The merits, weighed before the money.
Fulcra reads an asserted right and the product accused of infringing it, builds the claim chart limitation by limitation, and returns a calibrated read on merit, validity and enforceability — designed to compress a first-pass read from months toward hours. It exists to open enforcement to meritorious, under-resourced rights-holders — and to say no to the weak and the opportunistic. The guardrail is the product.
ELIGIBLE
Draft terms issued — qualified-counsel sign-off required before any capital moves.
Make diligence cheap, and you make nuisance cheap too.
The thing that quietly rate-limits bad litigation today is that analysing a case is expensive. Collapse that cost — which is the whole pitch — and the very same engine can either open the courthouse to people priced out of it, or industrialise shakedowns at volume. Fulcra is built, deliberately, for the first and against the second.
An anti-troll filter no merits number can override.
Before Fulcra weighs how winnable a case is, it asks whether this is a case that should be funded at all. Origin, real commercial harm and opportunism signals are checked first — and a Blocked verdict overrides even a strong profile. A pure-yield machine would chase this business; Fulcra is designed to turn it away.
Real origin, real harm
The right traces to a genuine inventor, company or lab — and the originating holder, or a real operating successor, still carries the economic interest and is the party funded. Demonstrable market overlap, real commercial injury, no mass-assertion pattern. Proceeds to scoring.
Mixed signals
Origin or harm is unclear, or the chain of title needs work. Routed to a human before anything is weighed — never auto-advanced.
Assembled to assert
A right acquired to assert, a holder that neither makes nor researches anything, serial or mass-identical demands, or claims priced to settle below the cost of defence — whoever first invented it. Declined at the door.
Docket the right. Chart the evidence. Weigh three axes. Decide — or abstain.
Docket
The asserted right is parsed into a structured record — a patent’s independent claims broken into limitations, a design’s figures, a trade-secret dossier — and the accused product is turned into a citable evidence store with stable locators.
Chart
For every limitation, the engine retrieves the specific evidence that reads on it and builds the claim chart — one row per element, each cell linked to its source. A limitation with nothing behind it renders loud: no evidence found.
Weigh
Three orthogonal axes — infringement, validity and enforceability — each returned as a band with a confidence interval, because they fail independently. The weakest essential link sets the ceiling; an adversary red-teams the strongest attack on each.
Decide — or abstain
Merits, quantum and cost resolve to one of four first-class outcomes. Where the evidence is thin or the construction unresolved, Fulcra returns Insufficient-Merit — an abstention, logged and explained, not a quiet yes.
Every verdict is a cited document, not a vibe.
Literal infringement needs all limitations present — one missing element defeats the case. So the chart is the credibility artifact: each element mapped to a specific evidence locator, each cell citation-linked, and a doctrine-of-equivalents fallback offered only where the literal read is absent. The cell that says no evidence found is the one that matters most.
Three axes, because a case can be winning and worthless at once.
A single number hides the killer. A claim can be 95% infringing and fatally invalid; averaging the two reads “fine.” So Fulcra reports three axes separately — each a band, a confidence interval and a state that can read Insufficient — and shows which one caps the case.
Infringement
Driven by the chart under the all-elements rule: the case is only as strong as its weakest essential limitation, with an equivalents path and dependent-claim fallbacks where the literal read fails.
Validity
An adversarial red-team runs the strongest invalidity theory — prior art, obviousness, eligibility, enablement — and scores survival against the best attack, not the average one.
Enforceability
Standing, clean title, remedy and injunction availability, parallel-invalidation exposure, the defendant’s design-around escape, and whether a judgment is actually collectable. A strong claim that can’t be realised is not a fundable one.
We keep the famous three-digit score — and we keep it locked.
A credit-style score is a friendly way in, so the composite merit index exists. But a three-digit number implies a precision no one has earned before the outcomes are in. So it renders greyed and calibration-locked — it only unlocks to a published figure once all three axes are calibrated and the reliability curve sits on the line for held-out cases. Until then, the honest surface is the banded, interval’d profile that is allowed to say “not sure.”
Four honest outcomes. Abstaining is one of them.
Calibrated merits, likely recovery, cost and duration resolve into exactly one of four verdicts — each first-class. A decline is not an error, and an abstention is not a failure. An engine that never abstains is simply miscalibrated, so abstain-rate is a headline quality metric, not something to hide.
Back the fight
Merits clear the hurdle, confidence is calibrated and the gate says Qualified. Produces a draft term sheet — capped funder return, non-control covenants, conflict screens.
No case to bring
The accused product doesn’t actually read on the asserted claim, or a clean design-around moots it. There is nothing meritorious to back — precisely the call a yield-chasing engine gets wrong.
Economics fail
The merits may be fine, but the recovery is too small, the cost too high, the duration too long, or the defendant judgment-proof. Returned with honest reason codes.
Insufficient merit
The engine can’t reach a calibrated verdict — an evidence gap, an unresolved construction, split prior art. Not a decline: a specific evidence-request list.
The moat isn’t the model — it’s being right about how right you are.
Anyone can call the same frontier model; that part is a commodity. What compounds is the ledger: every case Fulcra touches — funded, cleared, declined or abstained — is tracked to its real outcome, and those outcomes recalibrate the scores. More cases, better calibration, better selection, better outcomes. Accuracy is easy to claim; the reliability curve has to be earned, case by case.
Reported alongside accuracy, not buried. A system that always answers is one that isn’t calibrated — so abstention is a published quality signal.
You mostly see outcomes of cases that got funded. The curve is only honest if that is handled — propensity weighting, backtesting against historical public dockets, and synthetic hold-outs. Never funding a weak case to harvest data.
