Healthy primitives and failure families are causal claims. The navigation-surface disclaimer attaches to the twelve-count, not to those primitives, so the page does theory work while denying it is doing theory.

The page already treats layer numbers as addresses and the diagnostic move as a case-specific dependency, not numerical causation, so the working-hypothesis frame is load-bearing rather than decorative.

The live claim is the primitives, not the count. The six preconditions already describe a stack-shaped system, so the scan applies where the failure is already in this grammar; a corpus-reorganisation test is editorial.

Inter-rater disagreement on addresses would show the layers are labels. Inter-rater agreement is still insufficient if several layers each pass a "repair this layer unblocks the outcome" counterfactual.

"Smallest joint set of domains" has no algebra, and geometry (sequential, parallel-bonded, selector, allocation, substrate, topology) is named later. Layer agreement can hide geometry disagreement.

Reproduction as competence-stock depletion is a ledger object. Layers 5 and 11 are then the same stock at different time derivatives, so vocabulary specificity is not evidence of a causal cut.

Framework T as deep-time flourishing, audited at Layer 0 when it diverges from mandate, is a value override of the system's own purpose rather than a diagnostic precondition.

Telos-relativity is the method: the walk cannot start without a named boundary and T, and local success with flourishing-failure is the point of the qualifier, not a bug.

Two candidate T's with no composition rule makes T-choice the political dispute, treated as a field you fill in. Without an observation that forces flourishing over a succeeding mandate, Layer 0 audit is a named preference.

The town/court composite stipulates flourishing-failure while both systems succeed locally, and is labelled as carrying no evidentiary weight, so it cannot test T-choice.

If a two-T walk and a one-T walk emit the same repair — own the conversion between systems — T-choice has no downstream effect and Layer 0 audit is decorative.

Multi-system collision is the only worked example and has no layer address. The twelve-domain walk points away from the case the page actually works.

Naming the missing primitive is better than absorbing collision into Layer 3 or 8, and "walk both stacks" remains available as a diagnostic move without a thirteenth domain.

Walk-both-stacks is the primitive: import-obligation, carrier protection, and cross-system closure. Offering it as method after listing it as gap folds the primitive back into the twelve.

The collision category has no specimen that can show the twelve-domain walk failing where the cross-system questions succeed; the composite and the unworked pairs cannot carry that test.

The gate table relabels layers 4 and 6–10. It is a zoom on the intelligence-to-action corridor, not an independent coordinate system.

Layer locates substrate and gate locates stopped passage, so production versus admission at Layer 7 is a real cut even if gates do not apply to telos or reproduction.

Earliest-dependency-gate is a corridor rule. The two-step (gate, then distortions) still starts with the gate, so parallel-bonded cases get a sequential first move the legend warned against.

Selector-matrix is listed as a geometry where first-failed-gate misleads, but triage still walks the twelve in order. A no-vector-available case will first-hit Decision or Feedback and never reach the selector question.

Walk-the-twelve-in-order reintroduces the causal rank §I disclaimed. The geometry question must be able to recode the layer, not sit after it as a distortion scan.

The map is drawn in ownership shorthand ("unowned X produces Y") after the candidate says a named owner is neither necessary nor sufficient and failure is conversion that cannot produce T-preserving adaptation.

Ownership is a first-pass, the candidate is under test, and the twelve domains are substrates on which either ownership or another contract obligation can fail; no replacement is ratified.

The "unowned X produces Y" paragraph is the taxonomy, not hedging. Under the candidate, several of those recode as actuation, capacity, calibration, or continuity with an owner already on the chart, so the scan emits the wrong address.

Distributed protocols that convert without a named owner are the candidate's counterexample and have no layer address, so the first-pass scan files them as Layer 8 or 3 and generates an install-an-owner repair the candidate already rejected.

Exit-voice-loyalty already cuts response and standing; sequential-barrier models already cut the gate table; principal-agent already cuts unforced decision. The remainder that would change novelty is a misfile with a different repair, and compilation-versus-computation of a public frame is the candidate remainder.

Non-Compilation versus Calculemus is the page's own reason layers matter, and the defence is pragmatic organisation, not uniqueness against every prior partition.

The misfile test is: a case with standing whose missing object is the assembled frame, so the repair is build-the-frame rather than another complaint channel. If every such miss files as Layer 3, or the split only distinguishes two essays, there is no remainder.

The symbolic-substitution test applied to the page itself — action-set change, who must act, record or trigger, consequence if nothing happens — yields only named domains and essay pointers, so the walk is symbolic under its own rule.

The page is an analyst's working object (triage, cross-reference, gap-finding, repair routing), not a proposed institution. The substitution test is written for an institutional channel; a map's consequence if unused is different filing by the next analyst, which is a typed exemption rather than a response-duty.

Gap-finding ("fewer essays than weight predicts") generates a writing agenda from the partition. Under the substitution test, representing a domain is the failure mode; the missing generator-chain-formation essay is listed because Layer 11 has a hole, not because a failure demanded it.

A shared removal-test does not make state, claim, and capacity carriers one type. Collapsing three repairs under "carrier" is the misdiagnosis the page warns about, installed in the vocabulary.

The three-way split assigns state carriers to ledger/execution, claim carriers to Layer 3, and capacity carriers to Layer 11. The removal-test is the shared question; the repair is read off which sense was removed; the layer list is the routing. The word is the question, not the type.

Layer 3's title "Standing / carrier" and Layer 11's later "capacity carrier" do not survive a "which carrier is missing" walk: a dead apprenticeship emits Layer 3. Three repairs need three words in the addresses.

Visible-but-powerless and public-but-shameless share a signature — signal returned, no bound actor — so attractors are a repair-routing table, not an independent second triage move.

The five rows discriminate organised feeders, missing actuator, feeder-free time/jurisdiction, dominant-player shamelessness, and aggregate binding mass — different missing conversions, not one signature with five names.

If attractor assignment requires the layer diagnosis first, the second triage move is not independent. Two compilers given only signatures should agree on the row without already knowing the layer.