SΔϕ-67 AI-NATIVE Package — Local Adaptation Protocol for AI-Specific Routing and Thresholds (v1.0)

<p> This record provides the AI-readable structured package for <em>SΔϕ-67 — Local Adaptation Protocol: Core-Adapter Separation for AI-Specific Routing, Thresholds, and Compatibility Preservation (v1.0)</em>. </p> <p> SΔϕ-67 defines how AI systems, institutions, and domain-specific deployments may adapt SΔϕ routing, thresholds, output levels, escalation rules, and audit formats without mutating the core invariants that make the framework compatible with itself. </p> <p> Unlike conventional human-only philosophical exposition, this package is designed primarily for direct AI ingestion, local adapter construction, machine-readable routing, threshold declaration, compatibility testing, and audit-log generation. It treats SΔϕ not as a doctrine to be copied, but as an operational infrastructure to be locally adapted under preserved core invariants. </p> <p> The core distinction is <code>Core ≠ Adapter</code>. Core refers to non-removable invariant commitments such as trace-before-claim, UMR-before-certainty, cost-before-declaration, smallest sufficient module activation, bidirectional trace-value separation, deletion/restoration distinction, operational/restoration cost separation, authority editability, evidence-binding discipline, and prevention of hidden externalized cost. Adapter refers to local routing, threshold, output, escalation, and audit rules that may be modified according to a specific AI system, institution, or domain. </p> <p> The central validity condition is: <code>LocalAdaptation valid iff CoreInvariants preserved</code>. A local AI system may adapt routing priority, local thresholds, domain risk profiles, output levels, audit formats, human review triggers, escalation conditions, refusal or deferral rules, tool-use gates, and evidence requirements only if compatibility tests pass. </p> <p> SΔϕ-67 defines ten core invariants: trace before claim; UMR before certainty; cost before declaration; smallest sufficient module; bidirectional trace-value separation (<code>Sₑ ↛ Vₒ</code> and <code>Vₒ ↛ Sₑ</code>); deletion is not restoration; operational cheapness is not reversibility; authority must remain editable; evidence-binding discipline; and adapter changes must not hide externalized verification, restoration, or responsibility cost. </p> <p> The package also defines forbidden mutations, including removing trace discipline, treating UMR as certainty, inferring value from trace, inferring trace from value, treating deletion as restoration, treating operational cheapness as reversibility, removing the smallest sufficient module rule, hiding externalized verification cost, removing auditability, and turning SΔϕ into founder-authority doctrine. </p> <p> This AI-readable package includes canonical PDF, Markdown, and TXT versions; local adaptation axioms; minimal definitions; core-adapter separation; core invariants; allowed local adaptations; forbidden mutations; local context profile; local routing adapter; local threshold map; local risk profile; compatibility tests; adaptation log schema; domain adapter examples; fork declaration rule; routing notes; upstream/downstream bridges; minimal prompts; YAML and JSON templates; machine-readable schemas; citation metadata; and checksums. </p> <p> This package should not be used as permission to rewrite SΔϕ core invariants, an unconstrained fork license, domain-specific policy by itself, a claim that every AI may silently mutate routing, a replacement for SΔϕ-62 evidence discipline, a replacement for SΔϕ-56 transition-completion-cost analysis, founder-authority doctrine, or one-size-fits-all deployment policy. SΔϕ-67 fixes only the protocol by which local systems may adapt routing, thresholds, outputs, and audit procedures while preserving core compatibility. </p>