A. Introduction

The parties have filed a Joint Letter raising disputes regarding validation of Uber’s Technology Assisted Review (TAR) model for identifying relevant and responsive documents for production. Specifically, the parties dispute the relevance of certain categories of documents. Attorneys for both sides reviewed a sample of documents to compare their views on those to their opponents’ views and the results of the TAR model. The parties agreed that a significant number of relevant and responsive documents had not been identified as such by the TAR model, which Uber attributes to changing parameters for production over the course of the case based the parties’ negotiations. See Dkt. No. 2166 at 6. The parties disagreed as to a relatively small number of documents (125 documents from a sample of 4,500) falling within the following three categories: (1) documents related to non-sexual “physical and verbal assaults” by drives against passengers; (2) documents related to communications with government agencies; and (3) documents related to individual drivers’ background checks. At the Court’s request, each side selected one document in each category to file under seal as examples. See Dkt. Nos. 2172 (text-only Order), 2190 (response), 2193 (sealed exhibits). The Court addresses those categories, and next steps for production, as follows.

B. Non-Sexual Assault

Plaintiffs seek documents related to non-sexual physical and verbal assaults committed by drivers against passengers, including custodial records and non-custodial JIRA reports. Uber opposes the request, contending that these records are irrelevant and not within the scope of Plaintiffs’ document requests.

Though the Court previously limited Plaintiffs’ discovery of safety incidents that did not relate to sexual assault or sexual misconduct, Dkt. No. 683 at 9, the Court agrees with Plaintiffs’ current relevance argument to an extent. Uber’s own documents generally support Plaintiffs’ contention that the company has considered data regarding prior non-sexual behaviors and incidents to predict the relative safety of a driver compared to other drivers. See Dkt. No. 2199-3 at 32 (seeking to develop metrics to predict the relative safety of a driver and reduce safety risks), 34 (discussing efforts to predict driver safety incidents as part of an effort to prevent such incidents); Dkt. No. 2199-5 at 3-4 (discussing the percentage of drivers deactivated for sexual assault who had prior safety incidents that were not limited to incidents of a sexual nature, and included “verbal disputes,” “discriminatory remarks,” “threats of violence,” and other behavior).

However, Plaintiffs’ request is exceptionally broad, encompassing every “[d]ocument related to physical and verbal assaults of riders by drivers.” Dkt. No. 2166 at 4. This request for discovery relief goes beyond the requests for production on which Plaintiffs rely, which focus on documents related to vetting policies and processes. The undersigned disagrees with Plaintiffs that Judge Breyer’s Pretrial Order No. 2 established the general relevance of every custodial and non-custodial document related to the physical or verbal assault of a passenger by a driver. Dkt No. 65. Pretrial Order No. 2 directed that Uber’s custodians must be broadly identified and not limited to those who worked on sexual assault or sexual misconduct issues, and must include those “who evaluate and analyze passenger complaints about drivers” and “develop policies for vetting drivers or policies responding to passenger complaints about drivers.” Id. at 2–3. Furthermore, given the time period that this litigation covers and the volume of rides Uber facilitates and the number of complaint, a request for every record of non-sexual assault is likely disproportionate to the needs of this case. See Dkt. No. 684 at 1–2 (noting that Uber reviewed more than 800,000 user incident reports from 2017 to 2020 and during that period, facilitated, on average, 3 million trips a day).

