The America Invents Act, enacted in 2011, overhauled the U.S. patent system. One of its key components was the Inter Partes Review (IPR) process, whereby parties could challenge the validity of a patent at the Patent Trial and Appeal Board (PTAB), a theoretically lower cost alternative to validity challenges in district court. The tradeoff: if an IPR reaches a final decision, the challenger cannot re-litigate, at either the PTAB or in district court, invalidity “on any ground that the petitioner raised or reasonably could have raised.”

Simple enough, right? But fifteen years later, parties are still extensively litigating the meaning of whether a ground “reasonably could have [been] raised.” Few cases regarding this so-called IPR estoppel have reached the Federal Circuit. Last week, in a precedential decision, the Federal Circuit issued a significant ruling that sheds light on IPR estoppel and several important issues. In Ironburg Inventions Ltd. v. Valve Corporation, No. 2024-2088 (Fed. Cir. June 18, 2026), the court reversed a district court’s estoppel order for the second time in the same case, sending a clear message: pointing to a classification search that returns tens of thousands of results is not enough to prove that an invalidity ground based on one prior art reference in those results reasonably could have been raised.

Background: A Long-Running Patent Fight Over Game Controllers

The case traces back to 2015, when Ironburg sued Valve over its Steam Controller, alleging infringement of a patent covering a handheld game controller with flexible, elongated controls on the back. Valve filed an IPR petition in 2016, and the proceedings that followed spawned not one but three IPR petitions, two Federal Circuit appeals, and now a third round of district court proceedings — a reminder that patent litigation rarely ends quickly.

A third party (CMG) filed its own IPR petition, after Valve’s first two, against the same patent using two new invalidity grounds — one based on a reference called Kotkin, another combining references called Willner, Koji, and Raymond. Valve tried to add those same grounds to its district court invalidity case (CMG’s IPR was settled and resolved without a PTAB decision). Ironburg moved to estop Valve.

Round One: The Federal Circuit Shifted the Burden

In the first appeal (Ironburg I, 2023), the Federal Circuit held that the district court had improperly placed the burden on Valve to show it couldn’t have found the references. The correct rule: the patent owner bears the burden to prove that a “skilled searcher conducting a diligent search reasonably could have been expected to discover” the references at issue. The court remanded for Ironburg to try again.

Round Two: Same Result, New Problems

On remand, Ironburg came back with two categories of evidence. First, it pointed to Valve’s own pre-petition search records from a firm called Landon IP, which showed that classification codes covering Kotkin and Koji had been searched. Second, it hired Cardinal IP to run its own searches and demonstrate that the relevant references would have turned up.

The district court again estopped Valve on both grounds. The Federal Circuit again reversed — this time on two independent bases.

Problem 1: 26,333 Results Is Not a Discovery

On the Kotkin ground, the court held that the district court made a fundamental error: it treated a classification search that returned 26,333 references as sufficient proof that Kotkin was discoverable, without asking whether a skilled searcher would actually have reviewed that many results. The Federal Circuit said that’s not how searching works. Evidence in the record, including from both parties’ search experts, showed that classification searches are just a starting point, typically narrowed by keyword searches or citation searches before a searcher can meaningfully review the results. The court stopped short of requiring proof of manual review in all cases, but held squarely that “something more is required for estoppel when a classification search alone returns an unreviewable number of search results.”

Problem 2: Hindsight-Contaminated Searches Are Worthless

The Willner-Koji-Raymond ground failed for a different reason: hindsight bias. Cardinal’s searcher, conducting searches in 2023, had used forward citation searches that pulled in patents citing the disputed references, which didn’t exist when Valve filed its 2016 IPR petition. One of those post-filing citations came from Ironburg’s own design patent, issued in 2023, which cited Raymond because Raymond had been raised in CMG’s IPR. In other words, the very prior art history that post-dated Valve’s petition had infected Cardinal’s search methodology from the beginning.

Ironburg tried to cure this by having a different Cardinal employee apply date filters to the problem search strings, but the court was unimpressed. Because the hindsight bias had infected an early, foundational search whose results shaped the entire subsequent methodology, a belated patch five months later was too little too late. On top of that, Cardinal’s supplemental search, the one that finally located Raymond, was conducted only because the searchers knew Raymond still hadn’t been found. That’s textbook hindsight: structuring the search to find a specific target you already know exists.

How Ironburg II Evolved the Clearlamp Doctrine

To appreciate what changed last week, it helps to understand where the “skilled searcher” framework came from. The standard the courts have been applying traces largely to Clearlamp, LLC v. LKQ Corp., 2016 WL 4734389 (N.D. Ill. Mar. 18, 2016), one of the earliest district court decisions to give the § 315(e)(2) “reasonably could have raised” language practical content for prior art that was never included in a petition.

