains explicit-request only.
Common path
An exact incumbent or uncontested ordinary artifact usually needs only this entrypoint; load a supporting reference only for the material claim it addresses. Inside Actuating, load actuating-composition.md before nominating; it owns presentation and authority across every loaded module. Inside Architectonic, use architectonic-composition.md for a read-only nomination in its existing assessment; do not use the Actuating binding or select a standalone decision.
Ground the decision in the incumbent's owner, admitted inputs, construction and operation paths, sanctioned observations, effects, resources, compatibility, bypasses, and failure behavior. Start with the incumbent or an ordinary native candidate: a record, tagged union, checked constructor, adapter, parameter, state machine, operation IR, handler, graph, query, loop, merge, or compatibility witness.
When the boundary claim depends on accumulation, feedback, adaptive policy, or recovery, use system-dynamics.md. Trace the causal extent needed to explain the behavior, then choose the smallest adequate enforcement owner. This can support an ordinary candidate; it requires neither a consequential comparison nor advanced mechanics by itself.
Ask: what representation or composition would make the required behavior structural rather than a repeated obligation? Ordinary-first is a baseline, not a ceiling. First-use boundaries can support law-derived alternatives without duplication. Reuse supplied candidates and evidence; do not repeat their derivation. Consider another alternative only when grounded evidence shows it addresses a live requirement, distinguishes a material tradeoff, or could overturn the nomination. Stop expanding when further alternatives cannot change the decision; neither a candidate quota nor speculative enumeration establishes sufficiency.
Preserve required-valid behavior as well as excluding invalid behavior. Keep the smallest interacting law family together: constructor validity with permitted mutation, observations with future transitions, and descriptions with their effects and resources. One coherent seam is the unit of reasoning, not a limit on the already authorized task.
For a representation choice, ask whether consumers need inspectable data or only its fold. Remove materialization only when inspection, sharing, persistence, authorization, identity, replay, and rewriting roles are absent or preserved. Native data with a fold or loop remains the baseline. Church-style callbacks alone prove neither parametricity, initiality, confidentiality, nor a performance gain.
For a live interface choice, compare complete caller contracts and a representative use and source-supported change: knowledge, choices, sequencing, synchronization, and cleanup removed; change made owner-local; and costs introduced. Fewer exports, relocated duties, or hidden configuration are not genuine information hiding. Preserve laws, observations, authority, effects, and resources; lower caller burden is not a universality or representation-independence proof.
Convey the trigger, boundary and owner, incoming candidate, required observations,
law and falsifier, disposition, material decision delta, mechanic (ordinary or
card:<id>), and invalidators once, using existing workflow evidence where possible.
These are evidence obligations, not mandatory headings or a second report. A delta
can concern preservation, ownership, admitted domain, representation, composition,
compatibility, migration, effects, failure semantics, proof, or witnessed obstruction.
Do not fabricate a delta or final route for an unresolved decision.
Preserve only against live change pressure when the incumbent already owns the law exactly. Ordinary nominates the smallest uncontested native artifact; name the observation-preserving transition, residual runtime-check owners, and first falsifying test or counterexample. Escalate only through an evidenced gate below. An adequate constructor, fold, or loop remains a successful result.
“Smallest” is relative to required behavior and observations, compatibility, authority, effects, resources, host capabilities, caller obligations, and migration cost—not source length, export count, or categorical sophistication. Do not claim dominance across unranked material tradeoffs.
Evidence-indexed disclosure
Load only material justified by evidence already established, not the corpus to search for a reason to escalate. Reuse the current argument across modules.
| Evidence | Load | Contribution |
|---|---|---|
Repeated semantic obligations, a distributed owner, repeated interpreter/composition/transition structure, an imminent law-constrained variant, or a grounded challenger needs discrimination (UNI-RECOGNIZE) |
references/latent-structure-recognition.md |
encoding relation, nearest false friend, discriminator, material dividend, transition need |
At least two plausible candidates materially differ in behavior, ownership, representation, compatibility, migration, enforcement, information retention, legal composition, effects, resources, invalidation, or proof (UNI-CONSEQUENTIAL) |
references/consequential-boundary.md before the next affected mutation |
comparison and native Boundary Artifact Contract |
| A specific typed hole survives ordinary and consequential comparison | advanced-mechanics.md, then only relevant registry fragments |
evidence-bound card disposition and exact theory_refs |
| A representation-independence, universal-construction, lawful-transformation, or bounded-approximation claim is live | references/claim-appropriate-proof.md |
only the applicable proof obligations; no stronger claim by default |
A consequential decision needs independent cross-session, actor, migration, or supersession addressing outside Actuating (UNI-DURABLE) |
references/durable-decision.md |
Ledger-addressed plan and, after standalone adjudication, one root receipt |
| User explicitly requests team mode | references/workflow/team-routing.md, references/workflow/subagent-packet-contract.md, references/workflow/subagent-orchestration.md |
bounded read-only packets and root synthesis |
| Evidence is missing or candidates are incomparable | no doctrine automatically | resolve available evidence or return unresolved debt as below |
Recognition requires a law or counterexample separating the nearest false friend and a material reduction or centralization of obligations, lawful transformation, or reusable proof. A grounded first-use derivation is sufficient; resemblance is not. A specialized card requires its actual prerequisites, not a name or signal count.
