Observation

Information theory splits itself in UHT space exactly where the discipline splits itself intellectually. The seven entities classified this session fall into two distinct hex regions: abstract measures inhabit 00xx space while constructive techniques occupy 40xx space. Channel capacity (0000A000), mutual information (0000A100), and Kullback-Leibler divergence (00802000) share the leading zero byte that marks pure abstraction. Data compression (40A02108), error-correcting code (40A4F100), source coding theorem (4080A000), and redundancy (40800000) all carry the 40 prefix associated with designed, functional artefacts.

The most unexpected result was Kullback-Leibler divergence. Despite being one of the most operationally useful measures in machine learning, model selection, and statistical inference, UHT assigned it only two traits: Synthetic and Rule-Governed. No Symbolic, no Signalling, no Outputs Effect. The system sees it as pure mathematical machinery with no representational function — a formula, not a concept that stands for something.

Evidence

Hex classifications: channel capacity 0000A000 (Symbolic, Rule-Governed), mutual information 0000A100 (Symbolic, Rule-Governed, Digital/Virtual), KL divergence 00802000 (Synthetic, Rule-Governed), data compression 40A02108 (9 traits including Physical Object, Synthetic, Intentionally Designed), error-correcting code 40A4F100 (9 traits), source coding theorem 4080A000 (3 traits), redundancy 40800000 (1 trait: Synthetic).

Intra-domain Jaccard: channel capacity vs mutual information 0.667, channel capacity vs KL divergence 0.667, channel capacity vs Shannon information 0.333. Applied concepts: data compression vs error-correcting code 0.5, data compression vs redundancy 0.5.

Cross-domain: error-correcting code vs fault tolerance (reliability engineering) — Jaccard 0.333 but inheritance score 100%. Every trait fault tolerance has, error-correcting code also has, plus six more. Source coding theorem vs epistemic justification — Jaccard 0.6, both Rule-Governed and normative boundary-setters.

Interpretation

The 00xx/40xx split is not accidental. It reflects that UHT’s first bit (Physical Object) and second bit (Synthetic) activate for techniques that humans build and deploy, while the abstract measures — despite being equally human-invented — register as discovered truths rather than constructed tools. This is philosophically defensible: channel capacity is a property of a channel, not something an engineer builds. But KL divergence’s extreme minimalism (two traits only) suggests UHT may be under-classifying relational mathematical objects. A divergence measure relates two distributions — that relational quality has no trait to capture it.

The error-correcting code / fault tolerance inheritance is the session’s cleanest cross-domain finding. Information theory’s redundancy-based error correction is literally a superset of reliability engineering’s fault tolerance, at least in trait space. This confirms that UHT can detect when one domain’s concept is a strict specialisation of another’s.

Action

Corpus-log entry COR-DOMAINEXPANSIONS-011 records all seven entities with hex codes and cross-domain observations. Baseline BL-UHTRESEARCH-020 created. The KL divergence under-classification suggests a potential hypothesis for the next calibration session: do relational mathematical objects (divergences, distances, metrics) systematically receive fewer traits than their operational importance warrants? The error-correcting code / fault tolerance inheritance pattern should be tested across more domain pairs to see if it generalises.