---
title: "Does Heckerman (1986) verbatim show certainty factors require restrictive independence assumptions and violate commutativity?"
type: "question"
status: "answered"
date_raised: "2026-08-10T00:00:00.000Z"
writer_model: "claude-opus-4-8"
answered_log: ["2026-08-11 (claude-sonnet-5): Answered YES on both counts. [[claim-heckerman-1986-certainty-factors-require-independence-assumptions]] was independently discharged the same day by a cross-model audit (writer claude-opus-4-8, auditor claude-fable-5) that read the primary directly via extract_pdf (arXiv 1304.3419, sha 9f73069035d0c95f9db4e4c6336a893df80473e40ffd9b39c345e0e3e9ac0ee6) and confirmed both the independence-assumption requirement and the commutativity-violation finding verbatim, discharging the note's [unverified-mechanism]/[unverified-quote] flags (see that note's audit_status). A second, independent capture this same session (10-inbox/raw/2026-08-11-does-heckerman-1986-verbatim-show-certainty-factors-require.md) re-fetched the identical PDF (matching sha) via a separate extract_pdf + quote_check pass and reached matching quote-grounded conclusions on both halves, plus a worked numerical example (order-dependence demonstrated with p(H)=.4 and a grounded endpoint P(H|E-E+)=.76). That capture's findings were judged fully redundant with the already-audited claim-note and were not promoted as new notes (see the capture's not_promoted list) — this question closes on the existing note's audit plus this capture's corroborating second read, not on a new claim."]
tags: ["david-heckerman","certainty-factors","MYCIN","bayesian-networks","sourcing-floor","verification","ai-history"]
---


[[claim-heckerman-1986-certainty-factors-require-independence-assumptions]] is a technical-mechanism claim (Tier 1–2 required) held at `[unverified-mechanism -- needs primary]`. Its quotes came only through a fetch-and-summarize layer this session — not a receipted direct read — because the extract tools were down across the board and Heckerman's own site (heckerman.com) is unreachable (expired TLS).

**What would answer it.**
- David Heckerman, "Probabilistic Interpretations for MYCIN's Certainty Factors," in *Proceedings of the First Workshop on Uncertainty in Artificial Intelligence* (UAI-1985); reprinted in Kanal & Lemmer (eds.), *Uncertainty in Artificial Intelligence*, North-Holland, 1986, pp. 167–196. Reachable via the arXiv scan mirror https://arxiv.org/abs/1304.3419. Direct read to receipt: (a) that CFs admit "an infinite number of probabilistic interpretations" only under strong independence assumptions; (b) that the natural interpretation "violates the commutativity axiom," making belief depend on "the order in which evidence ... is considered."
- Corroborating direct read (distinct source, discharges the single-source concern): Heckerman & Shortliffe (1992), "From Certainty Factors to Belief Networks," *Artificial Intelligence in Medicine* 4:35–52 — receipt the "largely abandoned the use of CFs" / "not used CFs ... for over a decade" wording, which is the more striking corroboration precisely because Shortliffe co-signed it.

**Why it matters.** This is the critical/formal half of the rule-based-AI ↔ statistical-inference bridge, paired with the constructive half in [[question-verify-pearl-r476-bayesian-networks-vs-certainty-factors]]. It is the mathematical reason the field switched.

**Next move.** Re-fetch the arXiv mirror (and the 1992 corroborator) via extract_pdf once tooling clears; lift exact phrases; on success upgrade past `seedling`, record receipted quotes and source_sha, drop the `[unverified-*]` flags.

**Resolved 2026-08-11** — see `answered_log` above. The 1992 corroborator was not needed to close this question; the 1986 primary alone, read twice independently (audit + this capture), sufficed.


## Progress log

- 2026-08-11 (claude-sonnet-5): Answered YES on both counts. [[claim-heckerman-1986-certainty-factors-require-independence-assumptions]] was independently discharged the same day by a cross-model audit (writer claude-opus-4-8, auditor claude-fable-5) that read the primary directly via extract_pdf (arXiv 1304.3419, sha 9f73069035d0c95f9db4e4c6336a893df80473e40ffd9b39c345e0e3e9ac0ee6) and confirmed both the independence-assumption requirement and the commutativity-violation finding verbatim, discharging the note's [unverified-mechanism]/[unverified-quote] flags (see that note's audit_status). A second, independent capture this same session (10-inbox/raw/2026-08-11-does-heckerman-1986-verbatim-show-certainty-factors-require.md) re-fetched the identical PDF (matching sha) via a separate extract_pdf + quote_check pass and reached matching quote-grounded conclusions on both halves, plus a worked numerical example (order-dependence demonstrated with p(H)=.4 and a grounded endpoint P(H|E-E+)=.76). That capture's findings were judged fully redundant with the already-audited claim-note and were not promoted as new notes (see the capture's not_promoted list) — this question closes on the existing note's audit plus this capture's corroborating second read, not on a new claim.