How we modeled this
Modeled estimates, September 2026
The outcome matrices on this site are an estimate based on public research. They are modeled, not measured: no study has yet measured the effect of expert-witness simulation on case outcomes, ours included. This page traces every figure to its source or marks it as an assumption, and says how strong each one is.
The model
Each matrix models a close, contested matter in which both sides start at 50/50. The effect on the chance of winning is the share of the outcome decided by the expert fight, multiplied by the improvement in that fight from earlier, better-tested expert preparation. The shift is applied in log-odds, so it shrinks near 0 and 100 percent. Plaintiff-side use counts one and a half times defense-side use, because plaintiff experts draw most Daubert challenges. Every matrix uses the low end of the assumed improvement.
Expected recovery is the chance of winning times the value if won, adjusted by a small damages-size effect, plus the chance of losing times the value if lost. Defense payout is the same settlement money seen from the other side, so the figures are identical and only the reading is reversed. Defense legal fees are excluded.
What the figures do not show
- Real-world base rates. The matrices use a close 50/50 case. At real rates the effects are smaller: the securities plaintiff-only recovery lift falls from about +10.7 percent to about +5.7 percent at an 80 percent certification rate.
- Both sides using it. If both sides prepare this way, most of the edge cancels, leaving roughly +1.2 to +2.4 points.
- Blended cases. The general civil and commercial matrices model a blended matter rather than one practice’s published data, so they rest on wider assumptions than the securities and antitrust matrices.
- Selection. Trial and certification win rates describe cases that were not settled, so they overstate the odds in the median filed case.
Inputs used across every matrix
| Input | Value | Source | Strength |
|---|---|---|---|
| Improvement in the expert fight | +8 to 12 points | Our assumption. Anchored to Schwarcz et al., AI-Powered Lawyering (2026), a randomized trial finding reasoning models significantly improved the quality of legal work on most tasks. That trial measured work quality, not case outcomes. | Weak |
| Plaintiff to defense weighting | 1.5 to 1 | PwC Daubert studies: about 67 to 70 percent of challenges to financial experts target plaintiff-side experts. Peruzzi (2024): plaintiffs’ experts drew about 71 percent of challenges to antitrust economists. | Moderate |
| Damages-size effect | ±2 to 3 percent | Our assumption: partial-exclusion risk times the average damages cut times an assumed reduction in that risk. | Weak |
How much of the outcome the expert fight decides
All four shares are our assumptions, each argued from published evidence.
| Category | Share | Reasoning and evidence |
|---|---|---|
| Securities | 0.5 | After Goldman (2021), price-impact fights at class certification are expert battles, and defendants challenge price impact more often, increasingly with expert reports. In Rocket Companies (E.D. Mich. 2024) the court found the defense expert’s findings “largely dispositive” in denying certification. |
| Antitrust | 0.6 | The overcharge regression is at once the proof of common impact, the damages proof and the settlement anchor (Comcast v. Behrend). Of 286 Daubert challenges to economists from 1993 to 2021, 36 percent led to full or partial exclusion (Peruzzi, 2024). |
| Commercial | 0.35 | Experts drive the size of damages more than liability. PwC found a 44 percent full or partial exclusion rate for financial experts in breach-of-contract and fiduciary cases. |
| General civil | 0.3 | A blend of expert-heavy areas such as medical malpractice and product liability with expert-light ones such as employment and simple contract. |
Securities class actions
| Figure | Value | Source | Strength |
|---|---|---|---|
| Plaintiff-style damages | $490M | Cornerstone Research, 2025 review: median for cases settling after a certification motion was filed. | Strong |
| Settlement if certified | $39M | Cornerstone, 2016 to 2025: cases settling after a certification ruling settled at a median 8.0 percent of plaintiff-style damages, consistent with the $38M median post-motion settlement. | Strong |
| Value if certification denied | $6M | Proxy: Cornerstone’s 2025 median for cases settled before a certification motion. No public dataset tracks outcomes after denial. | Moderate |