Below ~$5M, the math says don’t look. Fulcra makes looking cheap — worldwide.
Hand-built diligence costs about the same whether a claim is worth $200,000 or $200 million — so beneath roughly $5 million in damages the unit economics collapse, and a perfectly valid claim simply goes unlooked-at. Not in one jurisdiction — everywhere, for every company on the wrong side of that line. That gap is the whole opportunity, and exactly what Fulcra is built to close: drive the marginal cost of a first-pass, evidence-linked read toward zero and the entire global long tail of sub-$5M disputes — the companies, inventors and labs the incumbent model was never going to screen — becomes a diversified, underwritable book.
Make enforceability cheap to assess, and the long tail opens.
The months of senior-lawyer diligence that price out every sub-$5M dispute are exactly what keep valid, under-resourced rights off the table. Collapse that cost and a calibrated enforceability read — infringement, validity, realisable recovery — becomes affordable across the long tail it was never economic to examine. The same engine can answer the inverse, defensive question too, but that is a byproduct, not the pitch.
Litigation outcomes are event-driven and largely uncorrelated with equity markets — which is exactly why disciplined, portfolio-scale underwriting matters. “Uncorrelated” is not “low-risk”; single cases are binary. That is an argument for calibration and abstention, not against them.
Four things compound. The language model is none of them.
A competitor has the same model but a cold calibration set and no loss history. What Fulcra accumulates instead is harder to copy and grows with every case it touches.
Published reliability curves, weakest-link gating, abstain discipline — aiming to be demonstrably calibrated, not merely accurate.
A proprietary dispute-outcomes dataset generated as a byproduct of operating — and barely in structured form anywhere else.
Claim-construction scaffolds, equivalents logic and jurisdiction rule-packs — accumulated legal-ops engineering, not a prompt.
Built to the strictest responsible-funding rules on the horizon, so a future directive is something to absorb, not scramble against.
We built this the way a careful underwriter would — starting from what we will not do.
Litigation finance has an obvious dark version: drive the cost of due diligence to zero and you can industrialise nuisance suits, arm patent trolls, and turn other people’s disputes into a derivative to be traded at speed. That version is easy to build and impossible to defend. So the guardrails here aren’t fine print — they are the structure of the product.
No funding for rights acquired purely to assert — no practicing or R&D origin, no fund.
No nuisance or shakedown campaigns — mass form demands or claims priced below the cost of defence.
No overriding an abstain with capital appetite — a weak case is never funded because the portfolio math clears.
No funder control of litigation or settlement, and no return structure that manufactures coercive pressure.
No outward-facing “who can I sue” targeting tool — the chart engine is not sold as an assertion weapon.
No decision without qualified-counsel sign-off, and no hidden black-box number presented as certainty.
No trading live disputes as instruments — no secondary market in other people’s lawsuits, and no designing around responsible-funding regulation.
The honest answers.
Isn’t this just a patent-troll machine with better marketing?
It is built to be the opposite. The commercial-nexus gate runs before any merits score and blocks rights held by an entity that neither makes nor researches anything and only asserts, dormant rights dusted off to assert, serial or mass-identical demands, and claims priced to settle below the cost of defence — and a Blocked verdict overrides a fundable profile. The purpose is the opposite of a troll shop: enforcement capital reserved for meritorious claimants — an inventor, a lab, an SME with a valid right — who otherwise couldn’t afford their day in court, with qualified counsel accountable for every funded case.
Why not just publish the single 321–987 score the pitch is famous for?
Because a three-digit number implies a precision no one has earned before the outcomes are in, and it hides which risk is fatal — a case can be strongly infringing yet fatally invalid. So the composite index renders greyed and calibration-locked, and only unlocks to a published figure once all three axes are calibrated and the reliability curve holds on held-out cases. Until then the surface is the banded, interval’d, three-axis profile that is allowed to read “Insufficient.”
Can the engine decide to fund on its own?
No. Claim construction, the doctrine of equivalents, eligibility and the final merits opinion that capital relies on are legally operative, contested and jurisprudence-dependent — the engine can scaffold them but not decide them. Every funded case requires sign-off from independent counsel admitted in the relevant jurisdiction, who reviews the chart and the reasoning, can veto, and whose edits to the engine’s draft feed back into calibration.
The source assumed litigation funding is deregulated. Is it?
No — that premise is the shakiest part of the original idea, and we corrected it. The absence of a single tailored law is a pause, not deregulation: responsible-funding principles, collective-redress rules and long-standing doctrines already bind, and the trend runs toward more oversight. Structures are designed to respect champerty, maintenance and claim-assignment limits jurisdiction by jurisdiction, and Fulcra holds to the strictest plausible standard now — capped funder returns, no funder control of the case, conflict and transparency disclosure — so an incoming directive is something to absorb, not scramble against.
What stops a sophisticated applicant from gaming the score?
The engine only sees what an applicant submits, so one-sided or unverifiable inputs lower confidence rather than raise the score, and load-bearing facts are flagged for independent corroboration. Selective documentation reads as a reason to abstain, not to approve. And because outcomes are tracked to real results, patterns that correlate with later losses are learned and priced — the ledger is what keeps the inputs honest over time.
The merits, weighed before the money.
Fulcra is an early concept from The Adjacent Possible. We’re building the claim-chart engine, the calibrated three-axis profile and the loss-history ledger — and we’d like to pressure-test the guardrails with funders, operating companies and under-resourced rights-holders who want enforcement to be fair, calibrated, and willing to say no.
Or write to [email protected]
Fulcra is not an insurer, a credit bureau or an investment product, and offers no regulated financial, investment or legal advice.