## 10 Native incorporation witness ### 10.1 An explicit logical realisation Consider two retained binary registers $(m,z)$, an admitted binary input $u$, and a synchronous native update $$m'=z,\qquad z'=m\oplus u.$$ The native record is $z'$. Each tick has unit logical cost. The nominated state carrier contains all four register states; each fixed-input update is bijective. The storage factorization, timing and intervention meaning are declared premises of this logical model. Under repeated $u=0$, the first two native records are $(m,z)$. Hence all four actual states have distinct native predictive classes, and the strong quotient is the identity. The internal dependency graph has $m\to z$ and $z\to m$. A two-tick $u=0$ schedule returns the initial value of $z$ through $m$ to $z$. Holding the other initial register fixed and varying $z$ changes the endpoint record deterministically. Thus the graph route is accompanied by an executable distinguishable return in the logical contract. This is a positive model-level witness, not a certificate for an unspecified physical circuit. An implementation must still justify its native clock, records, preparations, boundary isolation, owned resources and operative provenance. An external display copying $z'$ is not automatically the native record required by the model. The chosen record must have the declared endogenous physical role. ### 10.2 Installing and testing a relational result Let $\theta=A\oplus B$ be formed from two earlier inputs and install $(m,z)=(0,\theta)$. Isolate or remove the original input carriers from the future task boundary. After a first tick with $u=0$, the state is $(\theta,0)$ and the record is zero. On a second tick with query $u=q$, the record is $\theta\oplus q$. The stored relation has become a condition of later action. Before the test, overwriting $z$ with zero maps both possible initial installations to $(0,0)$. If $\theta$ is fair and no correlated variable remains available within the task boundary, no later procedure can recover $\theta$ with success greater than $1/2$. Restoring the correct $z=\theta$ rescues exact performance. This follows because the ablated future initial state and inputs are independent of $\theta$, whereas the rescued second record is the required parity. The lawful overwrite and rescue are manipulations of a realised register, with their preparation and control costs charged separately. The example demonstrates persistence, mediation, ablation and rescue. It does not show that dialogue was necessary. A static message containing the parity can install the same state if the receiver has the required permissions and resources. Nor does the model prove that an arbitrary artifact carrying that bit is conscious. Its recurrent realisation and native record contract are essential to its conditional use under A1. ### 10.3 Conditional phenomenal point separation **Proposition N1.** Suppose a certified qualifying realisation has one fixed complete endogenous predictive object and two lawful preparations $\mu_0,\mu_1$ concentrated on actual states $x_0,x_1$. If a common admitted native finite experiment has different laws from those states, then their native predictive classes differ. Under A2, their actual phenomenal points differ within the fixed calibrated structural copy. **Proof.** Equality of the native predictive classes requires equality of every admitted finite native law. One differing law contradicts that equality. A2 maps distinct classes to distinct points through its structure-preserving copy. $\square$ This modest statement is the exact bridge available from the constitution. In the register model, the two parity installations are distinguished on the second idle record, so they satisfy its model-level premise. A2 does not assign a named human quale to either state, and the argument does not establish primitive presence independently of A1 and A0. If model predictions for a chosen native experiment have contrast $\widehat\eta$, and the two actual-law discrepancies are bounded by $\beta_0,\beta_1$, the actual contrast is at least $\widehat\eta-\beta_0-\beta_1$ by the triangle inequality. A strictly positive lower bound establishes class separation for those nominated physical preparations. For mixture preparations, a positive diagnostic difference separates their predictive laws; it does not identify a single actual phenomenal point without further state information. The fixed carrier already includes the retained parity. Installation therefore changes the actual point, not the unpointed four-state predictive object. A genuinely changed native operation, boundary or domain would require a new justified realisation before claiming a changed constitutive object. ### 10.4 Countermodels that delimit the inference First, add a register that no native operation or native record consults, but that an external laboratory device can read. Altering that register can change an external score while leaving every native continuation law unchanged. Such a score is not a native diagnostic for N1. The register’s physical membership in a qualifying core is a separate realisation issue. Second, an arbitrarily long forward cascade can propagate a difference without returning it to its origin. A graph obtained by identifying repeated temporal roles can display a cycle even though the native physical carrier has no return. Time unfolding of a recurrent machine is itself a directed acyclic graph, so acyclicity of the unfolding is equally unable to settle native recurrence. The relevant question concerns the realised carrier and executable schedule. Third, causal influence does not generally compose transitively as endpoint influence. If $a$ affects $b$ and both affect $c$, cancellation can erase the total contrast. In a binary example let $b=a$ and $c=a\oplus b$. The direct structural dependencies are present, but changing $a$ while allowing $b$ to update leaves $c=0$. A route through dependency edges is not a proof of a distinguishable endpoint return. The native certificate must test the composed schedule. These countermodels explain why the developmental narrative cannot supply its own A1 proof. They also preserve a constructive path: identify a retained relation, establish its native continuation role, verify a genuine recurrent realisation, and only then apply the stated constitution.