Reproducible screening process combining multi-engine phonon consensus, fail-safe patent whitespace check, and proxy superconducting temperature triage

This process patent covers the reproducible screening pipeline used to identify superconductor and heavy-fermion candidates from the rare-earth silicide family. The core claim binds three independently-asserted operations: a four-engine machine-learning phonon consensus for dynamic stability, a fail-closed patent whitespace prescreen with a built-in sentinel control, and a disciplined handling of the estimated critical temperature as a triage signal rather than a claimed property. The integration of these three operations into a single, separately-claimed process gives any owner two independent layers of protection over the same verification engine — one covering the selection logic, one covering the reproducible pipeline that executes it. What makes this asset commercially distinctive is the fail-closed architecture of the patent prescreen. Most automated screening pipelines rely on live patent-search APIs that can silently return empty responses on service degradation, creating false-clean results that downstream automation accepts without alarm. This process uses a silicon nitride sentinel — a material known to return substantial prior art — to validate that each prescreen run is genuinely connected and returning results. The sentinel returned 1,005 claim hits in demonstrated runs, providing a concrete, calibrated threshold against which every production run can be checked. As IP-diligence and materials-discovery move toward continuous, automated candidate evaluation, a process that cannot silently fail is a meaningful differentiator. The claim is composition-agnostic and recurs on every new candidate set, structuring naturally as a platform license, subscription, or per-run fee rather than a one-time transaction. The independently-claimed operations also complicate design-around: a competitor would need to avoid the consensus step, the fail-closed prescreen, and the disciplined critical-temperature handling simultaneously to escape the claim.

Each candidate is validated by multiple independent machine-learning interatomic potentials. A material advances only when the engines agree on phonon (dynamic) stability — disagreement is surfaced, not hidden.

The process operates on a structured candidate set without being tied to any single chemical composition or space group. Dynamic stability is established by running finite-displacement phonon calculations under four independent machine-learning interatomic potentials — MACE, CHGNet, MatterSim, and ORB — and requiring consensus across those engines before a candidate advances. Across 99 four-engine phonon runs performed on this family, 77 returned consensus-stable results, a 78% pass rate. Requiring agreement across four independently-trained potentials is more stringent than any single-potential screen and is particularly important for f-electron intermetallics, where individual potentials can disagree substantially on zone-boundary phonon branches. The patent whitespace prescreen is conducted offline against a corpus of roughly 306,000 patent claims. Offline execution means the prescreen is not subject to live-API availability, but the fail-closed sentinel addresses the complementary risk: that a corrupted or zero-result response is misread as a genuinely clean landscape. A silicon nitride probe — chosen because its prior art is dense and well-established — must return at least its calibrated hit count (demonstrated at 1,005 claim hits) for the run to be treated as valid. If the sentinel fires below threshold, the run is flagged rather than accepted. This guard makes the prescreen auditable in a way that live-search-dependent approaches are not. The estimated critical temperature enters the pipeline strictly as a rank-ordering signal to prioritize which candidates receive more expensive downstream simulations. It does not serve as a selection criterion, it carries no claimed property value, and no performance representation is made on its basis. This disciplined scoping reflects the computational reality that MLIP-derived Tc proxies for rare-earth silicides carry meaningful uncertainty, particularly for materials without DFT-validated training data. The claim is about the process architecture, not about predicting any specific superconducting transition temperature.

We estimate the addressable market at $1 billion to $5 billion, spanning materials informatics platforms, IP-diligence services, and the superconducting and quantum-computing device sector. These estimates reflect the range of licensing contexts — from narrow field-of-use agreements with individual materials-discovery programs to broader platform rights with IP-analytics firms — and should be treated accordingly. The process claim monetizes differently from a composition patent: because it recurs on every new candidate set a licensee screens, the natural commercial model is a platform subscription or per-run fee rather than a lump royalty tied to a single material. The value anchor for pricing is displaced spend. A licensee running genus-scale campaigns periodically replaces paid live-search API calls and manual diligence reviews with an auditable, automated prescreen. Avoided wet-lab qualification costs for candidates that fail early — before expensive DFT runs or synthesis attempts — provide a secondary anchor. Both anchors scale with the size and frequency of screening campaigns, giving the licensor natural revenue participation in a licensee's growth. Named customer categories include materials-discovery houses running continuous candidate evaluation and IP-diligence firms offering clearance-adjacent analytics services. The recurring-use nature of the process, applied fresh on each new compound set or each update to the prior-art corpus, supports subscription structures with predictable revenue, which tends to command premium multiples relative to one-time royalty streams on composition patents.

fail-closed control prevents the silent false-clean failure mode of live patent-search services

process/verification-engine claim; not foreclosed by any composition prior art

Primary acquirers and licensees are materials-discovery platforms and IP-diligence firms seeking an auditable, automated screening engine. IP-diligence and patent-analytics companies are the strongest acquisition candidates: owning the fail-closed prescreen process directly differentiates their service from competitors relying on live search APIs and supports a premium, audit-grade offering that can be marketed to clients in regulated or high-stakes IP environments. Materials-discovery companies running continuous genus-scale candidate evaluation would license the process on a per-run or subscription basis, particularly those operating in superconducting electronics, quantum computing, or advanced magnet programs where candidate throughput is high and the cost of advancing a blocked material is significant. The license-versus-acquire calculus turns on the buyer's business model. A firm that intends to offer the auditable screening engine as a service to third parties has a strong incentive to acquire exclusively, both to control the brand and to deny rivals access to the fail-closed architecture. A firm that needs the pipeline only for internal candidate evaluation would find a non-exclusive field-of-use license sufficient, given the composition-agnostic scope of the claim. A strategic buyer seeking competitive positioning in IP-diligence analytics might acquire exclusive rights in the diligence field while leaving materials-screening field-of-use rights available for separate licensing.

The prescreen is explicitly a whitespace signal, not a formal freedom-to-operate opinion, and the corpus covers approximately 306,000 claims rather than the full patent literature. A buyer must not position it as legal clearance — it informs prioritization and flags obvious conflicts, but does not substitute for counsel review before commercialization. This is a structural limit of any large-scale automated prescreen, not a defect specific to this process, but it needs to be communicated clearly to downstream licensees to avoid misuse. The stability layer is MLIP consensus across four potentials, not first-principles density-functional perturbation theory. For f-electron systems, all four potentials share training-data limitations in the rare-earth intermetallic space, so majority consensus reduces but does not eliminate false-positive risk. The one open prosecution item — formally pinning the sentinel control-hit threshold through counsel — must be completed before conversion; until that term is fixed, the claim scope in the whitespace prescreen step is not fully defined. The critical temperature carries no claimed value and no performance representation, so no risk attaches there, but a buyer should ensure downstream marketing materials reflect that scoping accurately.