Litigation Finance Intelligence

Litigation risk
has a timeline.

Underwrite every case, price the capital at risk, and see what it does to the book — before a dollar goes in.

Forecast duration. Forecast the ending. Underwrite the book.
Built from federal civil cases.
Book a backtest on your closed cases →
Predicted resolution windowCadence · 80% interval
061218243036 mo
Illustrative distribution of time-to-resolution across comparable federal civil matters, with an 80% prediction interval shaded.
One model · the whole decision

Not just when a case resolves.
Whether the capital comes back, and what it does to the book.

Duration
Time to resolution per case, as an 80% interval whose measured coverage is published — not a point guess.
Disposition
How it ends: settlement, dismissal, judgment, or transfer — forecast, not assumed.
Capital at risk
The dismissal rate is your loss probability. Read per case type from millions of dockets.
Portfolio economics
Per-case distributions roll into IRR, capital lockup, recycling, and reserves.
Underwriting
What a new case does to the book's IRR — before a dollar is committed.
Filing signals
Jury demand, class allegations, pro se, district — read off the complaint on day one.
Benchmarked to the courts
Cross-checked against the judiciary's own published statistics, within a fraction of a month.
Auditable calibration
Every forecast ships with its own out-of-sample error record. No black box.

Timing is the engine. Everything above is what it drives. Tertius forecasts timing and disposition — never merits or damages.

01 — The clock

Every case has a clock.

Capital isn't committed indefinitely — it's committed until a case resolves, and that timeline determines when it returns, when it can be redeployed, and how a portfolio performs. Tertius makes it measurable.

Capital lockup · one case
CommittedRecyclable
Capital committed
Capital recyclable
COMMIT · mo 0
TEMPO FORECAST · mo 14
0612182430 mo
Tertius forecasts where that point lands.
Capital stays committed until the case resolves—Tertius prices when it's likely to come back.
02 — The ending

Duration is only half the forecast.

A three-year case isn't one forecast — it may settle, transfer to MDL, or end another way entirely. Tertius forecasts both the timeline and the path.

Forecast paths · one antitrust caseBar length = months to that outcome
FILED · mo 0
061218243036 mo
The same case ends many ways. Tertius forecasts each.
Disposition and timing, forecast together—so a transfer at three months and a settlement at three years are priced as the different outcomes they are.
03 — Capital at risk

The question isn't when.
It's whether your capital comes back.

A dismissal is a loss. A settlement is a return. Tertius reads the odds of each from millions of resolved dockets, so the ending is priced before a dollar goes in.

How capital ends, by case typeDismissed · Settled or won · Transferred
Loading capital outlook…
Dismissed — capital at riskSettled or won — capital returnsTransferred — rides on elsewhere

Procedural priors from mature cohorts — how cases end, not who deserves to win. Tertius forecasts timing and disposition, never merits or damages.

04 — The filing

The forecast begins on page one.

The complaint already carries measurable signals — jury demand, class allegations, pro se litigants, district selection. Read at filing, before the first scheduling order is entered.

Complaint · intake · example6 signals read
Jury demand
Demanded▲ longer
Class allegations
Present▲ longer
Representation
Counselnot pro se
District
S.D.N.Y.
Cause of action
Antitrust▲ longer
Amount in controversy
$1M+
The forecast is adjusted from the filing itself—before the first scheduling order is entered.
Read from page one
05 — Time is money

Time is part of the risk.

The same payoff, earned later, is a lower return.

Return sensitivity · illustrative
Baselinemo 14≈ 28% IRRDelayed +6 momo 20≈ 19% IRR↓ 9 IRR pts lost
Push a case out six months and its annualized return falls with it. Timing isn't operational—it's a financial variable.
01It changes IRR
02It extends capital lockup
03It delays recycling
04It increases uncertainty across the portfolio

Timing isn't an operational detail.

It's a financial variable.

06 — Portfolio impact

Plan capital with time in mind.

Cases don't move independently. Portfolios fail together. Courts slow together. Settlement behavior shifts together. Doctrine ripples through a book at once.

TPortfolio Planning/Fund III
80% intervalLast 20 closedExport ↧
Portfolio liquidityCapital lockupReserve requirementsIRR sensitivityCapital recycling
Open positions5 of 5
CaseForecast resolution · 0–36 moCapital
Anderson v. Meridian
Discovery
mo 14
$4.2M
Coastal Freight Co.
Pretrial
mo 8
$2.8M
re: Halcyon Labs
Appeal
mo 23
$6.1M
Prairie Union
Fact discovery
mo 16
$3.5M
Vertex Holdings
Trial-ready
mo 19
$5.0M
Capital returned over time
$21.6M
Committed
mo 15
50% returned
mo 23
Fully recycled
Tertius · illustrative exampleReproducible · version-stamped
01Capital lockup
02Reserve requirements
03IRR sensitivity
04Capital recycling
05Portfolio liquidity

Because knowing when capital returns is just as important as knowing if it returns.

