Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
A lawyer, funder, or firm wants to model whether a claim or portfolio is economically viable.
The question involves damages, probability of success, fee arrangements, ATE, adverse costs, funder return, or recovery waterfall.
The user needs sensitivity or scenario analysis rather than a single best-case number.
This skill models economics. It does not recommend whether to bring, settle, fund, or abandon a claim.
Audience and Work Shape
Audience: litigation partners, litigation-finance counsel, funding analysts, and legally supervised team members with enough quantitative fluency to review assumptions and formulas.
Work shape: bounded transactional calculation in calculation_mode; accretive judgment in specification_mode. The skill models assumptions and scenarios, but the responsible lawyer or funder decides what action to take.
Legal Failure Modes
Legal support, not legal advice: outputs are working calculations and assumptions registers, not merits, enforceability, tax, regulatory, or professional-conduct advice.
Privilege/confidentiality: claim economics may be privileged or work-product material. Sharing with funders, auditors, counterparties, insurers, or third parties can affect privilege/confidentiality and needs lawyer approval.
Accountability: a lawyer or authorised investment decision-maker must sign off before any output is used for fund/no-fund, bring/no-bring, settle/no-settle, or client-facing advice.
Access Modes
This skill works in three modes:
Calculation mode - use supplied assumptions and an available calculator, spreadsheet, code execution environment, or arithmetic worked in-session.
User-supplied model mode - review an existing spreadsheet, calculator output, term sheet, budget, damages model, or funding proposal supplied by the user.
Specification mode - build the model structure, assumptions list, and formulas, but do not invent calculated outputs.
If key inputs or a calculation method are unavailable, produce an assumptions register and model specification. Do not invent IRRs, percentiles, waterfalls, or Monte Carlo outputs.
How It Works
1. Define the scenario
Capture:
Claim type or portfolio type.
Jurisdiction and any regime-specific cost rules.
Claim count and cohort funnel if relevant.
Damages distribution.
Liability and quantum probability.
Expected time to resolution.
Settlement discount assumptions.
If a figure is unknown, ask whether to use a range, placeholder, or scenario variable.
Every input must be labelled as user_supplied, document_sourced, calculated, or assumption_to_verify. Keep that label through the output so the reader can separate facts from modelling assumptions.
Also record currency, VAT/tax inclusion or exclusion, nominal vs real basis, date of valuation, time unit, discount-rate assumption if used, and the formula used for each calculated output. Do not present calculated figures without these modelling metadata.
Counsel, expert, platform, marketing, and administration costs.
Inter-partes recovery assumptions.
ATE premium and adverse-cost exposure.
VAT or tax treatment only if the user supplies the correct assumption.
Flag internally inconsistent inputs and keep them out of headline outputs unless the user expressly asks to model the inconsistent scenario. Label any such output inconsistent_input_scenario.
3. Model funder economics
Support common return structures:
MOIC.
Percentage of proceeds.
Interest or structured debt.
IRR hurdle.
Time or stage ratchets.
Commitment and ancillary fees.
Hybrid or greater-of / lesser-of structures.
State clearly who receives what under each layer.
4. Apply recourse and waterfall
Build the waterfall:
Recoveries received.
Costs and disbursements.
ATE or insurer position.
Funder entitlement.
Firm entitlement.
Client residual.
Handle pari passu tiers, caps, floors, and shortfall allocation.
5. Run scenarios and sensitivity
At minimum, produce:
Base case.
Downside case.
Upside case.
Break-even claim count or recovery.
Sensitivity to win probability, damages, time to resolution, costs, and recovery rate.
If stochastic modelling is requested, specify distributions, correlations if any, iteration count, random seed if used, tool/calculation method, and outputs including P5, P50, and P95. If you cannot actually run the calculation, do not invent percentiles; provide the model specification or pseudocode instead.
6. Report decision-useful outputs
Recommended outputs:
Firm IRR / margin.
Funder IRR / cash multiple.
Client net recovery.
Worst-case firm exposure.
Probability of shortfall.
Key drivers.
Input assumptions requiring verification.
Distinguish calculated values from user-supplied assumptions and legal assumptions. Do not opine on merits, funding enforceability, tax treatment, regulatory capital, or professional conduct unless the user supplies the rule or counsel-approved assumption to model.
Confidence Bands
High: critical inputs are user-supplied or document-sourced, formulas are shown, and no unresolved legal/tax/funding warnings sit in the critical path.
Medium: at least one critical assumption is marked assumption_to_verify, but the calculation method is auditable.
Low: specification-only model, missing distributions for stochastic outputs, inconsistent inputs, or no runnable calculation method.
Regulatory capital or fund regulation -> regulatory specialist.
Professional-conduct issues -> professional responsibility counsel.
Escalation
Stop and route when the user asks for a legal/tax/enforceability conclusion, when a novel jurisdiction is involved, when funder/firm/client interests conflict, when DBA/CFA/PACCAR questions affect the economics, or when inputs are internally inconsistent and the user has not expressly asked to model the inconsistent scenario.
Example
Model this claim portfolio under a CFA plus third-party funding structure. Show base, downside, and upside cases, then run sensitivity on damages, win probability, ATE, and time to resolution.
For a compact output pattern, see examples/output.md.
For waterfall, recourse, return-family, Monte Carlo, and audit-bundle conventions, see references/engine-model.md.
Limitations
Legal costs and funding enforceability are jurisdiction-sensitive.
The model is only as reliable as the assumptions.
Tax, accounting, regulatory capital, and professional conduct issues may require specialist advice.
