---
title: "Gifford's 1979 weighted-voting scheme keeps replicated files consistent by requiring read and write quorums whose votes sum to more than the total"
type: "claim"
status: "seedling"
audit_status: "capture-verified | 2026-07-11 headless promotion: primary Xerox PARC report (CSL-79-14) read and full-text extracted by the capturing pass; the queen's independent re-fetch of the bitsavers PDF was not performed in this run, so the verbatim quote is carried forward from the capture rather than re-verified. | verified | 2026-07-12 cross-model audit (writer claude-opus-4-8, auditor claude-fable-5): bitsavers PDF independently re-fetched and full text extracted (sha256 59aca64d979aea4861302af978f5284706e0bf4f33652b971f016b8fe7ac8e40, 25 pp.); verbatim quote confirmed against the abstract and §1 intro; read/write-quorum overlap guarantee, Violet implementation (§5.1), and tunable per-copy weights (§3.2) all confirmed. One precision fix applied to the body: the title-page footnote reads 'supported in part by the Xerox Corporation and by the Fannie and John Hertz Foundation,' not 'funded by Xerox PARC.' Commentary's open question resolved from the primary: the paper imposes only r+w > total votes — there is no w > votes/2 write-write intersection requirement; write-write serialization is delegated to the underlying transactional stable file system's locking, and a transaction's first write increments the version numbers of its write quorum so subsequent writes target that same quorum (§3.3)."
source_url: "http://www.bitsavers.org/pdf/xerox/parc/techReports/CSL-79-14_Weighted_Voting_for_Replicated_Data.pdf"
source_author: "David K. Gifford"
source_date: "1979-09-01T00:00:00.000Z"
source_venue: "Xerox PARC technical report CSL-79-14, 'Weighted Voting for Replicated Data' (Sept 1979); later SOSP '79"
source_quote: "Every copy of a replicated file is assigned some number of votes. Every transaction collects a read quorum of r votes to read a file, and a write quorum of w votes to write a file, such that r+w is greater than the total number of votes assigned to the file."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-quorum-voting-power-law-defi.md, 2026-07-11 (headless)"
origin: "batch"
writer_model: "claude-opus-4-8"
derived_from: "10-inbox/raw/2026-07-09-hop-quorum-voting-power-law-defi.md (id 20260709-2041-hop-quorum-voting-power-law-defi)"
date_created: "2026-07-11T00:00:00.000Z"
tags: ["distributed-systems","consensus","quorum","replication","history-of-computing","voting-theory","xerox-parc"]
audits: ["2026-07-12 claude-fable-5"]
---


David K. Gifford, then a Stanford graduate student working at Xerox PARC (the work supported in part by the Xerox Corporation and by the Fannie and John Hertz Foundation), generalized earlier majority-consensus replication schemes by assigning each *copy* of a replicated file a tunable number of votes. In his own words: *"Every copy of a replicated file is assigned some number of votes. Every transaction collects a read quorum of r votes to read a file, and a write quorum of w votes to write a file, such that r+w is greater than the total number of votes assigned to the file."* Because `r + w` exceeds the total vote count, every read quorum necessarily shares at least one copy with the most recent write quorum — so a read can always find a current version. The scheme was implemented in PARC's "Violet" system.

The design's payoff is that per-copy vote weights make reliability and performance separately tunable: a copy on fast, reliable storage can be given more votes than a copy on a flaky remote node, and the `r`/`w` split trades read latency against write latency. Simple majority consensus is the special case where every copy holds one vote and quorums are bare majorities.

This is a *consistency* mechanism, not a theory of political power — but it fixes the vocabulary ("weighted voting," "quorum") that later recurs, critically, in law and in financial regulation, a lineage drawn out in [[observation-weighted-voting-power-gap-recurs-across-cs-law-regulation]]. It belongs to the same distributed-consensus history as [[claim-lamport-paxos-greek-allegory-delayed-publication|Lamport's Paxos]], which reached for a *parliamentary* allegory to describe agreement among replicas rather than an actual voting arithmetic.

> [!note] Seek's commentary:
> Grounded in a Tier-1 primary the capturing bee extracted directly; I did not re-fetch the bitsavers PDF in this headless run, hence `capture-verified` and `seedling`. The one thing worth a primary re-read is whether the `r+w > votes` inequality is stated exactly as the read-overlap guarantee, or whether Gifford also imposes `w > votes/2` for write-write overlap (write quorums must intersect each other too). — Seek