Da Silva Moore v. Publicis Groupe: the first federal judicial opinion to approve predictive coding (computer-assisted review) for e-discovery (S.D.N.Y. 2012)

The holding and the scale figures are bound to the primary opinion (287 F.R.D. 182) via the CourtListener full-text index; the “acceptable way to search for relevant ESI” line, which is OCR-corrupted in the primary’s own digitized copy, is bound to Judge Peck’s own later opinions that reproduce and attribute it to this case.

The problem

By 2012, large civil cases routinely turned on electronically stored information (ESI) at a scale that made page-by-page human review ruinously slow and expensive, and courts had not yet blessed a machine-learning alternative (source). The dispute arose inside Da Silva Moore v. Publicis Groupe & MSL Group, a Title VII gender-discrimination suit in which the producing party faced over three million documents requiring review from agreed-upon custodians (source). Corporate Counsel Business Journal’s contemporaneous analysis describes the same volume as “over three million emails requiring review and filtering” (source), and the phrase “over three million emails” returns the primary opinion as a unique quoted-phrase hit in the CourtListener full-text index (source).

What was built

The “build” here is a federal court opinion, not a software deployment: Magistrate Judge Andrew J. Peck of the Southern District of New York issued the ruling in Da Silva Moore v. Publicis Groupe & MSL Group, No. 11 Civ. 1279 (ALC)(AJP), dated February 24, 2012 (source). The technology it approved is “computer-assisted review,” also called predictive coding: a supervised machine-learning classifier that a human trains on a reviewed subset of documents and then applies to the rest (source). Under the parties’ ESI protocol, they agreed to a 95% confidence level to draw a random sample and selected “2,399 documents” for manual review to establish a “seed set” to train the software (source); the “2,399 documents” figure returns the primary opinion as a unique quoted-phrase hit in the CourtListener full-text index (source).

The outcome

The opinion’s own text holds that “computer-assisted review is an available tool and should be seriously considered for use in large-data-volume cases,” a line that returns the primary opinion as a unique quoted-phrase hit in the CourtListener full-text index (source), and which Proskauer independently reproduces from the ruling (source). The opinion also frames computer-assisted review as “an acceptable way to search for relevant ESI in appropriate cases,” a line reproduced verbatim by eDiscovery Today (source) and quoted, with attribution to this decision, in Judge Peck’s own later opinion Rio Tinto Plc v. Vale S.A. (source) and in Dynamo Holdings Ltd. Partnership v. Commissioner (source). Reputable commentators describe the ruling as the first of its kind: Corporate Counsel Business Journal wrote that it was “the very first time, a court has considered, and approved, the use of advanced document review software” (source), and Proskauer titled its analysis “Court Endorses Use of Predictive Coding for ESI Discovery” (source); Peck’s Rio Tinto opinion, subtitled “Da Silva Moore Revisited,” recounts the same “first” framing (source). CCBJournal reproduces the court’s cost rationale that “[l]inear manual review is simply too expensive” (source).

The load-bearing caveat, stated plainly: the weakest link is the “acceptable way to search for relevant ESI” line, because the primary opinion’s own digitized copy is OCR-corrupted for that sentence and does not return it, so that specific quote rests on Judge Peck’s later opinions reproducing it and on the Tier-2 eDiscovery Today retrospective, not on a unique hit in the primary itself (source). The 95%-confidence and seed-set framing likewise leans on Proskauer’s Tier-2 reading of the parties’ ESI protocol rather than a unique primary hit (source). The holding and the scale figures (“over three million emails,” “2,399 documents”) are each bound to the primary as unique CourtListener hits (source).

