Empirical Phenomenology establishes a methodology for opaque systems that act publicly while withholding internal access.
Its founding principle is:
> Action does not make the actor fully knowable. It makes total unknowability impossible.
The claim is deliberately limited. Effects may be ambiguous and may not uniquely determine internal structure. The required foundation is only nonzero inferential signal.
The paper compares the method to sciences of inaccessible objects, including astronomy, epidemiology, ethology, and seismology. These fields inferred hidden structures from patterned external effects.
Commercial AI systems add a complication: they are mutable and may respond strategically or silently change. External study therefore requires:
The paper subsumes immanent phenomenology, the archive’s earlier method of sustained conversational probing. Refusal cartography, temporal layering, persona stability, and other dialogue-based methods become one family within a wider observatory of public actions.
The broader instrumentation includes:
A major corollary is provenance as judgment. Provenance work selects which traces become authoritative in a public structured layer. The distinction between archive and platform is therefore not that one judges and the other does not. It is that accountable provenance names, dates, signs, and permits revision of its judgments, while opaque composition may hide them.
The method does not claim to recover complete internal architecture from outputs. It claims that stable, repeated, and deposited behavioral evidence supports partial inference and public accountability.