Search
60 of 60 terms
A
Acceptance Test
A defined, pre-agreed check a model, dataset, or system component must pass before it is accepted into production use. Acceptance tests convert a subjective judgment about whether something is good enough into a checkable pass-or-fail criterion set in advance of the result.
Advance Rate
The discount applied to a collateral pool's stated value to determine how much capital a lender will advance against it. In legal-asset lending, the advance rate should be applied against a stated percentile of the collateral's distribution rather than against its mean or a point estimate.
Appellate Base Rate
The historical frequency of affirmance, reversal, or modification on appeal, computed by jurisdiction and case type, the primary pricing input for a judgment's residual risk once merits uncertainty has already resolved at trial and liability and damages have been decided.
Asset Tracing
The investigative discipline of identifying and locating a defendant's assets, used to assess and manage collection risk on a judgment. Asset tracing draws on a different data set than pre-judgment outcome modeling and is a core competency for judgment preservation.
ATE (After-the-Event Insurance)
A policy taken out after a dispute has arisen to cover a claimant's adverse costs exposure if the matter is lost. ATE increasingly wraps litigation finance positions directly, requiring the insurer to price the same underlying litigation outcome risk the funder is also pricing.
B
Backtest
A validation method that freezes a model at a defined baseline and scores it against matters it has not seen, comparing the frozen model's stated probabilities to actual resolutions once they occur. A backtest run against a model's own training data measures nothing and is not a genuine backtest.
Base Rate
The underlying frequency at which a defined outcome occurs across a specified population, the foundation any calibrated probability estimate is built from. A base rate estimated from too small or too narrow a sample is noise dressed up as a finding, not a validated base rate.
Bellwether
A representative matter tried first within a multidistrict litigation to inform the valuation of the broader consolidated docket. Bellwether outcomes create a correlation structure across every other matter in the same MDL that does not exist in an independently proceeding case.
Borrowing Base
A defined pool of collateral against which a lender advances capital, discounted by an advance rate and certified on a regular cadence so the lender's exposure tracks the collateral's current value. Extended to legal receivables, the collateral itself requires re-scoring as procedural events occur, not only at each scheduled certificate.
C
Calibration
The property of a probability estimate matching realized frequency across a defined population. A model that states a given probability across enough comparable matters is calibrated if that share actually resolves that way over a sufficiently large sample, checked out-of-sample and out-of-time rather than against its own training data.
Capital Call
A fund's request for committed but undrawn capital from its limited partners, timed against the fund's actual deployment schedule. Duration-aware underwriting improves capital call planning by giving a fund a real distribution, rather than a single assumed date, for when deployed capital returns.
Co-Investment
An arrangement in which a capital partner invests directly alongside a fund in a specific position or portfolio segment, typically to gain concentrated exposure to opportunities the partner has separately underwritten confidence in. Co-investment terms in legal-asset finance turn on the same distributions a fund's own positions are priced from.
Collection Risk
The risk that a judgment, though legally valid, cannot actually be collected because a defendant lacks the capacity or willingness to pay. Collection risk is priced from defendant-specific asset and payment data, a different data set and discipline than pre-judgment outcome-probability modeling.
Concentration Risk
Exposure clustered along a specific axis, judge, legal theory, defendant relationship, or duration profile, that a conventional credit framework built around counterparty and sector does not capture. A legal-asset portfolio diversified on conventional axes can still be concentrated on the axes that actually govern this asset class.
Contingent Risk
Risk whose materialization depends entirely on a future, uncertain event, such as a litigation outcome, rather than on a counterparty's independent capacity to perform. Legal-asset positions are a pure form of contingent risk, which is why outcome-probability modeling, not conventional credit analysis, is the correct pricing tool.
Corpus
The full body of structured, resolved-matter records a platform's models are built from, whose real value lies in its verified unique record count rather than its raw ingested total. A corpus's scale should never be cited from raw counts without disclosed deduplication and provenance status.
Counterparty Tape
A dataset of actual, realized-dollar outcomes supplied by a capital partner who deployed real positions in a specific asset class, the specific evidence required to activate a realized-outcome prediction claim for that class through a preregistered backtest against unseen matters.
Coverage Snapshot
A stated, dated description of which jurisdictions, case types, or asset classes a model or corpus reliably covers, distinguishing deep, validated coverage from thin, nominal coverage. A coverage claim without a snapshot date and sample depth should be treated as potentially stale.