| Real-world certification rate | about 80% | Duane Morris Class Action Review 2026: 79 percent (26 of 33) granted in 2025. NERA: 81 to 86 percent of decided motions granted. | Strong |
| Settlement variation explained by case structure | about 75% | Cornerstone regression factors explain about 75 percent of settlement variation, mainly case size, which caps any preparation effect. | Strong |
Antitrust class actions
| Figure | Value | Source | Strength |
|---|---|---|---|
| Single-damages overcharge | $300M | Illustrative case size, above the median and at leadership-counsel scale. | Assumption |
| Settlement if certified | $60M (20%) | Connor and Lande, Not Treble Damages, 100 Iowa L. Rev. 1997 (2015): dollar-weighted average recovery about 19 percent of single damages; median 37 percent. | Moderate |
| Value if certification denied | $5M | Our assumption. | Weak |
| Real-world certification rate | about 77% | Duane Morris Class Action Review 2026: 77 percent (17 of 22) in 2025; 71 percent in the first half of 2026. | Strong |
Commercial litigation
| Figure | Value | Source | Strength |
|---|---|---|---|
| Claim size | $20M lost profits | Illustrative. | Assumption |
| Value if won | $8M (40% realized) | Our assumption, bracketed by CRA findings that large trade-secret awards were cut about 42 percent after trial or on appeal. | Weak |
| Value if lost | $1M | Our assumption: nuisance or cost-of-defense value. | Weak |
| Real-world win rate | about 60% | BJS, Civil Bench and Jury Trials in State Courts, 2005: plaintiffs won 66 percent of contract trials. CRA and Lex Machina 2025: trade-secret claimants won 59 percent of combined pre-trial and trial judgments. | Moderate |
Civil litigation in general
| Figure | Value | Source | Strength |
|---|---|---|---|
| Value if won, if lost | $600K, $50K | Illustrative blended matter. | Assumption |
| Real-world win rate | about 56% | BJS 2005: plaintiffs won 56 percent of state general civil trials. | Moderate, dated |
| Trial rate | 0.4% | Administrative Office of the U.S. Courts, Table 4.10, FY2025: share of federal civil terminations during or after trial. | Strong |
How these figures will be replaced
- A blind backtest. Run from the complaint alone on 50 to 100 resolved securities and antitrust class actions, then score the output against the opposing expert reports actually served and against the Daubert and certification rulings.
- A matched-matter study. Compare matters that used Supreme Mind with similar matters that did not, matched on circuit, damages size, lead plaintiff and defendant size.
- A controlled preparation trial. Randomize mock cross-examination preparation and have blinded evaluators score the resulting testimony.
Measured results will replace these modeled figures, here and on every page, once they exist.
Sources
Each opens in a new tab.
- Cornerstone Research, Securities Class Action Settlements: 2025 Review and Analysis (Feb. 2026).
- NERA Economic Consulting, Recent Trends in Securities Class Action Litigation: 2024 Full-Year Review (Jan. 2025).
- Duane Morris LLP, Duane Morris Class Action Review 2026, including the antitrust certification data.
- John M. Connor and Robert H. Lande, “Not Treble Damages: Cartel Recoveries Are Mostly Less Than Single Damages,” 100 Iowa L. Rev. 1997 (2015).
- Huntington Bank and the UC Law SF Center for Litigation and Courts, 2025 Annual Antitrust Report: Class Actions in Federal Court (2026), and the annual reports before it.
- Edoardo Peruzzi, An Empirical Study on Daubert Challenges to Economists in Antitrust Litigation (SSRN working paper, 2024).
- PwC, Daubert Challenges to Financial Experts studies (2000 to 2021).
- U.S. Bureau of Justice Statistics, Civil Bench and Jury Trials in State Courts, 2005 (2009).
- Lex Machina, Trade Secret Litigation Report 2026; CRA International, Trade Secret Litigation Watch (Feb. 2026).
- Administrative Office of the U.S. Courts, Judicial Facts and Figures, Table 4.10 (FY2025).
- Samuel R. Gross, “Expert Evidence,” 1991 Wis. L. Rev. 1113 (1991).
- Daniel Schwarcz et al., “AI-Powered Lawyering: AI Reasoning Models, Retrieval Augmented Generation, and the Future of Legal Practice,” Journal of Law and Empirical Analysis (2026).
- Fideres, research on price-impact challenges after Goldman (2021 onward).
- Halliburton Co. v. Erica P. John Fund (2014); Goldman Sachs Group v. Arkansas Teacher Retirement System (2021); Comcast Corp. v. Behrend (2013).