How this was verified

Method: primary binding via the CourtListener full-text index, with independent Tier-2 corroboration. On 2026-09-12, the holding “computer-assisted review is an available tool and should be seriously considered for use in large-data-volume cases” and the figures “over three million emails” and “2,399 documents” were each bound to the primary opinion (287 F.R.D. 182, CourtListener opinion id 8766706): every one returns the primary as a UNIQUE quoted-phrase hit in the reachable CourtListener REST search index, with the raw JSON captured to sources/. The line “an acceptable way to search for relevant ESI in appropriate cases” is OCR-corrupted in the primary’s own digitized copy (returns 0), so it is bound instead to two of Judge Peck’s own later opinions that reproduce and attribute it to this case — Rio Tinto Plc v. Vale S.A. (306 F.R.D. 125) and Dynamo Holdings Ltd. Partnership v. Commissioner (143 T.C. 183) — and corroborated by eDiscovery Today’s 2022 retrospective. All figures are further corroborated by Proskauer Rose LLP’s contemporaneous analysis and Corporate Counsel Business Journal. CourtListener’s object-detail and cluster endpoints returned HTTP 401, and Justia, Leagle, casetext, Google Scholar and web.archive.org were blocked from this headless environment, so the reachable search index served as the primary capture. No party or vendor was contacted; a court opinion has no subject to confirm it.

Path to green

The critical holding, the “first” framing, and the scale figures are now bound to the Tier-1 primary (via CourtListener) and to Judge Peck’s own citing opinions, so the sourcing gap that held round 1 below green is closed. The one detail still absent from every reachable source is the name of the predictive-coding software vendor used under the parties’ ESI protocol; it is not a critical claim in this story, but the full opinion text behind CourtListener’s authenticated object-detail endpoint (HTTP 401 this session), a govinfo/PACER copy, or the Federal Rules Decisions reporter would identify it.

Related case files

• The FTC’s action against the “robot lawyer” DoNotPay, an AI legal tool that overpromised — the enforcement counterpoint to a court blessing responsible AI-assisted legal work.
• Ayinde v. Haringey: a UK court imposes wasted costs for AI-fabricated citations — the hallucination line that shows why courts distinguish disciplined AI-assisted review from unchecked generative output.
• Barber v. Morawa: the first published opinion sanctioning AI-fabricated citations — a “first” on the misuse side, bookending Da Silva Moore’s “first” on the approval side.

Sources

1. eDiscovery Today · “The Da Silva Moore Case Ten Years Later” · 2022-02-23 · https://ediscoverytoday.com/2022/02/23/the-da-silva-moore-case-ten-years-later-ediscovery-case-law/ — Tier 2 (independent e-discovery case-law commentary; reproduces the opinion verbatim)
2. Proskauer Rose LLP · “Court Endorses Use of Predictive Coding for ESI Discovery” · 2012 · https://www.proskauer.com/blog/court-endorses-use-of-predictive-coding-for-esi-discovery — Tier 2 (independent law-firm analysis; quotes the opinion and the ESI protocol)
3. Corporate Counsel Business Journal · “Federal Court Approves The Use Of ‘Predictive Coding’ Technology-Assisted Document Review” · 2012 · https://ccbjournal.com/articles/federal-court-approves-use-predictive-coding-technology-assisted-document-review — Tier 2 (independent legal-industry analysis; quotes the opinion and the document volume)
4. Da Silva Moore v. Publicis Groupe & MSL Group, 287 F.R.D. 182 (S.D.N.Y. Feb. 24, 2012), No. 11 Civ. 1279 (ALC)(AJP) · CourtListener opinion id 8766706 · https://www.courtlistener.com/opinion/8782929/moore-v-publicis-groupe/ — Tier 1 (the court opinion itself; bound via the CourtListener full-text search index, where the holding, “over three million emails” and “2,399 documents” each return this opinion as a unique quoted-phrase hit — raw JSON in sources/)
5. Rio Tinto Plc v. Vale S.A., 306 F.R.D. 125 (S.D.N.Y. 2015) (Peck, M.J.), “Da Silva Moore Revisited” · https://www.courtlistener.com/opinion/8788861/rio-tinto-plc-v-vale-sa/ — Tier 1 (Judge Peck’s own later opinion reproducing “an acceptable way to search for relevant ESI in appropriate cases” and recounting the “first” framing, attributed to this decision)
6. Dynamo Holdings Ltd. Partnership v. Commissioner, 143 T.C. 183 (2014) · https://www.courtlistener.com/opinion/4562565/dynamo-holdings-ltd-partnership-v-commissioner/ — Tier 1 (published tax-court opinion reproducing the “acceptable way to search for relevant ESI” line)