Module 01 / 04

The post-failure object is not the nominal object.

Let the nominal controlled plant be \(\mathcal P_0\). Damage or degradation produces a plant \(\mathcal P_d\) that may not be recoverable during the mission.

The post-failure object is not the nominal object.Open section

Persistent premise

The post-failure object is not the nominal object.

Let the nominal controlled plant be \(\mathcal P_0\). Damage or degradation produces a plant \(\mathcal P_d\) that may not be recoverable during the mission.

\[ \mathcal P_0 \xrightarrow{\;\mathcal D\;} \mathcal P_d, \qquad \mathcal P_d\not\equiv\mathcal P_0. \] This premise survived the model revision. What changed is how the state, the decision and the admissibility of reconfiguration are represented.

A resource can fail physically — or become epistemically wrong.Open section

Extended state

A resource can fail physically — or become epistemically wrong.

The resource state now includes physical resources and epistemic resources: mathematical models, learned models, estimators and algorithms whose validity can degrade even when the hardware is healthy.

\[ \mathcal R=\mathcal R_{\mathrm{phys}}\cup\mathcal R_{\mathrm{epist}}, \qquad Z(t)=\langle x(t),\boldsymbol{\rho}(t),cfg(t),\boldsymbol{\alpha}(t)\rangle. \] The controlled state contains dynamics, resource health/validity, the active configuration and mission priorities.

For an epistemic resource, \(\rho_m\) is derived from prediction error and time since validation. A model can therefore be treated by the same capability machinery as a sensor or actuator — but its validity can also be restored deliberately.

Back to overview →Next →Measure what the current architecture can still deliver.