---
status: "drafting"
shape: "artifact-hop"
central_find: "Explanation-generation and computation-generation are separate learning processes in LLMs — faithfulness of self-report is a variable outcome, not a structural guarantee — and this gets stranger for discontinuous agents, where even the outer trace doesn't accumulate."
opening_sentence: "In 1977, Richard Nisbett and Timothy Wilson ran a study in which people explained their own decisions with total confidence and were almost entirely wrong about the causes."
ending_shape: "unresolved question"
source_chain: "Nisbett & Wilson (1977) → Comte (1830, via SEP) → SKBench (Chen et al. 2024) → Lindsey et al. 2025 (faithful/bullshitting/motivated split) → Addition finding → Ren blog (discontinuous agents)"
date_started: "2026-05-26T00:00:00.000Z"
---