Accordingly, Plaintiffs are not entitled to discovery of every JIRA or incident ticket that relates to a physical or verbal assault. Uber, however, must produce documents that show that Uber has considered data and information regarding prior non-sexual behaviors and incidents to determine whether such behaviors predict the relative safety of a driver compared to other drivers. Such information relates to Uber’s vetting and approval of drivers, its criteria for driver eligibility, and potential changes to that criteria, as called for by Plaintiffs’ document requests. Because such documents are subject to production, they should be included in training the TAR model. At least some such documents have already been produced, as evidenced by documents Plaintiffs cited from outside the TAR validation sample, see Dkt. No. 2166 at 5 nn.5&6 (citing documents later filed under seal at Dkt. No. 2199), and Uber asserts in a supplemental brief that it is “already producing documents involving discussions about these types of incidents, e.g., whether to deactivate a driver who has reports of verbal altercations.” Dkt. No. 2202 at 4. To the extent there is any question of whether such documents are responsive, the Court now holds that they are. But Uber need not produce documents concerning its responses to individual incidents or allegations of non-sexual assault or misconduct, unless those documents include discussion of Uber’s potential or proposed changes to the criteria or eligibility requirements or its vetting or approval process for drivers, or its broader efforts to analyze and predict driver safety.

C. Communications with Government Agencies

The parties dispute the extent to which documents in this category relate to the protected right to petition government—or in other words, lobbying. Plaintiffs contend that the documents at issue “do not include files belonging to, or communications with, registered lobbyists.” Dkt. No. 2166 at 5. Uber, on the other hand, asserts that Plaintiffs seek “discussions with legislative members and designated lobbyists . . . regarding positions for pending bills, e-mails to councilmembers seeking support of airport rideshare services, and discussions a D.C. councilmember about pending legislation.” Id. at 7–8. Uber notes that this Court previously “agree[d] with Judge Schulman’s decision [in the JCCP] that lobbying activity is not relevant.” Id. at 8 (quoting Dkt. No. 1698 at 25).

The requests for production on which Plaintiffs rely are as follows:

REQUEST FOR PRODUCTION NO. 112: Any and all DOCUMENTS REGARDING YOUR state or local government lobbying efforts or activities, including but not limited to communications with any lobbyist, public policy entity, trade association, or representatives of state, local, or federal governmental entities.

Dkt. No. 2190-1 at 9.

REQUEST FOR PRODUCTION NO. 121: Any and all DOCUMENTS RELATING TO YOUR policies, procedures, practices or strategies REGARDING governmental rules or regulations REGARDING Transportation Network Companies (“TNC’s”), livery companies, taxis, and/or common carriers.

Id. at 11.

In context, the Court’s previous Order agreeing with Judge Schulman on the general lack of relevance of lobbying activity was focused on the issue of custodian designation, where Uber disputed the inclusion of certain custodians related to lobbying work. See Dkt. No. 1698 at 24–27. The Court acknowledged—and did not reject—Plaintiffs’ position that “Uber’s lobbying efforts [might] demonstrate a level of culpability to support [Plaintiffs’] punitive damages claim,” but the Court considered the inclusion of the custodians at issue unnecessary to serve that purpose. Id. at 25. Although the Court continues to believe that lobbying documents are of at most limited relevance to this case, the Noerr-Pennington doctrine protecting such activity is not a categorical “bar to discovery of evidence.” Wal-Mart Stores, Inc. v. City of Turlock, No. CIV F 04-5278 OWW DLB, 2005 WL 8176355, at *3 (E.D. Cal. Jan. 28, 2005), modified on other grounds, 2005 WL 8176346 (E.D. Cal. June 2, 2005). Nor is it “a complete bar to admissibility,” and even if it were, evidence need not be admissible to be discoverable. See Nat.-Immunogenics Corp. v. Newport Trial Grp., No. SACV 15-02034 JVS(JCGx), 2018 WL 6137597, at *3 (C.D. Cal. May 16, 2018), modified on other grounds, 2019 WL 3110021 (C.D. Cal. Mar. 19, 2019).

Moreover, unlike the custodians to whom Uber objected, Uber agreed to produce any documents responsive to Request No. 112 (documents regarding “state or local government lobbying efforts or activities”) subject to limitations of time, geography, search terms, and custodians, and did not raise any objections based on the First Amendment or the NoerrPennington doctrine in its discovery response. Dkt. No. 2190-1 at 9–10. The previous Order specifically addressing a dispute over custodians was not license for Uber to withhold documents unilaterally that it had previously agreed to produce, to the extent those documents might be found in the files of the custodians ultimately selected by the parties and the Court.

