The Line Between Fact and Law

Teva v. Sandoz, claim construction, and the standard of review as a constraint engine.

Every legal system has rules about who is allowed to be wrong. A standard of review is one of them. It does not decide cases directly; it decides how much a higher court may second-guess a lower one — and in doing so it quietly allocates power between institutions. For seventeen years, patent law answered that question for claim construction with a single word: de novo. The Federal Circuit reviewed a trial court’s reading of a patent claim from scratch, owing no deference to the judge who had actually heard the evidence. In Teva v. Sandoz (2015), the Supreme Court took part of that word back. Teva and the Process of Claim Construction, by Lee Petherbridge and R. Polk Wagner, is the most useful account I have read of what that change does — and the reason it holds my attention is that the whole dispute turns on a problem an engineer will recognize on sight: what it means to read a measurement.

The disconnect

Claim construction — fixing the meaning and scope of the words in a patent claim — is the hinge on which most patent cases swing. Since Markman (1996), it has been a job for the judge rather than the jury. The Federal Circuit then went a step further in Cybor (1998), treating claim construction as a pure question of law subject to de novo review on appeal. The practical effect was that the appellate court could re-decide the meaning of a technical term off a cold paper record, without crediting the trial judge’s evaluation of live expert testimony. Reversal rates were high and famously panel-dependent, and the Supreme Court and the Federal Circuit had spent years talking past each other about whether claim meaning was “law” or “fact.” Teva was the Court closing that gap.

The dispute was a measurement

The contested term was “molecular weight,” in Teva’s patents on Copaxone, a multiple-sclerosis drug. Here is the engineering of it: a polymer is not one molecule but a distribution of chain lengths, and “molecular weight” can mean at least three different things depending on how you weight that distribution — peak, number-average, or weight-average. The number you report is a function of the method that produced it. Sandoz argued the term was therefore indefinite. The district court, crediting Teva’s expert on how to read a size-exclusion chromatogram printed in the patent, found that a skilled reader would understand it to mean peak average molecular weight. The Federal Circuit, reviewing de novo, disagreed and held the patent invalid.

Strip away the doctrine and the fight is about epistemic authority. Is “what a skilled artisan would understand this instrument trace to mean” the kind of thing decided by the people who can read the instrument, or by the people who can read the text? Calling it “law” routes it to the appellate court and the text. Calling it “fact” routes it to the trial record and the expert. The label is doing all the work.

What Teva held

The Court, 7–2, split the question rather than answering it whole. The subsidiary factual findings underlying a construction — typically findings about extrinsic evidence, such as what skilled people in the field take a term to mean — are reviewed for clear error under Rule 52(a)(6). The ultimate construction of the claim remains a legal conclusion reviewed de novo. Claim construction became a two-step operation: first the facts, then the law applied to them.

It is worth noting how little this changed the result that prompted it. On remand, the Federal Circuit invalidated the Copaxone claims again, even under the new and more deferential standard, because the ultimate indefiniteness holding was a legal one that survived deference to the underlying facts. The headline rule moved; the outcome did not.

Method, not incidence

This is precisely the gap Petherbridge and Wagner build their argument in. Their claim is that Teva will barely move the incidence of deference — how often a deferential standard actually changes who wins — but will substantially reshape the methodology of claim construction, and that the methodology is the part that matters. They identify two mechanisms. First, litigants now have a strong incentive to put extrinsic evidence about claim meaning into the district court record and to argue that it is decisive, because evidence-based findings are the part of the analysis an appellate court must respect. Second, the opinion itself prescribes a sequence — analyze the extrinsic evidence (fact), then weigh its direction against the patent (law) — that pushes trial courts toward grounding claim meaning in objective information about how skilled readers use terms, rather than in unaided textual intuition.

Their normative bet is that this is good for the system: anchoring meaning in objective evidence and running it through a settled process should make litigation-driven construction more predictable, encourage better-drafted claims over time, and let patent law tune its incentives more finely. I find the descriptive claim persuasive and the framing genuinely clarifying — they separate “does deference change who wins” from “does deference change how the work is done,” and show that the second question is where the action is.

The engineer’s reading

What strikes me, coming from measurement-heavy work, is that Teva is the patent system slowly conceding something engineers take as a starting point: the meaning of a technical term is rarely separable from how it is produced and measured. “Molecular weight” is method-dependent; so, in my own field, is something like threshold voltage, whose value is inseparable from the pulse history and the measurement protocol that produced it. De novo review implicitly modeled a patent as a text — a self-contained string of words an appellate judge could re-parse to the correct answer. Teva nudges the system toward modeling a patent as a description of a technical reality whose terms sometimes cannot be resolved without evidence about how practitioners actually use them. That is not a small move. It is the law admitting that some questions of meaning are irreducibly empirical, and relocating authority over them toward the part of the system closest to the evidence.

Read against the view of institutions as constraint engines, this is a more interesting lesson than any shift in win rates. The most consequential rules an institution has are often not about what gets decided but about who is allowed to decide it and how far they can be overruled. Teva moved that boundary a few inches — from the appellate panel toward the trial judge and the expert witness — and Petherbridge and Wagner’s optimism is, at bottom, a bet that better-located decision rights produce better-calibrated outcomes downstream. That is a systems argument, not a doctrinal one, and I think it is the right frame for the case.

The open question

I am less sure the optimism is fully earned, and the doubt comes from their own first mechanism. If the surest way to insulate a claim construction from reversal is to manufacture a factual dispute and bury it in expert testimony, then the same incentive that anchors meaning in evidence also rewards flooding the record with it. “More predictable” and “more expensive battle of experts” are not mutually exclusive; Teva may deliver both. Whether the methodology Petherbridge and Wagner praise actually disciplines claim construction or merely relocates the gamesmanship is, to my mind, the open empirical question their argument sets up — the kind of thing that, with a few more years of post-Teva data, is worth measuring rather than predicting.

Either way, the case is a clean instance of what this section is about. A standard of review looks like a piece of appellate housekeeping. It is in fact a guardrail that decides whom the system trusts when a technical word and a legal boundary turn out to be the same object. Teva adjusted that guardrail by a single category — fact instead of law — and, as Petherbridge and Wagner show, its most durable consequence was not who won, but how the question now has to be asked.

References

Lee Petherbridge & R. Polk Wagner, Teva and the Process of Claim Construction, 70 Fla. L. Rev. 379 (2018).
Teva Pharmaceuticals USA, Inc. v. Sandoz, Inc., 574 U.S. 318 (2015); on remand, 789 F.3d 1335 (Fed. Cir. 2015).
Markman v. Westview Instruments, Inc., 517 U.S. 370 (1996).
Cybor Corp. v. FAS Technologies, Inc., 138 F.3d 1448 (Fed. Cir. 1998) (en banc).