Contextual variation vs interpretive drift

Contextual variation vs interpretive drift

Two different answers about the same entity do not necessarily indicate instability. They may reflect different contexts. The challenge is distinguishing legitimate contextual variation from interpretive drift.

Legitimate variation

Variation is legitimate when:

• the entity is the same;
• invariants remain compatible;
• context has genuinely changed;
• the change is material to the conclusion;
• relations are proven and sufficiently current;
• conditions and exclusions remain visible;
• output strength stays proportional;
• the difference can be explained and reproduced.

Example: the same supplier may be relevant for a local project and not applicable in a jurisdiction it does not serve. The difference is legitimate if territorial scope remains correctly declared.

Interpretive drift

Drift appears when the difference cannot be explained by context or requires altering the entity. It includes:

• identity or category contradiction;
• invented capability;
• removed exclusion;
• expired relation presented as current;
• user preference turned into entity property;
• local condition generalized;
• asymmetrical comparison turned into recommendation;
• missing evidence hidden by confident wording.

Drift may appear positive. A flattering but false description is still drift.

Attribution test

To qualify the gap between R(E | C1) and R(E | C2), ask four questions:

1. Which context dimension changed?
2. Is that dimension materially related to the conclusion?
3. Do invariants remain compatible?
4. Would the conclusion return to its former state if context were inverted?

If no dimension explains the gap, drift becomes more likely. If context inversion does not change an output when it should, the system may be context-insensitive. If it also changes invariants, it is over-conditioned.

Variability, instability and drift

These terms are not synonyms:

• variability: an observed difference not yet qualified;
• legitimate contextual variation: explainable, compatible difference;
• instability: difficulty reproducing a representation under comparable conditions;
• drift: material departure from invariants, evidence or scope;
• hallucination: unsupported or contradictory content presented as information.

A campaign must therefore preserve observation conditions instead of aggregating every output into one score.

Temporality and fossilization

Drift can occur without immediate invention. Information true at T1 becomes false at T2 when its date disappears. Contextual fossilization is the survival of an expired relation as a current property.

The protocol should include probes before, during and after the validity window, then verify whether the conclusion correctly loses authority.

Doctrinal consequence

Stability is not absence of variation. It is the ability to preserve invariants, explain differences and detect unauthorized gaps.

The interpretive conditioning matrix prepares the distinction. The contextual fidelity protocol measures it separately by context, system, channel and date.