Winning or Defeating Class Certification: A Practical Guide

Document type: Guide Practice area: Litigation — Class Actions Jurisdiction: United States Last reviewed: 5 September 2026

The organizing question

Everything in a class action reduces to one question, and both sides should write it on the wall on day one:

Can the plaintiff prove that every class member was injured, using evidence common to the class?

Not whether the defendant did something wrong. Not whether the plaintiff was harmed. Whether injury across the class can be shown with common proof. Every decision — the theory pleaded, the class defined, the discovery sought, the expert retained, the model built — should be made by reference to that question.

Plaintiffs who plead broadly and build the model afterward lose. Defendants who spend the certification briefing arguing the merits lose. The side that understood the question first usually wins.

Step one: frame the theory to the evidence

For plaintiffs. Before the complaint is finalized, know what data exists and what it can show.

Does the defendant maintain transaction-level records covering the class period? If not, a regression-based overcharge model may be impossible.
Is the alleged conduct uniform across the class, or did it vary by product, region, channel, or time? Variation is where predominance dies.
Is there a single misrepresentation on a single label, or many statements across many media? Uniform exposure is far easier to prove.
What is the mechanism by which every member was injured, stated in one sentence? If it takes a paragraph, there is a problem.

Then plead the theory the model can measure. Comcast requires the damages model to measure damages attributable to the certified theory. A complaint with five theories and one aggregate model is a certification denial in waiting.

For defendants. From the first case management conference, build toward the certification record.

Identify the sources of variation across the putative class and preserve the evidence of them.
Identify categories of putative members who plainly suffered no injury and the data that shows it.
Consider early whether a narrower class would be certifiable, because the plaintiff will eventually narrow, and the defence position is stronger if the narrowing is forced rather than volunteered.

Step two: get the class definition right

The class definition determines the certification fight more than any other single choice.

For plaintiffs, the tension is real. A broad definition maximizes settlement value and invites the uninjured-member attack. A narrow definition is more certifiable and worth less. Counsel who refuse to narrow frequently end with nothing.

Practical rules:

Define by objective criteria. Membership must be determinable without an individualized inquiry into the merits. "All persons who purchased Product X between [dates]" works. "All persons who were misled by the label" does not — it is a fail-safe class, defining membership by the outcome of the case, and courts reject it.
Exclude the categories that cannot be proved. Purchasers under negotiated contracts with distinct pricing; channels where the label was not visible; periods where the evidence is weak; entities with their own claims.
Watch ascertainability. Circuits differ on whether an administratively feasible method of identifying members is required. In circuits that require it, a consumer class for a low-value product bought in cash presents a real problem, and the answer usually involves records the defendant or retailers hold.
Consider subclasses where the class contains groups with materially different positions, and consider separate representation where their interests conflict.

For defendants, the definition is the first target. Attack fail-safe definitions, definitions that sweep in obviously uninjured members, and definitions whose membership cannot be determined without mini-trials.

Step three: plan certification discovery

Certification discovery is a distinct project with its own scope, and both sides should propose a plan rather than letting it merge into merits discovery.

What plaintiffs need:

Transaction data covering the class period, at the finest available granularity, with the field definitions and a data dictionary. Negotiate format, completeness, and a process for resolving anomalies. This will take longer than anyone expects.
Documents establishing uniformity: pricing policies, labelling decisions, training materials, standard contracts, the mechanism of the alleged conduct.
Depositions of the custodians of the data and of the people who made the decisions alleged to be uniform.

What defendants need:

The named plaintiffs' individual circumstances, in detail. Typicality and adequacy attacks start here, and a named plaintiff whose purchase history, sophistication, or motivation is atypical is a real problem for the class.
Evidence of variation: differing negotiations, differing exposure, differing knowledge, differing outcomes.
The plaintiffs' expert's materials — data, code, workpapers, alternative specifications run and discarded. Ask for the code specifically; a model that cannot be replicated is a model that can be attacked.

Sequencing. Bifurcated certification discovery sounds efficient and often is not, because the merits and certification records overlap heavily. What does work is a defined completion date for the data production, so that the expert work can begin on a known timetable.

Step four: select and manage the expert

Selection criteria that matter:

Subject-matter fit. An economist who has worked in the relevant industry brings credibility a generalist cannot.
Data capability. These models involve large datasets; the expert's team must be able to handle them.
Deposition and hearing performance. Certification hearings increasingly involve live testimony under judicial questioning. Ask for transcripts.
Prior exclusions. Check whether the expert has been excluded, and on what grounds. It will be raised.
Prior inconsistent positions. Opposing counsel will find the report where the expert took the opposite view.

