Ironburg Inventions v. Valve Corp. — Federal Circuit Tightens IPR Estoppel Standard, Reverses Both Invalidity Bars

Background

This is the second Federal Circuit appeal in a decade-long patent battle between Ironburg Inventions — a British company whose founders designed a video game controller with extra buttons on the back of the device — and Valve Corporation, creator of the Steam gaming platform. Ironburg owns U.S. Patent No. 8,641,525, which covers a controller with “elongate” and “inherently resilient” back controls operated by the user’s other fingers. In 2015, Ironburg sued Valve for infringing the patent with its Steam Controller.

Before trial, Valve filed an inter partes review (IPR) petition in 2016 challenging the ‘525 patent’s validity (IPR2016-00948). The IPR proceeded and the PTAB issued a final written decision in September 2017. Separately, a third party, Collective Minds Gaming Co. (CMG), filed its own IPR petition in January 2018 raising two different invalidity grounds: one based on a U.S. application called Kotkin, and another combining three references — Willner, a Japanese application called Koji, and Raymond (“Willner-Koji-Raymond”). Under 35 U.S.C. § 315(e)(2), once Valve’s IPR yielded a final written decision, Valve was estopped at trial from raising any ground it “reasonably could have raised” during its own IPR.

Ironburg moved to invoke that estoppel to block Valve from using the CMG Grounds at trial. The district court twice estopped Valve. The first estoppel order was reversed in Ironburg I (2023) because the court had improperly placed the burden of proof on Valve — the Federal Circuit clarified that Ironburg must prove the grounds were reasonably discoverable using a “skilled searcher conducting a diligent search” standard. On remand, the district court took additional discovery and estopped Valve again (May 2024). Valve appealed, producing the present case.

The Court’s Holding

Judge Hughes, writing for the panel, reversed the district court on both grounds.

Kotkin ground — classification search insufficient when it returns 26,333 results. Ironburg relied on evidence that Valve’s own pre-petition search agent (Landon IP) had run patent classification searches that included the IPC and CPC codes under which Kotkin was indexed. The district court held that because the right codes were searched, Kotkin was discoverable. The Federal Circuit disagreed sharply. Landon IP’s searches returned 26,333 references from those classifications. A skilled searcher confronting a readout of that size would not simply declare victory — she would use keyword searches or other narrowing techniques to trim the results to a manageable set. The district court’s analysis stopped at the classification search without asking whether and how those 26,333 results would be narrowed. That analysis “does not go far enough.” The court reversed outright, finding the evidentiary record underdeveloped on remand, and directed the district court to allow Valve to assert Kotkin at trial.

Willner-Koji-Raymond ground — forward citation searches are hindsight-tainted. To prove Raymond was discoverable to a skilled searcher in 2016, Ironburg engaged Cardinal IP — a search firm — to run searches in 2023. The Federal Circuit found Cardinal’s entire approach was infected with hindsight bias. Among Cardinal’s first searches was string “number 3,” which used forward-and-backward citation searching without date restrictions. Forward citations are patents that later cited the subject patent — they did not exist in 2016 and thus could not have guided a diligent pre-petition searcher. Because Cardinal’s early forward-citation search returned results that then shaped later keyword and classification choices, the taint permeated the entire search. The patent examiner’s subsequent attempt to apply date filters to only one of the struck searches was “too little too late” — it could not retroactively cleanse an iterative search built on hindsight-contaminated results. The supplemental classification search that ultimately located Raymond was also tainted: Cardinal’s own project manager admitted it was conducted specifically because Raymond had not yet been found — itself evidence of hindsight-driven searching. The court reversed and remanded for the district court to consider the patent’s validity in light of Willner-Koji-Raymond.

Key Takeaways

• Classification searches need narrowing analysis. Establishing IPR estoppel via classification-search evidence requires more than showing the relevant codes were searched. When those searches return thousands of results, the proponent must also show how a skilled searcher would have narrowed the results — through keywords, citation searches (pre-petition ones), or other means — to a manageable, reviewable set.
• Forward citation searches carry hindsight risk. Forward citations — patents that later cited the relevant reference — are inherently dated after the IPR petition and therefore raise a structural hindsight problem. Experts using forward citation searches to reconstruct what a pre-petition searcher “would have found” must date-restrict those searches from the start, not as a post-hoc correction.
• Search contamination is holistic. Because Mr. Greenia’s early forward-citation search shaped his subsequent keyword choices and classification selections, a late-stage correction could not save the analysis. Courts will look at whether an initial hindsight-tainted search infected the overall methodology, not just the specific string that was later struck.
• Purpose of a supplemental search matters. Conducting a focused supplemental search specifically because a target reference has not yet been found is strong evidence of hindsight bias — the searcher already knows what she is looking for.

Why It Matters

IPR estoppel is one of the most consequential doctrines in modern patent litigation: it determines what prior-art grounds a petitioner can raise at trial after losing — or even winning — an IPR. For years, the Federal Circuit has deferred to district courts on the “skilled searcher” question without providing clear methodological guardrails. This decision supplies two important ones. First, a raw classification search result count matters: courts cannot declare a reference “discovered” when the relevant codes returned tens of thousands of results without narrowing. Second, the temporal discipline of citation searches is mandatory — forward cites must be date-restricted, and if they are not, the contamination can void the entire expert report.

The practical effect is that patent defendants who filed IPRs on certain grounds remain at risk of estoppel, but the bar for proving what a “skilled searcher” would have discovered is now measurably higher. Patent owners seeking to invoke estoppel will need more carefully designed search protocols — preferably ones that replicate 2016 conditions rather than using 2023 knowledge to reconstruct them. Judge Stark’s concurrence, which highlights the Federal Circuit’s ongoing reticence to define the skilled searcher standard comprehensively, signals that further guidance — perhaps from an en banc panel — will eventually be needed.