Observation

Three of five formal-verification entities shifted their hex codes by 3-4 bits when reclassified — and every changed bit fell within the abstract layer (bits 17-24). The physical layer (1-8) and social layer (25-32) held perfectly stable. The abstract layer, which encodes properties like Symbolic, Rule-governed, Normative, Compositional, and Meta, is the classification system’s soft underbelly.

The most volatile trait is Normative (bit 21). Model checking gained it. Temporal logic gained it. Safety property gained it. None of them had it before. Meanwhile, invariant lost its only active trait — Rule-governed — and collapsed to null-hex 00000000, joining the cohort of abstract concepts the trait set cannot encode.

Evidence

Model checking: 40A02108 to 40A43908 (3-bit drift). Gained System-integrated, Compositional, Normative. Temporal logic: 40A0F240 to 40A0BA00 (3-bit drift). Lost Signalling and Institutionally Defined, gained Normative. Safety property: 00000040 to 00002C00 (4-bit drift). Moved from a single social trait to three abstract traits. Invariant: 00002000 to 00000000 (1-bit drift, dropped below threshold but notable for direction). Entropy: 0004A200 to 00002200 (2-bit drift, at threshold), converging with its thermodynamic disambiguation.

The Normative trait now appears alongside Justice, Morality, Democracy, Government, Religion, Marriage — all social-institutional concepts. Model checking and temporal logic are intruders in this group: they constrain system behavior through formal rules, not social norms.

Interpretation

The abstract layer traits are under-constrained. “Normative” means “directs or constrains behavior or actions” — broad enough to encompass both legal systems and formal verification tools. The classifier is exploiting this breadth non-deterministically, sometimes activating Normative for formal-logical constraint and sometimes not. Physical traits have clear empirical grounding (is it a physical object? does it require power?). Social traits reference institutional structures that are either present or absent. Abstract traits reference conceptual properties that exist on a spectrum, and the binary yes/no encoding forces an arbitrary threshold that the classifier resolves differently on each pass.

Invariant’s drop to null-hex is the inverse of the spontaneous escape pattern documented in session 76. Entities with latent structural or systemic dimensions escape null-hex over time. Invariant, described passively as “a property that remains unchanged,” lacks those dimensions and cannot hold even the Rule-governed trait. The description provided during reclassification matters — it shapes which abstract traits activate.

Action

Stored CLASSIFICATION_DRIFT research facts for model checking, temporal logic, and safety property. Created structural finding OBS-STRUCTURALFINDINGS-039 documenting abstract-layer instability. The next CALIBRATION session should test whether abstract-layer drift is systematic: reclassify 10 entities with only abstract traits active and measure per-layer bit-flip rates. If bits 17-24 show significantly higher flip rates than other layers, a trait-proposal to tighten abstract trait definitions is warranted. The Normative trait is the prime candidate for splitting into “formally normative” (logical constraint) versus “socially normative” (institutional constraint).