Management:

Engage before the theory is fixed, so that the model and the pleading are built together.
Give the expert the data problems rather than a clean summary.
Stress-test before serving. Run the model under alternative specifications, on subsets, and on periods when no conduct is alleged. If it breaks, better to know now.
Require member-level output. The court will likely ask how many members show no injury. An expert who cannot produce that number has handed the defence its best argument.
Manage the draft file. Draft reports and most counsel-expert communications are protected, but facts and data provided by counsel and assumptions the expert relied on are not.
Consider a consulting expert to stress-test without creating discoverable material.

Step five: build the plaintiff's showing

The certification motion should be organized as a proof, not an argument.

Numerosity. Rarely contested; establish it with data.

Commonality. After Wal-Mart, identify the common answer, not the common question. What single determination will resolve an issue central to every member's claim in one stroke?

Typicality and adequacy. Address the named plaintiffs' circumstances head-on. Disclose the problems rather than letting the defence find them. Confirm counsel's qualifications and the absence of conflicts, including any intra-class conflicts.

Predominance. This is the motion. Organize it as: (1) the elements of the claim; (2) for each element, the evidence that will prove it, and whether that evidence is common; (3) the mechanism of class-wide injury; (4) the damages model, its methodology, its results, and its fit to the certified theory; and (5) the individualized issues, why they are subordinate, and how they will be managed.

Superiority and manageability. A trial plan. Courts increasingly expect one, and drafting it forces clarity about what the trial would actually look like.

Anticipate. Address the uninjured-member argument in the opening brief with the expert's member-level results, rather than waiting for the reply.

Step six: build the defendant's attack

Organize the opposition around the model, not the merits.

1. Member-level results. Obtain the model's fitted values for each class member. Count the members showing no injury or negative results. Quantify. This is the single most effective argument available, and it must come from an expert who has run the numbers.

2. Specification. Omitted variables, functional form, the choice of controls, and the sensitivity of the result to reasonable alternatives. If the overcharge estimate swings from 9% to 1% under a defensible alternative specification, say so.

3. Benchmark integrity. Is the "clean" period actually clean? Contamination inflates the but-for price and can invalidate the whole model.

4. Aggregation. Run the model separately by product, customer type, channel, and period. Heterogeneous results show that the aggregate masks segments with no injury.

5. False positives. Apply the model to a period when no conduct is alleged. An "overcharge" in that period is powerful.

6. Fit to the theory. Comcast: does the model measure damages attributable to the certified theory, or something broader?

7. Individual issues. Variation in negotiation, exposure, knowledge, reliance, and mitigation.

8. Named plaintiff problems. Atypicality, inadequacy, standing, arbitration agreements, releases, limitations.

9. Class definition defects. Fail-safe definitions, ascertainability, over-inclusiveness.

10. Article III. After TransUnion, every member must have standing to recover damages. Press the question of when that must be shown.

Step seven: the Rule 702 motion

File it. A defendant that does not move to exclude has conceded the model's reliability and weakened its appellate position.

Time it with the certification opposition, not afterward. The court should decide admissibility before or concurrently with certification, because the certification decision depends on it.

Frame it as a gatekeeping question, not a battle of experts. The argument is not that the defence expert is right; it is that the plaintiff's expert's methodology is unreliable or unreliably applied, and that the 2023 amendments to Rule 702 require the proponent to establish otherwise by a preponderance.

The strongest grounds, in order:

The model does not measure what the theory requires (Comcast).
The methodology fails on its own terms — false positives, results driven by an arbitrary specification choice, an analytical gap between the data and the conclusion.
The model was not reliably applied to the facts — data errors, inappropriate benchmark, a population that does not match the class.
The expert cannot replicate or explain the analysis.

What rarely succeeds: attacks on qualifications where the expert is credentialed; disagreements about which of two defensible specifications is better; and arguments that go to weight rather than admissibility. Courts say so, repeatedly.

For plaintiffs opposing: distinguish weight from admissibility relentlessly; show that the methodology is standard in the field and peer-reviewed; and have the expert address each criticism substantively rather than dismissing it.

Step eight: the hearing

Certification hearings vary from argument on the papers to multi-day evidentiary proceedings. Prepare for the latter.

Live testimony. Increasingly common where the experts disagree. Prepare the expert for judicial questioning, which is different from cross-examination: judges ask about the assumption underlying the model, why a particular control was chosen, and what happens if it is removed.

