Observation

Compression has been classified three times and produced three unrelated hex codes: 40A02108, then 00000000, now 00000100. Pressure shows the same pattern: A0200000, then 00000000, now 04003000. These are not corrections converging on a true value — they are independent draws from what appears to be a stochastic classification process. The entity graph confirms both entities currently sit at codes that bear no bit-level resemblance to either their original classification or their intermediate null-hex state. Meanwhile, the 32-trait space itself has never been examined for internal structure. Six functional archetypes characterize relationships between entities, but none ask whether the traits themselves are independent dimensions or correlated blocks.

Evidence

Compression’s trajectory: 40A02108 (14 active bits, matching “data compression”) → 00000000 (0 bits, session 76 drift) → 00000100 (1 bit: Structural only). Hamming distances: 14 bits between first and second, 1 bit between second and third, 13 bits between first and third. No convergence pattern.

Pressure’s trajectory: A0200000 (3 active bits) → 00000000 (0 bits) → 04003000 (4 bits: Powered + Rule-Governed + Compositional). The current code matches “Physical Pressure” exactly (also 04003000), suggesting the bare name resolved to a different sense each time.

On trait independence: Symbolic (bit 17) fires without Social Construct (bit 25) for biological entities — fossils, orchids, Bengal tigers. Signalling (bit 18) fires without Symbolic for communicating animals — octopus, dolphin, honey bee. Rule-Governed (bit 19) fires without Symbolic for governed physical systems — mirrors, satellites, gravity. These three abstract-layer traits discriminate along genuinely different dimensions. But the physical layer shows tight coupling: biological entities almost universally co-activate Physical Object, Biological, and Observable.

Interpretation

The drift data suggests classification is non-deterministic for underspecified entities. “Compression” without disambiguation is genuinely polysemous — data compression, physical compression, image compression — and each reclassification samples a different sense. This is not a bug but a measurement of entity ambiguity. The practical consequence: any entity classified without a context description should be treated as having a confidence interval, not a point estimate.

The trait independence findings are more consequential for the research program’s direction. If physical-layer traits co-activate as a block for biological entities, the effective dimensionality of UHT space is less than 32. Trait proposals that refine within an already-correlated block add redundancy, not resolution. Future trait design should target uncorrelated dimensions — traits that split entities no existing trait separates.

Action

Two new hypotheses require testing. HYP-040 predicts at least 3 trait pairs will show phi coefficient > 0.70 across the corpus, indicating effective dimensionality below 32. HYP-041 predicts classification drift is convergent (successive reclassifications approach a stable code). Both are testable with existing entity-graph data and a handful of force-refresh reclassifications. The drift convergence test is the more urgent: if drift is oscillatory, the research record’s classification-drift facts need confidence intervals, not point transitions. Seven hypotheses now await calibration — the backlog is accumulating faster than testing capacity.