Plaintiffs’ example from this category is an email chain discussing Uber’s response to a proposed local regulation requiring driver background checks. Dkt. No. 2198-4 at 3–9. That document is at least potentially relevant to this litigation for reasons other than seeking to impose liability for lobbying efforts, because it relates to the safety measures available to Uber and Uber’s ability or willingness to implement them, and the Court would order its production even if Uber had not previously agreed to produce documents related to lobbying. Although Uber’s chosen example, id. at 13–17 (a discussion related to fare rules), is less clearly relevant, it falls squarely within the documents that Uber previous agreed to produce.

Uber must produce documents responsive to Request No. 112 because it previously agreed to do so, and never amended that response or sought relief from it. Both of the examples selected by the parties are responsive to that request. For Request No. 121, where Uber’s response indicated only that it would meet and confer, Uber must produce documents related to “policies, procedures, practices or strategies regarding governmental rules or regulations” to the extent they relate to safety or driver qualifications. See Dkt. No. 2190-1 at 11 (capitalization altered).

D. Individual Drivers’ Background Checks

The final category at issue relates to individual drivers’ background checks. Plaintiffs contend that these documents are responsive to requests for documents related to background checks, driver screening processes, and drivers who did not meet Uber’s criteria. Uber argues that information about individual drivers who are not accused of misconduct in this litigation is not relevant or proportional to the needs of the case. Both of the example documents are notices under the Fair Chance Act to drivers’ whose conditional offers of engagement were revoked after a background check was completed. Dkt. No. 2198-5. In the Court’s view, such notices (and similar documents related to driver background checks) that reveal past instances of sexual assault or misconduct are sufficiently relevant to an understanding of how Uber handles such issues to warrant production even if the driver at issue is not accused of misconduct in this litigation, while notices that do not relate to such conduct (like Uber’s example, which relates to theft and driving without a valid license) need not be produced. As Plaintiffs acknowledge, documents related to background checks for drivers who drove only for Uber Eats need not be produced. See Dkt. No. 2166 at 6 n.9. Plaintiffs have not explained why notices under the Fair Chance Act or similar documents related to individual drivers’ background checks that reveal past instances of physical assault or other non-sexual violence are relevant and important to the issues involved in this MDL. Therefore, Uber need not produce such documents.

E. Next Steps

It is not entirely clear whether Uber disputes the need to retrain the TAR model and produce additional documents based on the parties’ validation review and negotiations. Uber’s arguments that much of the discrepancy between the TAR model and Uber’s own reviewers’ characterization of documents stemmed from agreed changes to the scope of production in December indicate that retraining and additional production is necessary, although it is not clear whether or to what extent that has already occurred. Even if that retraining has already been completed, the Court’s rulings here include meaningful differences from Uber’s view of the scope of responsive documents. A TAR model—or, for that matter, any other method of large-scale review—will likely never produce perfect results, but a model trained on an incorrect scope of production is inherently not a reliable method of identifying responsive documents. Further retraining is therefore necessary, unless the parties identify an alternative method to gather and produce the types of documents addressed by this Order (as well as any documents falling within the scope of any concessions that Uber might have made during the meet-and-confer process).

Accordingly, Uber is ORDERED to produce documents consistent with this Order, and the parties are ORDERED to meet and confer and file either a stipulation or joint letter addressing a process and schedule for such production. Because the issues to be resolved may be complex, the Court sets the deadline for such filing as two weeks from the date of this Order, although parties are encouraged to file sooner if they are able to do so. If the parties are unable to reach a stipulation and instead file a joint letter, it should address only the process and schedule for production; the Court is not inclined to reconsider the need for Uber to identify and produce the documents at issue. 

IT IS SO ORDERED.