Rankings methodology
Repleader rankings and indexes are computed from public records by a published method. They describe litigation and business activity. They are not legal advice, investment advice, or a recommendation to buy, sell or hold any security. A lawsuit is an allegation, not a finding of liability. They are not a recommendation to use or avoid any product.
Repleader publishes two separate numbers for companies in its universe: a momentum score for legal AI companies and a litigation exposure index for public companies. They are never blended. Both are labeled modeled and versioned, and every input behind them is shown with its source.
The universe
The companies come from config/universe.yaml, a founder-approved list of legal AI companies and product lines, and of public companies that are legal technology providers, litigation funders or defendants in large litigation. Every fact in the file carries the URL of the page it came from; a field that could not be verified from a fetched page is left out rather than guessed. Product lines owned by a company (for example a research product inside a publisher) carry a parent and roll up into that company: their product line, job boards, headlines and dockets count toward the parent, and they are never ranked on their own. A company leaves the universe only when it ceases to exist; a change of status, such as a distressed sale or a run-off, is published as a fact.
Momentum score, rank-1.0 (weekly)
Each component is a percentile within the universe: among the companies that have that component, a company's value is ranked and given the midrank percentile (ties share it). The score is the weighted mean of the component percentiles times 100, with the weights renormalized over the components the company has. A company with data for fewer than three components is listed as "unranked: insufficient public data", with the reason.
| Component | Private, subsidiary or nonprofit | Listed company |
|---|---|---|
| Hiring | 30% | 20% |
| Product | 25% | 15% |
| Funding | 25% | not used |
| Voice | 20% | 15% |
| Revenue growth | not used | 35% |
| Scale | not used | 15% |
| Component | Raw input | Source | Window |
|---|---|---|---|
| Hiring | Open roles on the company's public job board, summed over its boards and those of its product lines | Public Greenhouse, Lever and Ashby board APIs. A board counts only when its identity was verified as the company's (a board named like the company but belonging to another business is excluded). | Count no more than 30 days old. The 90-day change enters once 90 days of daily counts exist. |
| Product | Product and category breadth: the number of product lines the company has in the universe (its own row plus product rows it owns) plus the number of distinct categories among them | The universe file, each line linked to its website | Current universe |
| Funding | The latest disclosed round with a US dollar amount, a full date and a source, weighted by age: amount x 2^(-age in days / 365) | Company releases, wire services and trade press, as linked | 24 months. Acquisitions and control investments are shown as facts but are not rounds. |
| Voice | News headlines linked to the company | Headline, publisher, date and link from news feeds (never article text) | 90 days. Pending until headline items exist. |
| Revenue growth | Latest annual revenue over the prior year, same XBRL concept and unit | SEC EDGAR company facts | Latest fiscal year, at most 18 months old |
| Scale | Natural log of the latest annual revenue, US dollar filers only | SEC EDGAR company facts | Latest fiscal year |
Ranks are positions by score among the ranked companies of each list; equal scores share a rank.
Litigation exposure index, lei-1.0
For each matter m in which dockets name the company as a defendant: exposure_m = (I_wave / 100) x type_w x ln(1 + n_m), where I_wave is the matter's Repleader index, n_m the number of such dockets, and type_w is 3 for enforcement, 2 for mass tort, class action, securities or any MDL stage, and 1 otherwise. The index is round(100 x (1 - exp(-sum of exposure_m / 5)), 1). Each company page shows the matters, their inputs and the contribution of each, so the index recomputes from what is displayed.
Only exact name, CIK or approved-alias matches count; there is no fuzzy matching. While the matter index is unpublished (it waits for its backtest, see inertia), the exposure column shows the observed count of dockets naming the company as a defendant instead, labeled observed. While the docket table is empty it reads "pending". The public list is ordered by this number; companies with no dockets naming them are listed but not ranked.
SEC data
- Source:
data.sec.gov/api/xbrl/companyfactsfor each company with a CIK, fetched with a declared User-Agent at no more than 2 requests per second, with a conditional GET by ETag. Tickers are confirmed againstwww.sec.gov/files/company_tickers.jsononce a week. - Only annual-report values are kept: forms 10-K, 20-F and 40-F (and amendments) with fiscal period FY, full-year durations of 350 to 380 days or year-end balances, as last reported, for the latest six fiscal years. Each value links to its filing by accession number.
- Revenue: the concepts Revenues and RevenueFromContractWithCustomerExcludingAssessedTax (Revenue for IFRS filers) are both checked and the one with the latest fiscal year is used, because companies switch tags over time.
- Litigation figures (loss contingency accruals and estimates, litigation reserves, settlement payments) are shown only when reported for the latest fiscal year. Otherwise the page says "no litigation-specific reserve reported in the SEC data feed" and links the annual report. Company facts carry only values in standard taxonomies without segment detail, so amounts a company reports under its own tags or by member are not shown, and a blank does not mean zero. IFRS filers have no litigation-specific concepts in this feed.
- Three states for public companies: an SEC filer with XBRL annual data; SEC-registered without annual reports (a CIK exists, but company facts has nothing); not an SEC filer. The last two show no fundamentals and say why.
- No stock market data of any kind is requested, stored or shown.
Refresh cadence
- Weekly, Mondays: SEC company facts (conditional GET), the ticker check, then the ranking computation.
- Job-board counts and funding records come from the universe file, refreshed by its own review cycle; a count older than 30 days drops out of the score instead of going stale.
- Pages rebuild hourly from the latest computation; every page states the date its data is as of.
What is not used
No paid data, no login-gated sites, and none of G2, LinkedIn, Crunchbase or similar aggregators. No stock market data. No model writes any number or ranking.
Not investment advice, not a recommendation to use or avoid any product.