D
Decision Record
The timestamped artifact documenting an underwriting decision, price, structure, capital committed, and portfolio context, captured at the moment capital is committed and referencing the specific prediction record it relied on. Kept separate from the prediction record, it is what lets a later review distinguish a bad model from a bad price.
Deduplication
The process of identifying and merging records that describe the same underlying event, entered into a corpus more than once through overlapping sources, repeated harvesting, or unverified merged datasets. Deduplication must run continuously as a corpus grows, not as a one-time cleanup.
Disposition
The specific, defined resolution class a matter concludes with, dismissal, settlement, judgment for the claimant, judgment for the defense, within a regulated market's own outcome taxonomy. The disposition taxonomy differs across regulated markets even where the underlying subject matter looks similar.
Docket
The chronological record of filings, rulings, and procedural events in a matter, maintained by a court, and the primary source of the historical ground truth that makes court-record corpora verifiable rather than self-reported. Docket data also supports duration modeling, since procedural events are recorded with dates.
Duration Distribution
A probability distribution over the time a matter takes to reach resolution, built from a corpus of matters that actually resolved, rather than a single expected date. It typically carries a right-skewed shape, with a median shorter than the mean because a minority of matters extend well past the bulk of the population.
E
Eligible Collateral
The subset of a legal receivables pool that qualifies for inclusion in a borrowing base, defined by exclusion criteria built for this asset class's specific weaknesses, procedural age, defendant or judge concentration, independently disputed claims, rather than the criteria conventional receivables lending uses.
Evidence Class
A category of underlying data distinguished by its provenance, sensitivity, and handling requirements, such as public court records versus medical lien data versus seller-stated inputs. Treating every evidence class identically in a governance program overlooks distinctions that materially affect how each class should be verified and used.
F
Feature Engineering
The rules-based process of constructing the specific inputs a model scores against from raw ingested data, built to be inspectable and reproducible rather than generated by a free-form reasoning process. Feature engineering is one of the stages a rules-based pipeline keeps deliberately narrow and auditable.
Fund Life
The defined remaining term over which a fund can hold positions before it must wind down, a structural constraint that has to be checked against a portfolio's duration distributions, not only its expected durations, to avoid holding positions the fund cannot wait out.
L
Labeled Distribution
A settlement or award distribution built from comparable resolved matters with every input tagged by source, distinguishing what a party stated about a matter from what the corpus independently corroborates. Labeling discipline is what prevents a distribution from silently blending unverified assertions with verified evidence.
Leakage
A modeling defect in which information that would not actually be available at prediction time enters a model's training process, inflating its apparent accuracy without producing a genuinely predictive model. Leakage findings are disclosed and the affected models retired or restricted rather than left in production quietly.
Lien
A legal claim against settlement proceeds asserted by a medical provider, insurer, or other party with a financial interest in a claimant's recovery. Lien behavior, aggregated across resolved matters, functions as a structured data asset for modeling injury severity, case duration, and net-to-claimant proceeds.
M
Mandate
An institution's stated risk appetite, portfolio context, and capital availability at the moment a position is committed, information a prediction model was never built to hold and should not be built to hold. The human decision at the point of commitment exists specifically to apply the mandate to the model's output.
Material Change
A shift in methodology, training corpus, or target definition significant enough to require disclosure against a fixed prior state. A material change is only checkable if a sealed baseline captured before the change exists to compare the new state against.
MDL (Multidistrict Litigation)
A federal procedure consolidating similar civil matters filed across different districts into one court for coordinated pretrial proceedings. MDL structures introduce bellwether trials and aggregated settlement mechanisms that behave differently, as an asset class, from a single-plaintiff matter proceeding independently.
Model Registry
A full accounting of production models at the level of signature uniqueness, target definition, filter, sample size, and validated metric, distinguishing true predictive models from governance artifacts and reference tables. Registration asserts that two models sharing an identical signature are one model, not two.
N
NAV Lending
A financing structure advancing capital against the net asset value of a fund's existing portfolio, rather than against a single receivable. Applied to legal-asset funds, NAV lending requires the same portfolio-level distribution and duration modeling this platform builds for underwriting individual positions, applied at fund scale.
Non-Recourse
A structure in which a capital provider's recovery depends entirely on a specific matter's resolution, with no independent claim against the claimant's other assets if the matter is lost. Non-recourse structures concentrate a lender's risk directly in the outcome distribution, with no secondary repayment capacity to fall back on.