07 — The model

Built from outcomes.
Measured against reality.

Every forecast is grounded in real federal civil cases. Every model is graded on held-out cases it never saw during training. Performance isn't estimated—it's published.

Introducing
Most litigation forecasts ask for trust. Tertius publishes evidence.
Current release
Federal cases in the dataset
Out-of-sample grading cases
Median error
Inside the predicted 80% interval
BenchmarksOut-of-sample
Benchmark
Docket-average baseline
Tertius
Loading live benchmark comparison…
Docket-average baseline: a single constant prediction (the training set's own unconditional median — no case-type, district, or amount signal), scored on the identical held-out cases with the identical rules. Not a strawman — it's what "use last year's docket average" gets you.
New in Cadence 3What this release adds over Cadence 2
01
Conditions on pro se status, class actions, jury demands, and jurisdiction
02
Surfaces how cases end—settlement, dismissal, judgment, or transfer—not just when
03
Correlated portfolio sampling with explicit, editable assumptions—independence is a toggle, not a hidden default
Loading calibration…
Measured on held-out cases where follow-up already covers the predicted range — still-open cases included, never dropped.

Every release includes:

Historical training data
Held-out validation
Calibration against federal court statistics
Stress testing against systemic slowdowns
Version history for reproducibility

Reproducible, versioned, and open to inspection.

The flywheel

Graded on your own book.

Everyone gets the same public-record model. Only you see it scored against your cases.

01
Public record is the model
Every forecast comes from a model trained exclusively on public federal dockets — 9.1M cases, no customer data inside it.
02
Your outcomes grade it
Each outcome you record scores Tertius against your own realized results, out of sample — accuracy on your book, not a vendor average. Your outcomes don't train the current model.
03
Milestones, disclosed
Recorded outcomes count toward future recalibration milestones. If a future model version trains on customer-contributed outcomes, its calibration page will say so before it ships — and captions, parties, and amounts never leave your organization.

And you can see it working: every forecast is graded against your own realized outcomes, out of sample — the accuracy number no in-house spreadsheet can produce.

08 — Trust & control

Built for trust.

No black boxes, no invented milestones. Just duration forecasts you can reproduce and defend.

Benchmarked against the courts themselves
Duration forecasts only
No invented milestones
Every export is version-stamped
Historical forecasts remain reproducible
Your data stays yours
Know what generated every forecast—and exactly which model
Cross-checked against the federal judiciary's own published statistics (AO Table C-5)
Portfolio correlation assumptions stated explicitly and adjustable — never hidden in the math
Forecast provenance · every forecastReproducible
Forecast ID
Unique per forecast
Model
Engine
Input snapshot
SHA-256 of inputs
Prediction interval
80% · calibrated
Method
Duration forecast only
Version-stamped
Tertius
Re-run this forecast anytime — byte-identical output, from the exact model that produced it.
09 — FAQ

Fair questions.

The ones skeptics ask first — answered the way everything on this page is answered.

Why not just use the average time to resolution?

Because the risk lives in the spread, not the middle. Cases of the same type resolve anywhere from under two years to past fifteen, and the median itself moves by multiples with court, claim, and amount. Published averages also count only closed cases — the slowest cases are still open, so the average can't see them. Tertius forecasts the full distribution, counts every case, and updates the forecast as a case ages.

Case length depends on attorney strategy. What does a timing forecast actually mean?

A calibrated range, not a theory of the case. Twenty years of real dockets already contain every delay tactic and settlement posture attorneys actually use — the forecast is the distribution those behaviors produce. A life table prices lifespan without knowing anyone's story. This is the same discipline, applied to dockets.

Does a longer case mean a worse outcome?

No — it means a worse return on the same outcome. The same recovery, arriving two years later, can cut annualized return in half, and capital locked in one case can't fund the next. Duration is a financial variable, independent of merits. Tertius forecasts the clock; the merits stay yours.

How is the model graded?

Out of sample, on more than a million held-out cases the model never saw in training — still-open cases included, never dropped. The scorecard above is live, every forecast ships with the error record of the model that produced it, and the underlying data is cross-checked against the federal judiciary's own published statistics.

Which courts does it cover?

Federal civil, by design. Patent, antitrust, securities, and MDL mass torts live in federal court — and so does most of the capital deployed against them. Every case in the dataset is a real federal docket.

Every forecast is measured against held-back cases and published with its historical accuracy.

No analyst estimates.
No hidden adjustments.
No artificial certainty.

Just forecasts, measured against reality.

Get access

See it on your
own book.

Book a backtest and Tertius is scored against your own closed cases — outcomes you already know — before you rely on a single forecast.

Book a backtest on your closed cases →