Planner-LM Composites — the grounding problem expressed as “no state simulator” and its architectural solution
Energy-Based Models — constraint-satisfaction landscapes as the substrate for grounded reasoning; Aleph + Lean as verifier-backed grounding in formal math
Transpilation as a Grounding Strategy — formal-language grounding is a gradient; deterministic transpile to a grounded language as the fix (6502→C, COBOL modernization)
Repairing LLM Code — The Two Oracles — correctness grounds downward (diff vs a lower artifact); readability has no lower oracle, only transfer-to-a-reader
Comments and the Distance to an Oracle — oracle availability is a distance, not a binary — and the class with no oracle (intent) is the class that must be kept
The Contract Model vs. the Substrate Model — a per-delivery contract (delivery / constraints+guardrails / proof artifact / outside-verification / owner) is the vault’s architecture with the accumulation stripped out — contracts don’t compound, substrate does; plus the artifact-vs-reader split in verification independence, and tests written only to pass as oracle collapse
Oracles Are Objective Functions — verification and optimization are one structure from two ends: an objective you can evaluate is not an optimum you can locate, which is why oracle work exists at all and why every oracle in use is partial by construction
Eyes and Ears for LLMs — the Audio Translator — sound’s three grounding regimes — echolocation (physics), speech (transcripts), music (half-grounded); draw the verifiable/statistical line per layer and let the verifiable one overrule
The AI Decomp Wave — Outside Confirmation — agent decompilation’s “answer key” = the vault’s correctness oracle; flashy ports/mods vs. mechanics; idea/expression legality