P
Percentile Pricing
Pricing a position off a stated percentile of its underlying distribution rather than its mean, with the choice of percentile reflecting the capital's actual risk tolerance. Percentile pricing lets different capital layers price the same underlying distribution differently and honestly.
Point Estimate
A single number, an average settlement value or a median duration, that collapses an entire distribution into one figure and discards the width and tail information a capital allocator actually needs. Pricing capital from a point estimate treats genuinely different risk profiles as though they were identical.
Portfolio Finance
Capital deployed against a diversified pool of legal-asset positions rather than a single matter, priced using portfolio-level concentration and correlation analysis across judge, legal theory, defendant, and duration axes, not only conventional counterparty and sector diversification a borrowed credit framework checks.
Preregistration
Locking a backtest's population, cutoff date, and success criteria before any result is generated, preventing the evaluation from being adjusted after an unfavorable early result to find a framing that passes. Preregistration is what separates a genuine test from a search for a favorable description.
Procedural Posture
A matter's current position within its regulated market's defined process, the specific stage, motion pending, discovery underway, appeal filed, that determines which portion of the outcome and duration distribution is actually still live. Procedural posture is what triggers re-scoring between scheduled reviews.
Production Model
A model actively deployed to generate outcome or duration predictions used in real underwriting decisions, distinct from an experimental, retired, or reference model still present in a broader registry. Only production models should be counted toward a claim about an institution's active predictive capability.
Provenance
The documented record of where a specific piece of data actually came from, what license or public-record basis it rests on, and its verification status. Provenance documentation is what makes a corpus's scale claim checkable rather than a figure a counterparty is asked to trust.
R
Realized Outcome
The actual, structured result of a resolved matter, matched back to its original prediction. Realized-outcome claims about actual dollar returns activate per asset class, only once a model trained on counterparty tapes with real realized dollars has passed a preregistered backtest for that specific class.
Receivable
A right to future payment, in this asset class typically a litigation finance advance, a law firm fee receivable, or a judgment, used as collateral in a borrowing base or securitization structure. Legal receivables lack a fixed maturity and are frequently non-recourse, distinguishing them from conventional trade receivables.
Reconciliation
The process of identifying and merging duplicate records describing the same underlying event across sources, a required step before a corpus's scale or a base rate's sample size can be reported honestly. Reconciliation has to run continuously as new data enters a growing corpus, not once.
Regulated Market
A body of matters resolved under one procedural and regulatory regime, with its own defined outcome taxonomy, timeline structure, and recordkeeping conventions, the correct analytical unit for organizing outcome intelligence in place of a practice-area or industry-vertical label a market map might use.
Reliability Curve
A plot of stated probability against realized frequency across bins of a model's score, the specific artifact that demonstrates a calibration claim rather than merely asserting it. A reliability curve without the sample size behind each bin disclosed is incomplete evidence.
S
Sealed Baseline
A frozen, timestamped snapshot of model versions, inputs, and outputs, locked against retroactive edit at the moment it is captured. It gives every later claim of improvement, drift, or material change a fixed prior state to be measured against, rather than a description someone wrote from memory.
Servicing
The ongoing administrative and monitoring function that tracks a legal-asset position's procedural status, re-scores its distributions as events occur, and manages collection once a matter resolves. In this asset class, servicing is inseparable from monitoring, since procedural events change the position's priced value continuously.
Settlement Band
A labeled distribution of settlement or award value, built from comparable resolved matters and bounded by a stated percentile range, rather than a single dollar figure. It states what a population of comparable matters has historically realized, not what any one pending matter will realize.
Signature Uniqueness
The registration principle that two models sharing an identical target, filter, sample size, and validated metric are counted as one model, not two, preventing a model registry's headline count from being inflated by duplicate entries describing the same underlying model.
W
Warehouse Facility
A revolving credit facility that finances a growing pool of legal receivables before permanent capital or securitization takes them out, structured around a borrowing base that re-scores continuously as the underlying collateral's procedural posture changes, rather than only at each scheduled reporting date.
Waterfall
The contractual order in which proceeds from a resolved legal-asset position or portfolio are distributed among capital providers, typically senior debt, then subordinated capital, then equity or residual claimants. Waterfall mechanics determine how a labeled settlement distribution translates into what each capital layer actually receives.
Requesting a term added or clarified?
Institutional partners, allocators, and counsel are welcome to suggest additions to this glossary or ask for clarification on how a specific term applies to their own portfolio.