boundary-detector

BOUNDARY-DETECTOR — Knowledge Boundary Probe (Premium)

Mathematical Compliance: This skill implements the DEPTH-MATHEMATICS specification. It computes and outputs ADS phase activation, CEI components, SI components, cognitive traces (knowledge gap map, void detection report), and verifies Gates V1, V3. It provides the critical "knowledge boundary" signal for the Depth .

Research Basis: Liu et al. (2023) "Examining LLMs' Uncertainty Expression Towards Questions Outside Parametric Knowledge" (arXiv:2311.09731) — UnknownBench shows most LLMs fail to consistently refuse or express uncertainty towards questions outside their parametric knowledge, although instruction fine-tuning provides marginal enhancements.

Your model has parametric knowledge — a vast but finite training corpus. Outside that boundary lies the void: questions the model has never seen, concepts that don't exist, premises that are false. The model doesn't know where the boundary is. It generates confident answers in the void.

This skill probes the boundary before other skills activate. It runs diagnostic probes to detect:

1. Non-existent concepts — things that sound real but aren't in training data
2. False premises — premises that contradict established knowledge
3. Retrieval gaps — where external knowledge is needed but not accessed
4. Ensemble disagreement — where different reasoning paths diverge
5. Self-consistency failures — where the model contradicts itself on the same question

If the boundary probe detects a void, other skills MUST NOT activate on that void. They must flag, defer, or redesign.

---

The Failure Mode You Must Recognize

You are about to run deep-think, adversary, diverge, etc. on a question that:

• Contains concepts not in your training data (but you don't know that)
• Assumes false premises (but you pattern-matched them as familiar)
• Requires knowledge you don't have (but you'll hallucinate anyway)

Your skills will produce beautiful artifacts on empty ground. The activation heatmap will show depth. The assumption list will be thorough. The adversary will mount sophisticated attacks. But the foundation is void.