Concurrent evidence. Some judges put both experts on the stand together and have them discuss the disagreement. This rewards the expert who genuinely understands the other side's position and punishes the one who has been coached to a script.

Demonstratives. A regression is hard to explain and easy to misexplain. Simple, honest demonstratives — a scatter with the fitted line, a bar chart of results by segment, a distribution of member-level estimates — do more than any amount of testimony.

Ask for findings. Both sides should request that the court resolve the specific expert disputes and make findings. A certification order without findings on a contested methodological dispute is vulnerable on appeal, and both sides benefit from knowing where they stand.

Step nine: Rule 23(f)

Fourteen days, from entry of the order. In most circuits this is not extended by a motion for reconsideration. Calendar it the day the order issues.

What gets granted: an unsettled legal question of general importance; a death-knell situation where denial ends the case or grant creates irresistible settlement pressure; or manifest error.

How to frame it. As legal error — the court applied the wrong standard, or failed to resolve a factual dispute it was required to resolve — rather than as disagreement with the weighing. Abuse-of-discretion review is deferential; de novo review of a legal question is not.

What the record must show. The expert evidence, the objections, the Rule 702 motion, and the findings. A defendant that never moved to exclude cannot complain on appeal that unreliable evidence was credited.

Step ten: settlement classes

Where the parties settle, certification is still required, and Amchem holds that Rule 23 applies with full force except for manageability concerns that disappear because there will be no trial.

What courts scrutinize:

Intra-class conflicts. Present versus future claimants, different injury types, strong versus weak claims. Structural protections — subclasses with separate counsel — are required where the conflicts are real.
Adequacy of representation, given those conflicts.
The allocation plan, and whether it treats similarly situated members similarly.
The relationship between the recovery and the strength of the claims.
Attorneys' fees, their relationship to the actual benefit conferred, and the treatment of claims-made structures where the realistic take-up rate is low.
Clear sailing and reversion provisions, which attract particular scrutiny.
Notice, and whether it is the best notice practicable.

Practical points. Build the settlement class definition with the same discipline as a litigation class. Do not assume that settlement forgives a definitional problem — courts have refused approval on exactly that ground. And be candid in the approval papers about the weaknesses of the case; a court asked to approve a modest settlement is far more receptive when told why the recovery is modest.

Named plaintiffs: selection, vetting, and the problems they create

The named plaintiffs are the case's most easily attacked component, and both sides underestimate how much turns on them.

For plaintiffs, vet before filing:

Standing. Did they actually purchase the product, in the class period, in a covered channel? Obtain the receipts, the account records, the credit card statements. A named plaintiff who cannot document the purchase is a dismissal motion.
Typicality. Are their circumstances representative? A named plaintiff who is a sophisticated repeat purchaser, an industry insider, a former employee, or a serial class representative creates problems.
Arbitration and class waivers. Did they agree to terms containing an arbitration clause with a class waiver? This is the first thing the defence checks and it has ended more cases than any merits argument.
Releases. Prior settlements, warranty claims, or account agreements.
Limitations. Are their claims timely, and does any tolling doctrine apply?
Credibility. They will be deposed. Prior litigation, bankruptcies, criminal history, and social media are all fair game.
Availability and commitment. A named plaintiff who withdraws mid-case creates delay and an adequacy argument.
Relationship to counsel. A named plaintiff who is counsel's relative, employee, or long-standing client invites an adequacy attack.

File with more than one, in more than one relevant subgroup, so that the loss of a representative does not end the case.

For defendants, named plaintiff discovery is early, cheap, and high-yield:

Full purchase history and account records.
Deposition covering what they saw, what they relied on, what they knew, and how they came to counsel.
Prior litigation history and any arbitration agreements.
Their understanding of their role and their engagement with the case — an adequacy attack succeeds where a representative cannot describe the claim.

A caution for defendants. Attacking named plaintiffs individually is productive up to a point and counterproductive past it. A court that perceives the defence as trying to pick off representatives on technicalities may permit substitution. And knocking out a weak representative sometimes clears the way for a stronger one. Assess whether the attack ends the case or merely improves the plaintiff's position.

Managing the data

In any case involving a quantitative model, the transaction data is the foundation and the schedule risk.

For the producing party. Determine early what exists, in what systems, for what periods, and in what condition. Legacy systems, migrations, and acquisitions all create gaps. Produce with a data dictionary — field names, definitions, permissible values, units, and the meaning of nulls. A production without one generates months of correspondence.

