What happened
Acciona sued the Greater Vancouver Sewerage and Drainage District after the district ended its contract for the North Shore Wastewater Treatment Plant. Under an agreed protocol, TransPerfect Legal ran two TAR 1.0 predictive coding models on about 7 million of Acciona's documents. Dentons lawyers coded 3,812 English and 2,295 Spanish documents to train them. The models reached 75% recall, with precision of 81% in English and 71% in Spanish. The district said that Acciona's production of almost 4 million documents included some 750,000 irrelevant documents, and it asked for an amended list. On 7 July 2025 Justice Elwood dismissed the application. He accepted the evidence that the models were trained to the point of diminishing returns. Based on what Deloitte billed the district for a smaller set, he found that a TAR 2.0 review would have cost Acciona more than $10.6 million. He held that a model's precision rate is not an appropriate measure of compliance. He accepted that the production holds a large number of irrelevant documents. He did not address a late concern about underproduction.
Sources
-
Court document, by authors with no tie to the system.
Supreme Court of British Columbia (Justice Elwood). Acciona Wastewater Solutions LP v. Greater Vancouver Sewerage and Drainage District, 2025 BCSC 1256, Reasons for Judgment, 7 July 2025, paras 13-16, 24-36 and 96-103. BC Courts text, read through the Internet Archive.
In the circumstances, it would be a mistake in my view to reason backwards from the large number of irrelevant documents in the lists of documents to a conclusion that Acciona adopted a dilatory or casual attitude to training the TAR 1.0 models.
Details
AI system
- Workflow
-
The parties agreed on nearly 500 shared search terms and on the use of TAR 1.0 or TAR 2.0. Dentons lawyers coded 3,812 English and 2,295 Spanish documents to train the models. TransPerfect trained the models until recall and precision could no longer be improved without diminishing returns.
Follow-up
- Conditions for replication
-
The document set is large and relatively rich in relevant documents. Lawyers review by hand the documents that TAR cannot process and those flagged as possibly privileged. The parties cooperate to cull irrelevant and duplicate documents further.
Entry details
- Sector
-
Legal services / e-discovery.
- Purpose
-
TransPerfect Legal built the TAR 1.0 models to predict which of Acciona's documents were relevant, after training on documents that Dentons lawyers coded.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). BC court declines to make Acciona redo its TAR 1.0 review (AIBD-0991). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-0991.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)