Lattice Graph × Escape Velocity (EV3)

Lattice Graph operates a computational materials-discovery platform built around a knowledge graph that spans millions of compositions, drawing on experimental literature, computed databases, and proprietary internal results. The core engine cross-validates any candidate material through multiple independent physics engines — including state-of-the-art machine-learning interatomic potentials such as MACE and CHGNet alongside density functional theory — and requires consensus across them before a result is trusted. When engines disagree, that disagreement is itself a signal: a quantitative calibration marker that tells you whether a predicted property is robust or cherry-picked from whichever method gave the most favorable number. For technical diligence in the AI-for-materials and critical-minerals space, two additional layers matter most. The first is a large-scale patent corpus — more than 300,000 materials patents — screened for whitespace, freedom-to-operate exposure, and claim density by composition family. This lets an investor ask whether a startup's claimed discovery lane is genuinely unoccupied or whether it runs straight into a thicket of prior art. The second, and harder to replicate, is a corpus of labeled negative results: 23,196 failed-experiment kill edges that document what does not work and why, accumulated from internal records that never appear in the published literature. Most AI-for-materials models have never been trained or benchmarked against a systematic failure set, because almost none exists outside of private industrial labs. Lattice Graph holds one. The platform wraps all of this in a governed, provenance-linked knowledge graph that connects composition to structure to predicted and measured properties to synthesis routes to IP status. Every claim in the graph traces back to its source, its validation history, and the disagreement profile across physics engines. For a diligence reader, that provenance chain is the difference between a result you can put in an investment memo and a result you have to trust on the founder's authority alone.

Escape Velocity's thesis — that the most defensible companies of the next decade are those that design matter computationally rather than discover it by brute force — puts the firm in exactly the position where independent technical verification matters most. AI-for-materials and critical-minerals decks are now easy to construct. Every pitch claims a proprietary model, a data moat, and a credible path to a well-capitalized buyer. The problem is that almost none of those claims can be stress-tested using only publicly available tools and datasets, because the public materials corpus is overwhelmingly positive, broadly shared, and systematically blind to failure. A model trained on it will tend to over-predict success; a founder benchmarking on it will tend to over-claim defensibility. EV3 needs an independent read that does not take the target's own numbers at face value. That is the specific gap Lattice Graph fills. The platform was built to do exactly what a sophisticated technical advisor does — verify opportunity, characterize the IP landscape, and probe model quality — but with data that is not available to the market and at a speed consistent with a live deal process. The Opportunity and Buyer Intelligence engine answers the commercial question: is this discovery lane real whitespace, and does a funded buyer actually exist for the material or capability being proposed? The negatives atlas answers the technical question: does this company's model know anything the public corpus does not already give away, or does it re-predict known dead ends as successes when benchmarked against an independent failure set? Together, those two signals address the two questions a diligence committee needs answered before committing at frontier valuations. The strategic fit is also structural. EV3 invests across AI-for-science foundation models, autonomous lab platforms, generative materials companies, and the refining and separation businesses that must actually produce materials at scale. Lattice Graph's coverage spans the full stack — from discovery and prediction through IP whitespace and supply-chain concentration — so a single diligence subscription can serve the firm across its entire thesis rather than requiring a different specialist for every deal type.

Escape Velocity's thesis — that the most defensible companies of the next decade are those that design matter computationally rather than discover it by brute force — puts the firm in exactly the position where independent technical verification matters most. AI-for-materials and critical-minerals decks are now easy to construct. Every pitch claims a proprietary model, a data moat, and a credible path to a well-capitalized buyer. The problem is that almost none of those claims can be stress-tested using only publicly available tools and datasets, because the public materials corpus is overwhelmingly positive, broadly shared, and systematically blind to failure. A model trained on it will tend to over-predict success; a founder benchmarking on it will tend to over-claim defensibility. EV3 needs an independent read that does not take the target's own numbers at face value. That is the specific gap Lattice Graph fills. The platform was built to do exactly what a sophisticated technical advisor does — verify opportunity, characterize the IP landscape, and probe model quality — but with data that is not available to the market and at a speed consistent with a live deal process. The Opportunity and Buyer Intelligence engine answers the commercial question: is this discovery lane real whitespace, and does a funded buyer actually exist for the material or capability being proposed? The negatives atlas answers the technical question: does this company's model know anything the public corpus does not already give away, or does it re-predict known dead ends as successes when benchmarked against an independent failure set? Together, those two signals address the two questions a diligence committee needs answered before committing at frontier valuations. The strategic fit is also structural. EV3 invests across AI-for-science foundation models, autonomous lab platforms, generative materials companies, and the refining and separation businesses that must actually produce materials at scale. Lattice Graph's coverage spans the full stack — from discovery and prediction through IP whitespace and supply-chain concentration — so a single diligence subscription can serve the firm across its entire thesis rather than requiring a different specialist for every deal type.

For an EV3 analyst running a diligence pass, the most-used surfaces in the platform are the opportunity and buyer leaderboard views and the knowledge-graph explorer. The leaderboard views display the ranked discovery landscape for a given material family or application domain — showing which opportunity lanes the engine rates as genuine whitespace, which are crowded, and which have documented buyer demand — alongside the funded-buyer affinity signals that let an analyst pressure-test a go-to-market claim in minutes. These are the same signals available through the API, surfaced as a structured dashboard that drops directly into a diligence workflow without requiring a data-engineering setup. The knowledge-graph explorer gives analysts the ability to trace any specific claim — a composition, a predicted property, a stability result, a synthesis route — back to its provenance, its multi-engine validation history, and the negatives attached to that material family. Composition-intelligence reports generated through the platform produce provenance-backed, readable summaries per material that can be dropped into investment memos or shared with a partnership committee. Batch screening workflows allow an analyst to push a target's full list of claimed candidate materials through the negatives check and the trust-and-disagreement signals at once, returning a structured view of how many candidates survive contact with an independent failure corpus. These are evaluation and grounding tools sized for an investor's workflow: fast, scoped, and designed to produce a defensible technical read rather than to replicate the discovery platform internally.

The natural structure for EV3 is a scoped diligence engagement rather than an asset license or co-development program. A standard pass focuses on a single target or active round: Lattice Graph runs the opportunity and buyer reality check against the claimed discovery lane, executes the negatives reproducibility test against the target's model or candidate list, and delivers an independent technical memo with findings grounded in data rather than the company's own benchmarks. Additional layers — supply-chain concentration and import-risk analysis, freedom-to-operate and patent-whitespace screening, or multi-engine trust-and-disagreement scoring on specific predicted properties — are scoped in as the thesis requires, not sold as a fixed bundle. Repeat engagements across EV3's pipeline roll into a standing diligence subscription that gives the investment team ongoing, metered access to the opportunity intelligence engine and the negatives atlas across inbound deal flow. This allows the firm to triage AI-for-materials and critical-minerals decks against a consistent independent baseline at the speed of a live process. Per the firm's lead framing, fast scoped engagements are priced in the range of thirty to sixty thousand dollars per pass, scaling with the depth of coverage and the number of material families in scope. No exclusivity or IP transfer is involved — the service is an independent technical-diligence capability plus metered data access, and findings are structured for direct use in investment committee materials.