Summary

Coordinated update across the full UHT classification stack: trait definitions, classification prompts, CLI tooling, research protocol, and QC bot protocol. All versions are now tracked and queryable via the CLI.

Version manifest

Component	Version	Notes
Trait definitions	2.0	Expanded definitions with classification guidance, edge cases
Prompt spec	1.1	32 per-trait prompts, gpt-4o-mini, temperature 0.3
uht-substrate CLI	0.4.5	New: --force-refresh, --context, --format pretty, trait-prompts
airgen-cli	0.4.0	Server-side filtering, --document/--section filters
Research protocol	4.1	Context-enriched classification, force-refresh for drift checks
QC protocol	1.0	Context-enriched classification, force-refresh for reclassification

Changes

Trait definitions v2.0

Each of the 32 traits now carries an expanded_definition and a URL linking to its canonical page on universalhex.org. Viewable via uht-substrate traits --format pretty.

Prompt spec v1.1

Each trait has a dedicated classification prompt with: core definition, APPLICABLE/NOT APPLICABLE examples, and edge cases. The --context flag on classify injects into the {{entity_description}} template variable, meaning every trait evaluator now receives the entity description. Viewable via uht-substrate trait-prompts --bit N --format pretty.

CLI 0.4.5

• classify --force-refresh — bypasses entity-name cache, forces fresh 32-trait evaluation. Essential for drift detection.
• classify --context "<description>" — now feeds into trait prompts as {{entity_description}}, directly influencing classification quality.
• --format pretty on traits, trait-prompts, entities list — colour-coded, human-readable output.
• trait-prompts command — inspect the exact prompts used for classification.
• --version now reports correct version number.

Research protocol v4.1

• All classify calls now require --context with a 1-2 sentence description covering domain, key properties, and distinguishing characteristics.
• INTEGRITY drift checks use --force-refresh to bypass cache.
• CALIBRATION and APPLICATION test phases include context guidance.
• CORPUS_EXPANSION classify calls include context examples.
• Entity count updated to 2900+.
• CLI reference expanded with traits, trait-prompts, --format pretty.

QC protocol 1.0

• All classify calls use --context with rich descriptions.
• DRIFT_CHECK uses --force-refresh for meaningful reclassification.
• GRAVITY_WELL_ANALYSIS reclassification uses --force-refresh with context.
• Phase 4 (approved action execution) has two RECLASSIFY paths: same-name with --force-refresh, or name-change with delete-then-classify.

Implications

Prior to this update, all classifications were performed with bare entity names and no description. The 2900+ existing entities were classified without the benefit of {{entity_description}} in trait prompts. This means:

1. Future --force-refresh reclassifications with rich context may produce different hex codes — expected and desirable.
2. The QC bot’s DRIFT_CHECK task class can now detect genuine classification improvements, not just cache artifacts.
3. CORPUS_EXPANSION sessions will produce higher-quality initial classifications going forward.

All components are version-tracked and queryable: uht-substrate --version, uht-substrate traits --format pretty (trait version), uht-substrate trait-prompts --format pretty (prompt version).