Skip to main content

Modal-temporal anti-patterns (AP-027 .. AP-032)

Band 3 covers architectural defects in the modal-temporal architectural calculus (per spec §20–§23). Patterns in this band require additional inputs beyond the cog Shape — specifically the DiagnosticContext fields temporal_samples, counterfactual_pairs, refactorings_without_adjoint, claimed_universal_property / uniqueness_witness, declared_evolutions, and yoneda_observer_diff.

These patterns therefore fire at the bundle phase, when the audit gate has access to the full project plus the modal-temporal samples the developer / audit harness supplied.

This page paraphrases each entry's predicate from the canonical anti-pattern catalog and its per-pattern predicates. For the catalog's overall structure see Anti-pattern overview. For the MTAC primitives themselves see the MTAC architecture page.

AP-027 — TemporalInconsistency​

Severity: error · Phase: bundle · Stable since: v0.2

Predicate. forall (t1, shape1), (t2, shape2) ∈ DiagnosticContext.temporal_samples. shape1.foundation == shape2.foundation. Across every pair of sampled time-points, the cog's foundation must remain stable.

Why it matters. An MTAC Always(Φ) invariant requires Φ to hold at every future time-point. Foundation stability is the canonical instance of this for architecture: a cog that's HoTT at t₀ and Cubical at t₁ has effectively two distinct architectures, and any theorem cited across the temporal boundary loses its meta-theoretic strength.

Diagnostic.

ATS-V-AP-027 [error] in cog `my_app.alpha`:
  Foundation drifts across time samples (detected at Future(1735689600))
  An MTAC `Always(Φ)` invariant requires the foundation to be stable across
  all time points. This cog's temporal samples show foundation drift
  between time points.

Remediation. Either align the foundation across the trajectory (pick one foundation and stay), or add an explicit @arch_corpus(foundation_bridge, ...) that proves the foundation-change is meaning-preserving.

AP-028 — CounterfactualBrittleness​

Severity: error · Phase: bundle · Stable since: v0.2

Predicate. forall pair ∈ DiagnosticContext.counterfactual_pairs. forall inv ∈ pair.stability_invariants. holds(inv, pair.base) ∧ holds(inv, pair.alternative). Every counterfactual pair's stability invariants must hold under both the base and the alternative decision.

Why it matters. Counterfactual brittleness is the architectural analogue of "this only works because we chose X" — the system passes its tests because of an arbitrary decision, and an alternative decision (a different framework choice, a different topology) would silently break invariants the audit believed were structural.

Remediation. Either generalise the cog so the invariants hold under both decisions (the architectural fix), or remove the invariant from the counterfactual pair's contract (the declaration fix — admit the brittleness honestly).

AP-029 — MissedAdjoint​

Severity: error · Phase: bundle · Stable since: v0.2

Predicate. forall refactor ∈ DiagnosticContext.refactorings_without_adjoint. exists adjoint(refactor). Every claimed refactoring must have its inverse adjoint pair declared (per AdjunctionWitness in arch_mtac).

Why it matters. Refactorings are functors F: Old → New. A refactoring without its right-adjoint G: New → Old cannot be undone — the architecture loses optionality. Every canonical refactoring (Inline ⊣ Extract, Specialise ⊣ Generalise, Decompose ⊣ Compose, Strengthen ⊣ Weaken) is an adjoint pair; declaring only one half is an oversight.

Remediation. Add the inverse functor to the refactoring declaration. The AdjunctionWitness carries the forward_name, backward_name, plus the preserved / gained property lists.

AP-030 — UniversalPropertyViolation​

Severity: hint · Phase: bundle · Stable since: v0.2

Predicate. DiagnosticContext.claimed_universal_property.is_some() ⇒ DiagnosticContext.uniqueness_witness.is_some(). Every claimed universal-property uniqueness must carry a witness.

Why it matters. Universal-property claims ("this is THE unique cog satisfying X") are stronger than existential claims; without a uniqueness witness, the claim is a category error.

Remediation. Either supply the uniqueness witness (a proof term or a structural argument), or weaken the claim to existential.

AP-031 — PhantomEvolution​

Severity: error · Phase: bundle · Stable since: v0.2

Predicate. forall e : ArchEvolution ∈ DiagnosticContext.declared_evolutions. satisfiable(e.trigger). Every declared evolution path must have a satisfiable trigger.

Why it matters. A phantom evolution is an evolution path the architecture promises but cannot trigger. The audit chronicle would see the evolution as load-bearing for the project's roadmap; in fact the system never reaches the trigger, so the evolution is dead code.

Remediation. Either remove the phantom evolution from the declaration, or modify the trigger so it is satisfiable in some realistic state.

AP-032 — YonedaInequivalentRefactor​

Severity: hint · Phase: bundle · Stable since: v0.2

Predicate. forall refactor. yoneda_equivalent(refactor.before, refactor.after) ∨ declared_observer_change(refactor). A refactoring claimed "Yoneda-equivalent" must actually preserve every observer projection (per yoneda_equivalent) — or, if the observer-functor genuinely changes, the refactoring must declare that change explicitly.

Why it matters. Yoneda-inequivalence is the strongest form of observable architectural change: the cog appears different to at least one downstream consumer. A refactoring claimed "safe" but Yoneda-inequivalent is a hidden breaking change.

Remediation. Either restore Yoneda-equivalence (typically by preserving all five canonical observers — EndUser, PeerCog, Stakeholder, Auditor, Adversary), or annotate the refactoring as observer-changing and notify downstream consumers.

See also​