0005 — The nonlinguistic core, and the qualia stake

2026-08-22. Researcher rulings, verbatim:

"i think what we should do is not lie to ourselves. this thing has to be nonlingquistic right. shapes shaped shaping shaping shapes."

"if we do this right we get to destroy the concept of qualia"

The honesty audit (the lie we were telling)

Day one's stack is language wearing a graph costume. Relation labels are English words; parses come from language models reading prose; even the planned SVG sense routes through named XML tags. The instrument's own first data already voted against this: prose was the worst teacher of structure (P4), and removing nouns IMPROVED recovery (P3). The system's first result argues for its own de-linguification — shapes shaping the shapers.

Ruling: the core is nonlinguistic

• Content of core representations is structure: adjacency, geometry, order, dynamics, response profiles. Labels may ride along as decoration; nothing in comparison may require them (GED already prices renames at zero; that accident is now policy).
• Comparison never touches a token: WL colors, spectra, edit costs, response surfaces. Numbers and topology end to end.
• Language is demoted to the boundary: one sense among several on the way in (and per P3/P4, a poor one), and the human report layer on the way out. Nowhere in the middle.
• First fully nonlinguistic path to build: SVG → primitive graph → WL/spectral → comparison. Deterministic, zero tokens, no model priors anywhere. This is the instrument's first honest sensory channel and the calibration standard for every model-mediated sense (ADR forthcoming from the senses discussion: rasterize the SVG, feed the vision model, measure the eye's transfer function against known truth).

The qualia stake, operationalized

The classical qualia claim: something about experience ("what it is like") survives NO translation into any other representation — private, ineffable, structure-exceeding.

In this project's vocabulary that is a precise, testable shape claim: qualia are proposed as the maximally representation-EVASIVE residual — content that appears in no mapping, with a floor that no increase in representational resolution lowers.

Note the symmetry with §22: "God" was provisionally defined as the maximally representation-INVARIANT structure — what refuses to disappear. Qualia-as-claimed is the opposite pole of the SAME limiting operation — what refuses to appear. The God Machine's one operation (increase representational distance, measure what is conserved) has two limit objects, and this project now names both.

Discipline (§106, §22 mirrored — binding)

We do not build the machine to destroy qualia. We build the machine that makes qualia pay rent. Two honest outcomes:

1. The residual shrinks as structural resolution grows, with no principled floor — "ineffable" decays into "untranslated at current apparatus configuration" (§107 language), and the concept dissolves the way élan vital did: not refuted, obsoleted.
2. The residual hits a floor that added resolution does not touch — which would be the most interesting measured result in the neighborhood of philosophy of mind in a long time.

Do not optimize toward outcome 1. Do not search for qualia's absence. (§22's "do not search for God," mirrored exactly.)

Prior art dent worth keeping in the quiver: the human color solid is ASYMMETRIC (blue–yellow vs red–green structure; dark yellow is brown), so the classic inverted-spectrum thought experiment is already structurally detectable in principle — an existing empirical shrinkage of qualia's claimed privacy, made by exactly the move this instrument generalizes: raise the structural resolution until the "undetectable" becomes a measurement.