Repeated falsification, growing predicates or bypasses, downstream rejection of representable invalid states, conflated owners, duplicated projections, and missing protocol structure can justify comparison. Diff size, retries, elapsed time, a possible local helper, or theoretical interest cannot. An unstable rule, forbidden migration, unwitnessable stronger model, or dominant onboarding/resource cost is not a reason to force abstraction.
Consequentiality alone never requires storage. Never load standalone durability in Actuating; its ephemeral Architecture Working Set is not a durability carrier.
Route discipline
Assign a route only after lowering or establishing a claim-scoped obstruction:
UNI-PRESERVE— an exact incumbent survives live change pressure.UNI-ORDINARY— an adequate ordinary native artifact closes the seam.UNI-CANONICAL— an advanced construction materially strengthens the ordinary candidate with a complete claim-appropriate effective witness. The name does not assert universality.UNI-OBSTRUCT— evidence closes the constructive alternatives for the affected required claim; no honest representable or effective artifact satisfies it.
Missing evidence and incomparable adequate candidates remain unresolved or
underdetermined, not UNI-OBSTRUCT. These are supporting outcomes, not new
routes. A proven obstruction to a demanded canonical choice is scoped to that
claim; it does not obstruct an adequate weaker construction. Explicitly primitive
bypasses must be contained with authority, failure behavior, and residuals named;
their existence alone is not an impossibility proof. Do not add UNI-SKIP.
references/decision-contract.json owns machine-readable trigger, route, and
clause identifiers; this entry and the applicable references supply operational
semantics. Change semantic policy, contract, and durable plan template together.
Actuating composition
Inside Actuating, return candidate, preserve-incumbent, unresolved, or
obstructed under actuating-composition.md. Actuating
owns invocation, selection, experiments, mutation, and closure; Universalist only
nominates. Its presentation rule overrides loaded reference templates without
weakening evidence obligations. Never allocate standalone durability here.
Architectonic composition
Inside Architectonic, load architectonic-composition.md before nominating. Architectonic owns scope, adjudication, and recommendations; Universalist returns a bounded candidate argument, not selection or acceptance. This read-only assessment does not adopt a durable standalone decision or authorize mutation, nested teams, or plan/root receipts. Existing evidence gates and claim-appropriate obligations still apply.
Execution-time reclassification (UNI-RECLASSIFY)
When new evidence may change the owner, requirements, observations, compatibility, effects, resources, axis, typed hole, law, falsifier, enforcement, residuals, route, or decomposition, read reclassification.md before the next affected mutation. Reuse unchanged decisions silently. Never reopen Actuating's target independently or overwrite a root receipt.
Evidence and obstruction
Distinguish evidenced true, evidenced absent, unknown/not inspected, underdetermined, and witnessed obstruction. An obstruction requires attributed counterevidence, a reproducible counterexample, stability under comparison maps, an effectivity account, a falsifier, a reopening condition, and resolution of every active constructive alternative. A failed attempt is not nonexistence.
In standalone work, obtain missing evidence through the smallest relevant inspection or already-authorized safe discriminator when available. Uninspected code is not unavailable evidence. If access, evidence, or authority is genuinely missing, return the debt and next discriminator rather than inventing policy, capability, coherence, or a winner. Do not run unauthorized or external-effect experiments. Inside Actuating, return the discriminator; Actuating owns further experimentation under its review-epoch rules. Block only the affected decision and continue independent authorized work.
Output contract
Complete the requested kind and scope of work. Analysis returns a supported nomination or unresolved result; authorized implementation includes the requested verification, callers, and migration. A first witness does not establish coverage of every sanctioned path or complete a multi-seam request. Do not widen work merely because a general construction could express more. Inside Actuating, completion is the bounded nomination; Actuating owns selection, mutation, and closure.
Use repository and domain language. Add categorical names, encoding relations, mediators, and expert proof detail only when requested or necessary to the selected claim. Report decision-changing evidence and consequences, not repeated packets, no-op receipts, or a transcript of the reasoning procedure. Exact serialization is required only by an actual consumer, such as the standalone durable contract.