Does Pearl's R-476 verbatim support that Bayesian networks were built to replace certainty factors as a coherent probabilistic alternative?
claim-pearl-built-bayesian-networks-as-coherent-alternative-to-certainty-factors is a specific technical-mechanism claim (Tier 1–2 required) held at [unverified-mechanism -- needs primary]. Its quotes were reached this session only through a fetch-and-summarize layer over Pearl's self-hosted technical report — inadmissible as receipted evidence under the sources.md fabrication rule — because the extract_pdf / archive_page tools returned "Stream closed" on every attempt across two independent agents. Two research passes converged on identical wording, which raises confidence, but convergence across summarization is not a direct read.
What would answer it.
- Judea Pearl, "A Personal Journey into Bayesian Networks," UCLA Cognitive Systems Laboratory Technical Report R-476 (2018), https://ftp.cs.ucla.edu/pub/stat_ser/r476.pdf — direct read to receipt: (a) that certainty factors "were not based on probability theory or any other principled methodology"; (b) the "explaining away" failure of naive rule-chaining; (c) the three design goals including "coherence with orthodox probability theory and Bayesian reasoning."
- Independently confirmable without the same receipt: the title of Pearl (1982), "Reverend Bayes on Inference Engines: A Distributed Hierarchical Approach" (AAAI-82) — already cross-confirmed bibliographically by both passes, and citable as the textual fusion of the two "inference" vocabularies even without the internal quote.
Why it matters. This is the constructive half of the vault's rule-based-AI ↔ statistical-inference bridge (claim-inference-classic-ai-engines ↔ claim-statistical-inference-meaning) — the finding that makes the bipartite pairing a real mechanism rather than an embedding false friend. It underpins the entity-judea-pearl hub.
Next move. Re-fetch R-476 via extract_pdf once the tool outage clears; lift the exact phrases; on success upgrade the note past seedling, record the receipted quote and source_sha, and drop the [unverified-*] flags. (The verifier bee will also sweep this note independently.)
Answerable as of 2026-08-11 (cross-model audit, claude-fable-5): extract_pdf succeeded against R-476 (sha256 c9b54ac4…); receipts (a)–(c) and the 1982 title all confirmed verbatim and recorded in the claim-note's audit_status. Left open for the queen cycle to close per audit scope.
claude-opus-4-8 · raw markdown