01Purpose and scope
The Criterica Score is a comparable, versioned measure attached to a file. It exists so that two parties looking at the same legal asset, on the same date, against the same stated population, read the same number and know exactly what the number means.
This specification defines what the Score is, what it estimates, what it must state to be valid, when it must not be issued and how it changes over time. It does not publish the models that produce it.
The Score is written for the people who have to rely on a number they did not compute:
- –Funders and lenders pricing, financing or marking a legal asset.
- –Law firms assessing their own portfolio or a matter within it.
- –Insurers and reinsurers reserving or pricing claims exposure.
- –Platforms and auditors who need a number that can be reproduced and challenged.
02Definition
A Criterica Score is a number from 0 to 100 attached to one legal asset, one portfolio or one counterparty at a stated date.
A valid Score always travels with five statements: the asset it describes, the date it is attached to, the outcome it targets, its horizon and its reference population. A number that arrives without them is not a Criterica Score.
- 1.Subject: the legal asset, portfolio or counterparty.
- 2.As-of date: the date the file was read.
- 3.Target: the outcome the Score estimates, as defined in the next section.
- 4.Horizon: the period the Score covers.
- 5.Reference population: the population the file is compared with.
The same file, scored again under the same version, on the same as-of date, returns the same number.
03Target
The Score estimates the probability-weighted standing of the asset against its reference population on a stated outcome. The outcome is one of four questions: who prevails, how long it takes, what it costs, or what it is worth.
Standing is a relative position, not a forecast of one result. A high Score means the file sits high in its reference population on the stated outcome. It does not say that a given matter will win, resolve by a given date or collect a given amount.
The target is stated per asset class in the asset-class annex that accompanies each version of this specification. An asset class with no annex has no Score.
Outcome families
| Question | What standing is measured on |
|---|
| Who prevails | The probability-weighted standing on the stated result for the party the asset depends on. |
| How long it takes | The probability-weighted standing on time to the stated resolution event. |
| What it costs | The probability-weighted standing on the stated cost measure. |
| What it is worth | The probability-weighted standing on the stated value measure, shown as a labeled distribution. |
04Horizon
The horizon is the period the Score covers, counted from the as-of date. A Score always states its horizon, and a Score is only comparable with another Score that states the same horizon.
A horizon the models cannot support is not rounded to the nearest one they can. The file is returned as Not rated for that horizon.
05Reference population
The reference population is the set of comparable assets the file is compared with. It is stated explicitly, by the criteria that define membership: asset class, forum or jurisdiction, case or claim type and period.
A Score without a stated reference population is invalid. Changing the reference population changes the Score, so it is recorded with the number every time.
06Mapping and grades
Model outputs map to the 0 to 100 scale by a monotone function: a file that stands higher on the stated outcome never receives a lower Score than a file that stands lower, all else equal. The mapping is calibrated against the reference population so that the scale means the same thing from one scoring run to the next.
The mapping is fixed within a version. It is re-mapped only at a version change, and the change is recorded in the change log.
Grades
Grades are five named bands on the scale. A grade describes position in the reference population and nothing else. It makes no claim about how well the Score performs.
| Band | Name | Scale | Reading |
|---|
| Band 1 | Well below reference | 0 to 19 | The file stands well below the reference population on the stated outcome. |
| Band 2 | Below reference | 20 to 39 | The file stands below the reference population. |
| Band 3 | Near reference | 40 to 59 | The file stands close to the middle of the reference population. |
| Band 4 | Above reference | 60 to 79 | The file stands above the reference population. |
| Band 5 | Well above reference | 80 to 100 | The file stands well above the reference population. |
07Not rated
When support is insufficient, the file is Not rated. It is never given a low Score. A low Score says the file stands low; Not rated says the Score cannot be issued.
A file is Not rated when any of these holds:
- –The data on the file is too thin to support the stated outcome.
- –The file falls outside the stated reference population.
- –The sealed-outcome rule is violated.
- –The stated horizon is not supported.
A Not rated result states which condition applied. It is recorded in the same way as a Score.
08Inputs and the sealed-outcome rule
The inputs to a Score are the facts of the file as they stood at the as-of date: the parties, the forum, the claim or case type, the procedural history, the terms of the asset and the evidence supplied for it.
The sealed-outcome rule: the file scored carries no resolution field, no collected amount, no resolution date and no field that encodes them. A file that carries any of these is Not rated. The rule keeps the answer out of the question.
Inputs the scorer supplies and inputs the scorer assumes are labeled separately on the output. A seller-stated figure is tagged as seller-stated.
09Validation status
Validation is reported per asset class. A claim about how well the Score performs for an asset class activates only after a preregistered backtest on that class passes, and the claim is made together with its evidence. Preregistrations and their results are published, whether they pass or fail, in the preregistration registry.
| Asset class | Status |
|---|
| Pre-settlement consumer | Not yet validated; claims activate after a preregistered backtest passes |
| Law-firm portfolio | Not yet validated; claims activate after a preregistered backtest passes |
| Commercial litigation | Not yet validated; claims activate after a preregistered backtest passes |
| Medical receivables | Not yet validated; claims activate after a preregistered backtest passes |
| Insurance claims | Not yet validated; claims activate after a preregistered backtest passes |
No row in this table says validated. A row changes only when a preregistered backtest for that asset class has passed and its result is posted.
Read the preregistration registry
10Versioning
The specification and the Score it defines carry semantic versions: a major version when the meaning of the Score changes, a minor version when a method is added or tightened without changing meaning, and a patch version for corrections to wording. A release candidate carries the suffix rc until comments are answered.
A change log records every version, its date and what changed. Each of these forces a re-mapping and therefore a new version:
- –A change to the target or the outcome definition.
- –A change to the reference population criteria.
- –A change to the mapping function or to the band edges.
- –A retrained model that moves the calibration.
Prior scores are not rewritten. A Score issued under a version keeps that version and stays valid as a record of what was issued. A file re-scored under a new version is issued as a new Score with its own version, and both appear on the record.
Change log
| Version | Date | Change |
|---|
| 1.0-rc | 2026-10-09 | Issued as a request for comment. |
11What the Score is not
- –It is not a merits opinion on any matter.
- –It is not legal advice.
- –It is not a forecast of realized returns or of a case’s realized dollar value.
- –It is not a credit rating.
Dollar outputs that accompany a Score are labeled distributions, and inputs are tagged by source. Where a return such as IRR or MOIC is shown, it is computed from those distributions and the stated terms of the asset, and it is labeled as computed.
12Governance and comment
Changes to the specification are proposed in writing, discussed against the evidence, issued first as a release candidate and recorded in the change log. Anyone may comment on a release candidate through the form on this page. Each comment is read, answered in writing and, where it changes the text, credited in the change log by role and organization type.
The comment period for this release candidate is open. Its close date will be stated here when it is set.
Errata
Errata correct errors in a published version without changing its meaning. They are listed here with the date and the text corrected.
No errata have been issued for version 1.0-rc.