Clearlamp did three things. It adopted, from the AIA’s legislative history, the test that estoppel reaches grounds a skilled searcher conducting a diligent search reasonably could have been expected to discover before the petition was filed. It tied estoppel to specific prior art references rather than to broad subject matter. And, most influentially, it offered a two-part evidentiary roadmap: to show a skilled searcher would have found a reference, a party should (1) identify the search string and search source that would locate the reference, and (2) present evidence, likely expert testimony, explaining why that search criterion would be part of a diligent searcher’s work.

For nearly a decade, many district courts treated that two-part test as close to a safe harbor. If you could name the search string that surfaced the reference and explain why a competent searcher would run it, you had generally done enough. The district court in this very case applied Clearlamp exactly that way for Kotkin: Valve’s own searcher had run classification codes covering Kotkin, so, under the Clearlamp framework, the court deemed Kotkin discoverable and stopped there.

Ironburg II does not overrule Clearlamp. It endorses the standard as a starting point, but makes clear there is more to consider:

The two-part test is not sufficient when a search returns an unreviewable number of results. The Federal Circuit held that the Clearlamp approach “does not go far enough” when a classification search returns too many references. A diligent searcher must not only run the search that locates the reference, but also must actually pull it out. The court demanded evidence of the narrowing steps (keyword searches, citation searches, or equivalents) that would reduce the results to a set a searcher would realistically review.

An accessibility/findability distinction. Judge Stark’s concurrence formalized a line Clearlamp never drew. A reference is accessible if it sits somewhere within a search’s output, but it is only findable if a diligent searcher reasonably would have encountered and reviewed it.

A second analytical step Clearlamp never contemplated. Judge Stark’s concurrence also articulates that proving estoppel may require two distinct showings: first, that the references were findable, and second, that a skilled searcher, once holding those references, reasonably would have recognized the specific invalidity ground the challenger seeks to assert. Clearlamp was concerned only with locating references and said nothing about the reference-to-ground inference. The majority did not need to reach this step here because Valve did not contest it, but the concurrence frames it as a separate, live requirement.

A hindsight-bias discipline Clearlamp did not address. Clearlamp gave no guidance on how a litigant should design an after-the-fact search commissioned to demonstrate discoverability. Ironburg II imposes a demanding check: the search must be designed and run as if the target references were unknown, and contamination at an early, foundational step propagates through the entire methodology. Filtering only the most obviously tainted step at the end will not save the rest.

Judge Stark’s Concurrence: A Useful Two-Step Framework

Judge Stark wrote separately to highlight the contributions the opinion makes to the developing law of IPR estoppel, and to suggest a framework the majority didn’t formally adopt. He distinguished accessibility from findability: a reference sitting in a searched classification code is accessible, but accessibility alone doesn’t mean a skilled searcher would actually find it. He noted that some grounds may be so apparent on the face of a reference that finding the reference necessarily discloses the ground, but that this is a separate inquiry from findability, and one the courts and the Board will have to develop case by case.

What the Federal Circuit Left Open

For all that Ironburg II clarifies, the panel was careful to flag several questions it did not decide. Practitioners should treat these as unsettled:

• The standard of review for the skilled-searcher inquiry. The court expressly declined to decide whether the inquiry is a question of law reviewed de novo or a question of fact reviewed for clear error.
• When “something more” is required, and what it is. The concurrence emphasized that determining when accessibility must be supplemented by additional proof, and what that proof must be, is a context-dependent inquiry that will turn on the field of the invention, the state of the prior art, and the search strategies a diligent searcher would use. The court left development of this requirement to the district courts and the PTAB.
• The reference-to-ground (step two) inquiry. Because Valve challenged only whether the references were findable, the court did not address what a patent owner must show to prove that an invalidity ground — as opposed to the underlying references — reasonably would have been discovered.

What This Means Going Forward

For patent owners seeking estoppel, this decision raises the evidentiary bar meaningfully. It is no longer enough to show that the relevant classification codes were searched. Owners will need to demonstrate that narrowing techniques – keyword searches, citation searches, or something equivalent – would have reduced the results to a reviewable set and that the target reference would have survived that narrowing. They will also need to be careful that any expert search they commission is run blind, without any methodology step that draws on knowledge of what the challenger eventually found.

For challengers defending against estoppel, the opinion offers useful ammunition when a patent owner’s search evidence was conducted years after the relevant IPR petition using modern tools or with knowledge of the prior art that post-dates the petition.

The case now returns to the Western District of Washington, where Valve will finally get to argue Kotkin and Willner-Koji-Raymond on the merits. After nearly a decade of litigation, the invalidity fight is just beginning.