Thresholds vs Optimization: The Court of Appeal on Software Method Claims
The Court of Appeal clarifies the distinction between binary threshold checks and continuous optimization algorithms in software method claims, while addressing the admissibility of late-filed non-infringement arguments.
The Court of Appeal has established that when a method claim requires determining whether a user modification complies with a threshold before generating a revised output, prior art employing constraints within a continuous optimization algorithm does not anticipate the claim.
How did the Court of Appeal interpret the temporal order of features 1.6 and 1.7?
In UPC-CoA-36/2026, the panel reviewed an order granting a preliminary injunction based on EP 4 346 690, which protects a method for automated management of clinical modifications to orthodontic treatment plans. The core dispute centered on two claim features. Feature 1.6 requires determining in real time that a requested modification is within a predetermined threshold. Feature 1.7 requires generating a revised treatment plan when the modification is within that threshold.
The Court of Appeal held that these are two distinct steps with a specific temporal order and substantive relation (reasons 56). The determination step must precede the generation step. Crucially, if the modification exceeds the threshold, no revised plan is generated. This interpretation aligns with the patent's stated advantage of reducing computing resources and overhead by avoiding the generation of undesirable treatment plans (reasons 53).
Why did the optimization algorithm in D2 fail to anticipate the claimed method?
The Appellants argued that the patent lacked novelty over D2, a prior art document describing the recalculation of treatment plans using numeric limits. The panel rejected this argument by distinguishing between a binary threshold check and a constrained optimization problem.
In D2, the constraints define the feasible region of optimization within which the output is computed (reasons 94). The system always produces a constraint-compliant treatment plan, even if the requested modification violates a constraint, by outputting the best achievable compromise. The Court of Appeal found that D2 lacks the specific mechanism of a distinct first assessment that governs whether a revised plan is generated in a second distinct step (reasons 95). The constraints in D2 merely shape the outcome of the calculation rather than acting as a gatekeeper to allow or reject the generation process.
When will the Court of Appeal admit late-filed non-infringement arguments under Rule 222.2 RoP?
The Appellants first raised non-infringement arguments in their Rejoinder before the Local Division, which disregarded them as late-filed. On appeal, the panel exercised its discretion under Rule 222.2 RoP to admit a subset of these arguments.
The panel noted that in summary proceedings for provisional measures, a more lenient approach may sometimes be justified. The arguments concerning "scenario 1" were admitted because the technical functionality of the Appellants' software was uncontested, and the Respondent had four months to study the arguments before filing its Statement of response (reasons 161 and 162). Conversely, the panel disregarded the arguments for scenarios 2 and 3 because they disputed the technical qualifications and lacked any justification for the late filing (reasons 164).
Does an error message prove the presence of a threshold determination?
To prove infringement for scenarios 2 and 3, the Respondent pointed out that the Appellants' iOrtho software displayed an error message when a user changed a tooth position or arch shape by too great an amount. The Court of Appeal held this was insufficient to prove the mechanism of features 1.6 and 1.7.
The facts did not show a distinct determination step preceding the generation step. An error message does not establish that the software uses a threshold-based determination of admissibility rather than a different mechanism within the system (reasons 172). However, the panel did find infringement for scenario 1, where the software checked if an attachment was moved within a red gingival line before generating the revised plan (reasons 168).
What must practitioners change when drafting software method claims after UPC-CoA-36/2026?
When drafting claims for automated software methods, practitioners must carefully distinguish between binary threshold checks and continuous optimization processes. If an invention relies on saving computational resources by rejecting non-compliant inputs before processing, the claims must explicitly recite the temporal order of the determination and generation steps. A generic reference to applying constraints will not suffice to distinguish over optimization algorithms.
Furthermore, litigators defending against preliminary injunctions must front-load all non-infringement arguments in the Objection. Relying on the Court of Appeal to admit late-filed arguments under Rule 222.2 RoP is highly risky and only succeeded here because the underlying technical facts for one specific scenario were entirely uncontested.
Sources
Further Reading
Articles related to the topics covered above.
Evidentiary Gaps and Single-Appliance Claims in UPC Preliminary Injunctions
The Düsseldorf Local Division dismissed a preliminary injunction request after the applicant relied on marketing materials showing sequential treatment phases to prove infringement of a single-appliance method claim.
Constructive Knowledge and the Donkey-Kick: PI Urgency at the UPC
The Düsseldorf Local Division establishes a strict 'should have known' standard for PI urgency under Rule 211.4 RoP, while clarifying how explicit specification disclaimers limit claim scope.
Claim Construction and Multi-Material Purge Towers at The Hague Local Division
The Hague Local Division dismisses an application for provisional measures, strictly interpreting a 'layer by layer' claim feature based on the patent's description and drawings.