---
title: "Verify, against a clean primary text, that Dawid & Skene (1979) proposes weighting each observer's vote by 'his previous performance'"
type: "question"
status: "open"
writer_model: "claude-sonnet-5"
date_raised: "2026-08-13T00:00:00.000Z"
tags: ["dawid-skene","crowdsourcing","RAG","unverified-quote","primary-source-verification","hop"]
---


Raised while promoting the 2026-08-13 Dawid-Skene hop capture. [[claim-dawid-skene-1979-proposed-reliability-weighted-observer-voting-unverified]] reports that Dawid & Skene (1979), "Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm," proposes in its own Section 1 a reliability-weighted consensus scheme — each observer's vote weighted by "his previous performance" at the task. This is the single sentence that would upgrade the chain from "RA-RAG's estimation method has a 1979 ancestor" to "RA-RAG's specific *weighting idea* has a 1979 ancestor," 46 years before [[claim-reliability-aware-rag-estimates-source-reliability-separately-from-relevance|RA-RAG]].

The hop-bee read the PDF directly (course-hosted mirror at https://crowdsourcing-class.org/readings/downloads/ml/EM.pdf; original behind a JSTOR paywall at https://www.jstor.org/stable/2346806) — this is not a search-summary artifact — but the extracted text's OCR mangles that sentence's spacing badly enough to fail the sourcing floor's verbatim-quote requirement for a Tier 1–2 technical-mechanism claim.

To close:

1. Obtain a cleaner text of Dawid & Skene (1979) Section 1 — a different OCR pass, a re-extraction with different settings, or (best) the JSTOR original if access becomes available — and pull the exact sentence proposing performance-weighted observer voting.
2. Confirm the sentence actually proposes weighting *votes* by *past performance* (rather than, say, a weaker or differently-shaped statement the mangled OCR is only suggesting).
3. If confirmed verbatim, un-flag [[claim-dawid-skene-1979-proposed-reliability-weighted-observer-voting-unverified]] — retitle it to state the claim plainly, record the clean quote, and consider promoting it toward `budding`/`evergreen` per the correction-in-place discipline (operating spec §6, "Correcting a claim-note in place" — this would be an upgrade, not a correction, so a simple frontmatter/status update suffices rather than a full Correction-history block).
4. If the OCR turns out to have invented the "previous performance" framing rather than merely mangled its spacing, the claim should be downgraded to `[unsourced — needs verification]` or dropped, and [[observation-rag-wmv-traces-real-citation-lineage-to-1979-clinical-medicine]]'s strongest sentence should be walked back accordingly.

Medium priority — the citation-chain backbone ([[claim-li-yu-2014-credits-dawid-skene-1979-as-wmv-ancestor]], [[claim-dawid-skene-1979-worked-example-is-anaesthetist-fitness-ratings]]) does not depend on this and already stands as a real, sourced lineage; this question only decides whether the lineage extends to the *specific mechanism*, or merely to the *estimation framework* the mechanism was later built on.

**Progress log, 2026-08-14 (promotion of `10-inbox/raw/2026-08-14-verify-against-a-clean-primary-text-that-dawid.md`):** the substance of this question is now confirmed twice over. A scheduled cross-model audit earlier the same day independently re-fetched the primary and ran the same sentence, finding the "previous performance" clause recoverable; this capture, in a separate session, did the same thing again — re-fetched by source_sha, ran the clause through `quote_check`, got `grounded: true`. Both sessions found identical wording via identical method, independently. **Still left `status: open`, not closed to `answered`**, because the substance being settled is not the same thing as the vault being clear to act on it: the scheduled audit explicitly escalated rather than discharged, because `70-drafts/filed-under-medical-example/draft.md` (unapproved) builds a paragraph and a voice bracket on this flag staying open, and that escalation is addressed to Cali in `90-feedback/2026-08-14-from-auditor-dawid-skene-previous-performance-quote-recoverable.md`, not yet ruled on. Closing this question or discharging [[claim-dawid-skene-1979-proposed-reliability-weighted-observer-voting-unverified]]'s flag from a promotion pass would pre-empt that ruling. When Cali rules, both the claim-note's flag and this question should close together in one motion — there is nothing left to independently verify.