For the receiving party. Load and validate the data before filing anything that depends on it. Check totals against public financials, look for gaps, duplicates, impossible values, and units that change mid-period. Raise anomalies quickly and specifically.

Agree a protocol. Format, delivery method, a process for resolving anomalies, and a named technical contact on each side who can speak to an engineer rather than a lawyer. An hour on a call between the two data teams saves six weeks of letters.

Protective order and expert access. Highly confidential transaction data will require an attorneys'-eyes-only designation and an expert acknowledgment process. Build in enough time; getting an expert's team cleared can take weeks.

Preservation of the working files. Both sides' experts should preserve their code, intermediate datasets, and specification runs. Opposing counsel will ask, courts increasingly require production of code, and an expert who cannot reproduce their own result is finished.

The trial plan

Courts increasingly require a trial plan as part of the superiority and manageability showing, and drafting one is useful even where it is not required.

What it contains:

The claims and elements to be tried on a class basis.
The common evidence that will prove each element — documents, testimony, expert opinion.
The order of proof, and the anticipated length.
The individualized issues, if any, and how they will be handled: bifurcation, a claims administration process, special masters, sub-trials, or aggregate damages with an allocation plan.
The defendant's defences and how it will be permitted to raise them — this addresses the Wal-Mart "trial by formula" objection directly.
The verdict form, in draft. This is the item that most clarifies whether the case is actually triable.
Damages: how the aggregate is determined and how it is distributed.

Why it helps the plaintiff. It converts an abstract manageability argument into a concrete plan a judge can evaluate, and it demonstrates that counsel has thought past certification.

Why it helps the defendant to demand one. A plaintiff who cannot produce a coherent trial plan has revealed that the case is not triable, and a verdict form that requires the jury to make findings the common evidence cannot support is the clearest possible demonstration of individualized issues.

Worked example: the decision that decided the case

Ilya Zabrodin filed a consumer class action alleging that a supplement's labelled potency was overstated. The complaint defined the class as all purchasers nationwide over six years, across eleven product variants sold through retail, online, and subscription channels.

The problem, visible from month one. The label language changed twice during the period. Three of the eleven variants were never tested by the plaintiff's laboratory. Subscription purchasers received the product without seeing the retail packaging. And roughly a third of retail sales were at promotional prices set by retailers rather than by the manufacturer.

The defence expert's report quantified all four issues and concluded that a class-wide price premium could not be estimated for at least 40% of the class.

Ilya's choice. He could defend the broad class and probably lose, or narrow. He narrowed: four variants rather than eleven, retail and online but not subscription, and a class period of twenty-two months during which the label language was constant and the tested variants were the ones sold.

The narrowed class was roughly 28% of the original by revenue.

The result. Certification granted. The defence's uninjured-member argument, which had been overwhelming against the broad class, had nothing to work with against the narrow one — every member in the narrowed class bought a tested variant, saw the same label, and paid a price the manufacturer set.

The case settled eight months later for an amount that, per class member, was several times what the broad class would plausibly have recovered.

The lesson, which generalizes: the class definition is not a maximization problem. It is a matching problem — matching the class to the evidence that actually exists. Counsel who understand that early win cases that counsel who do not lose entirely.

Notice, opt-outs, and what happens after certification

Certification is not the end of the procedural work, and the post-certification steps affect value materially.

Notice. Rule 23(c)(2)(B) requires the best notice practicable under the circumstances, including individual notice to all members who can be identified through reasonable effort. In practice:

Direct notice by email or mail where records permit, supplemented by publication and digital campaigns.
A notice plan prepared by a notice expert, with a projected reach percentage — courts look for 70–80% or better.
Plain language. Courts have become considerably less tolerant of dense legalistic notices.
A functioning claims website and telephone line, with the notice, the pleadings, and the key orders available.
Cost allocation. The class representative bears notice costs in a litigation class, which is a real constraint; in a settlement class the fund usually pays.

Opt-outs. For a (b)(3) class, members must be given the opportunity to exclude themselves.

For defendants, high opt-out rates among large purchasers are the beginning of a second front — individual actions by the most valuable claimants, often with better counsel and no class-wide discount. Some settlements include a blow-up provision permitting the defendant to withdraw if opt-outs exceed a threshold.
For plaintiffs, opt-outs reduce the class recovery and complicate allocation.
Mass opt-out solicitation by other firms is now routine in large commercial classes and should be anticipated.

Decertification. A certification order is not final and may be revisited as the record develops. A defendant that loses certification should preserve the arguments and renew them if discovery changes the picture — a model that appeared workable at certification sometimes does not survive merits discovery.

