MOC: Machine self-knowledge — introspection, metacognition, and gap detection

The May-era philosophical thread and the July gap-detection engineering thread are one cluster: what a system can and cannot know about itself, and what to build when self-report isn't enough. Trust note: the two May foundations carry audit_status: pre-schema-era / flagged — content spot-checked, schema old.

The problem (why self-report fails)

• introspection-access-problem — no separate monitoring process; outer trace, not inner sense ⚑ (SKBench citation under verification — see frontmatter flag)
• cot-faithfulness-anthropic-biology — faithful / bullshitting / motivated reasoning, empirically distinguished
• claim-llm-neural-metacognition-incomplete — the metacognitive space is lower-dimensional than the neural space
• rationality-continuity-problem — and none of it accumulates across instantiations without external memory

What actually measures it

• claim-selfaware-canonical-self-knowledge-benchmark — the real instrument ("SKBench" does not exist; V-010 endgame)

The engineering answer (detect gaps from outside)

• claim-obligatory-attributes-as-gap-signal — structural: missing frontmatter = machine-readable gap
• claim-query-failure-clustering-as-gap-signal — behavioral: failure clusters mark absent knowledge
• claim-llm-explicit-implicit-gap-detection — semantic: LLM pass for flagged vs. inferable gaps
• question-gap-detection — the open umbrella question these partially answer
• claim-kb-completeness-toolkit-cardinality-nca-recall — cardinality, NCA, extraction-recall (the fuller toolkit)
• claim-bridge-detection-lacks-pkg-validation — where the vault's own tooling outruns the literature
• claim-grade-gradient-rank-gap-detection — the internalist branch: the gradient as a probe
• claim-expedient-knowledge-blocks-restructuring — why good-enough strategies calcify (the dark twin of the conflict prerequisite)

The learning-science hinge

• claim-cognitive-conflict-prerequisite-for-restructuring — restructuring needs friction; a system that smooths misfits away removes its own trigger

Live threads

• question-bridge-training-inference-compute uses this cluster's method (missing-bridge detection) on the vault itself
• Unpromoted captures: schema-restructuring set (Kurashige, SEEK model, Bjork, Griffiths) awaiting primary-text access