v1.1.0 — Oracle, Gap Detection, Strategist

What's New

Knowledge Gap Detection

• knowledge_boundary + knowledge_audit prisms detect what analysis can't verify
• Found real errors: asyncio.RWLock doesn't exist, O(n)+O(n) mislabeled as quadratic
• Two prisms are complementary — neither catches all errors alone

Self-Correcting Analysis

• L12-G: single-pass analyze → audit → self-correct. 90% zero-confabulation rate (N=10)
• Oracle: 5-phase max-trust analysis (depth → typing → correction → reflection → harvest)
• Every claim tagged [STRUCTURAL], [DERIVED], [KNOWLEDGE], or [ASSUMED]

Strategist

• Meta-agent that knows all 48 prisms + 12 modes
• Plans optimal tool sequence for any analytical goal
• 2-call: plan + adversarial critique

New CLI Flags

• --trust — Oracle mode (max epistemic trust)
• --confidence — tag claims HIGH/MED/LOW/UNVERIFIED
• --provenance — source attribution per finding
• --depth shallow|standard|deep|exhaustive — aspiration level control
• --models — show current model routing table

Key Finding

A prompt with scrambled vocabulary (nonsense words, preserved format) scored 10/10. Format carries meaning independently of content.

Stats

• 1000+ experiments across 41 rounds
• 48 prisms, 12 scan modes, 7 CLI flags
• 30 literature reviews (~400 papers surveyed)
• 19 new principles (P206-P224)
• Persistent knowledge base (.deep/knowledge/)