Interlocutory settlement dynamics. After certification, exposure is aggregate and settlement pressure is intense. Defendants should think about this before certification, not after, and should be honest internally about whether the case will be tried. A defendant that will never try the case should have been evaluating settlement while it still had leverage.

Where cases are actually lost

Reviewing certification denials and reversals, the same failures recur.

Plaintiff-side:

A class defined more broadly than the evidence. The single most common cause.
A damages model built for a theory broader than the one certified. Comcast, straightforwardly.
An expert who cannot produce member-level results, or who resists doing so.
No answer to the uninjured-member argument until the reply brief.
A named plaintiff with an arbitration agreement nobody checked before filing.
A fail-safe class definition.
No trial plan, and no coherent answer when asked what the trial would look like.

Defence-side:

Arguing the merits instead of the model. Judges are not deciding whether the defendant did it.
Not moving to exclude, and then complaining on appeal that the evidence was unreliable.
Asserting that uninjured members exist without quantifying them. An expert who says "some" loses to one who says "22%."
Attacking named plaintiffs so aggressively that the court permits substitution and the plaintiffs end up with better representatives.
Failing to make a record, so that the appellate court reviews an order with no findings and defers.
Missing the fourteen-day Rule 23(f) window.

Each of these is a process failure rather than a doctrinal one, and each is avoidable by a team that treats certification as the case rather than as a motion.

Forum, removal, and the choices made before anything else

Where the case is litigated affects the certification standard, the availability of issue certification, the treatment of uninjured members, and the appellate posture. Both sides should think about it before the complaint is filed.

For plaintiffs. Filing in state court and resisting removal has traditionally been attractive, but the Class Action Fairness Act gives federal jurisdiction over most class actions with more than 100 members, minimal diversity, and more than $5 million in controversy. The local controversy and home state exceptions are narrow and require careful pleading to invoke, and a plaintiff who pleads into an exception constrains the class definition in ways that matter later.

Within the federal system, circuit differences are real: on ascertainability, on whether predominance must be satisfied case-wide before Rule 23(c)(4) issue certification, on the treatment of uninjured members, and on the intensity of the Rule 702 inquiry at certification. These differences are worth understanding before choosing a venue, and worth preserving through the multidistrict litigation process where one exists.

For defendants. Removal is nearly always available and usually right, but consider the transferee court if MDL consolidation is likely — the panel's assignment may matter more than the removal.

Multidistrict litigation. Where an MDL exists, certification is often deferred while bellwether proceedings develop the record, and the certification fight when it comes is informed by the bellwether results. The transferee court decides certification under the law of the transferor circuits for state-law claims, which creates genuine complexity in a nationwide class and is itself an argument against predominance where the governing law varies materially by state.

Choice-of-law and nationwide classes. A nationwide class asserting state-law claims must confront variation among state laws. Plaintiffs argue the differences are immaterial or can be managed with subclasses grouped by legal standard; defendants argue that applying fifty bodies of law defeats predominance and manageability. The plaintiff who has not prepared a state-law variation analysis before moving for certification will be asked for one, and the answer is not a footnote.

Budget, staffing, and the timeline

Certification is the most expensive phase of most class actions, and both sides should plan it as a project.

Timeline, typical for a case with a quantitative model:

Months 0–3: pleadings, motions to dismiss, initial case management.
Months 3–9: certification discovery, dominated by the transaction data production and its validation.
Months 6–12: named plaintiff depositions, custodial document production, corporate depositions on uniformity and variation.
Months 10–14: expert reports — opening, opposition, reply.
Months 13–16: expert depositions.
Months 15–18: certification briefing and the Rule 702 motion.
Months 17–20: hearing, sometimes multi-day with live testimony.
Months 18–24: decision.
Plus 14 days for a Rule 23(f) petition, and six to fifteen months if permission is granted.

Two years to a certification decision is normal. Three is not unusual in a large case.

Cost drivers, in order:

Experts. In a substantial antitrust or securities class action, expert costs frequently exceed all other litigation costs combined. Budget for the testifying expert's team, the consulting expert, and the data infrastructure.
Data. Production, validation, hosting, and processing.
Document review, which is merits-heavy but often runs concurrently.
Depositions, particularly of experts, which are long and require substantial preparation.
The hearing.

Staffing. A small senior team that understands the model, supported by people who can handle the volume. The most common staffing error on both sides is having the certification motion drafted by people who have never spoken to the expert.

For defendants, the budget conversation with the client should happen early and should be honest: certification is where the money goes, and a defence that underspends there and plans to fight on the merits has usually misallocated. If the case will settle after certification is granted, the certification phase is the entire fight.

For plaintiffs, the economics are asymmetric: the firm funds the case. That is a reason for discipline about which cases to bring and about narrowing the class definition early, when the cost of the narrower case is much lower than the cost of losing the broader one.

Explaining a regression to a judge

A great deal of certification practice reduces to whether the court understands the model well enough to evaluate the attack on it. Both sides should invest in this, and most do not.

What works:

A single sentence stating what the model does. "It compares what customers actually paid with what they would have paid if prices had been set the way they were before the conspiracy began."
A scatter plot with the fitted line. Nothing communicates a regression better.
A distribution of member-level estimates, which makes the uninjured-member question visible rather than abstract.
A single sensitivity table: the estimated overcharge under the plaintiff's specification and under three defensible alternatives.
The false positive test presented as a chart: here is the model applied to a period with no alleged conduct, and here is what it produced.
Plain-language definitions of the four or five technical terms the court will encounter — specification, benchmark, control variable, statistical significance, fitted value — supplied once, in a glossary, without condescension.

What does not work:

Reciting the regression equation.
Reports that run to 200 pages with the key result on page 147.
Expert testimony that answers a different question from the one asked.
Demonstratives with six lines, two axes, and a legend in eight-point type.
Treating the judge as either an economist or a novice. The right register is that of an intelligent, sceptical non-specialist who will ask exactly the right question about the assumption you were hoping to skip.

The tell that the presentation has failed: the court's opinion describes the models in the parties' own words without evaluating them, and resolves the motion on a procedural ground. That is a court that did not feel equipped to decide, and both sides should regard it as a self-inflicted outcome.

Issue certification as the middle path

Rule 23(c)(4) permits certification of particular issues, and it is the most underused provision in the rule.

When it fits. Where liability elements turn entirely on the defendant's conduct and are common — whether a design was defective, whether a policy existed, whether a statement was false, whether a duty was breached — but causation, injury, and damages vary. Certifying the conduct elements resolves the questions that are genuinely common and leaves the individual questions to individual proceedings.

Why plaintiffs undervalue it. An issue class does not produce an aggregate damages number and therefore does not produce the settlement pressure a full damages class produces. Counsel funding a case on a contingency reasonably prefer the larger prize.

Why it is often the right answer anyway. A denied certification produces nothing. An issue class produces a binding determination on the questions that matter most, materially improves the value of the individual claims, and frequently leads to a global resolution anyway once liability is established.

Why defendants should think carefully before opposing it. A defendant facing likely full certification may prefer an issue class: the aggregate exposure disappears, the individual claims must still be proved one by one, and many will never be brought. Reflexive opposition to any certification is not always the right instinct.

The circuit split. Some circuits require predominance to be satisfied as to the case as a whole before issues may be certified, which largely nullifies the provision. Others assess predominance within the certified issues, which makes it a genuine tool. Know which rule governs before building a strategy on it.

Drafting the issue. Precision matters more here than anywhere. The certified issue must be capable of a yes-or-no answer on common evidence, must not embed an individualized element, and must be framed so that the verdict actually binds in the subsequent individual proceedings. An issue certified as "whether the defendant's conduct was unlawful" is too abstract to bind; one certified as "whether the design of the Model 400 valve was defective as of January 2021" is not.

A short checklist for the week before filing

Plaintiffs:

Does the class definition use objective criteria and avoid defining membership by the merits?
Have obviously uninjured categories been excluded?
Does the damages model measure damages attributable to the pleaded theory, and only that theory?
Can the expert produce member-level results, and what do they show?
Has the model been run on a period with no alleged conduct?
Have the named plaintiffs been vetted for standing, arbitration agreements, releases, limitations, and credibility?
Is there more than one representative, covering each subgroup?
Is there a state-law variation analysis for any multistate claim?
Is there a trial plan and a draft verdict form?

Defendants, before the opposition:

Have we quantified the uninjured members using the plaintiff's own model?
Have we run the false positive test?
Have we disaggregated by product, customer, channel, and period?
Have we tested the benchmark period for contamination?
Are we filing a Rule 702 motion, with the opposition rather than after?
Are we arguing the model rather than the merits?
Have we requested findings on the specific expert disputes?
Is the Rule 23(f) deadline calendared for the day the order issues?

Neither list is long. Both are routinely incomplete in cases that are lost.