How we score firms

Every firm page on insights.ipauthor.com carries a Quality Metrics panel comparing the firm's prosecution outcomes against the same examiner's and the same art unit's baseline performance over the trailing five years. This page explains where those numbers come from.

The baselines

• examiner_baselines_5yr — for each examiner, the 5-year allow rate, mean office-action count to first allowance, and rejection-statute mix across every firm that prosecutes in front of them. The firm-page z-scores measure how the firm's outcomes for that examiner deviate from this baseline.
• art_unit_baselines_5yr — same shape, aggregated by the USPTO art unit. Used when the examiner sample is below threshold (n < 50 resolved cases) so a single low-volume examiner doesn't dominate the score.

The aggregate

For each firm we aggregate over the firm's pending and recently-resolved applications, weighting each case by the resolved case count of the matched baseline so a 5-app outlier doesn't outweigh a 500-app baseline. This produces the firm's allow-rate-norm and OAs-to-allow-norm z-scores shown on the page.

What we deliberately exclude

• Cases where the firm is the assignee, not the prosecutor. Mapped via company_alias/company_canonical; only rows with entity_kind = 'law_firm' contribute.
• Examiners with fewer than 50 resolved cases in the 5-yr window. The art-unit baseline takes over below that threshold.
• Continuations + RCEs counted as one "case" when the examiner is the same — otherwise a re-examination pulls the OAs-to-allow metric in the wrong direction.

What's coming (2026-05-23 pivot)

The pure-numeric Wave-1 ranking has a known ceiling: a firm that wins almost-allowable cases looks great even if the win is mostly attributable to client mix. We're pivoting to a hybrid that combines these numeric scores with an LLM-judged qualitative read on the firm's actual response arguments, filtered to investment-tier matters where the stakes are high enough for the qualitative signal to be reliable. The current page reflects Wave 1 only.

Update cadence

The baselines refresh weekly; firm aggregates recompute on the same cron. The Generated timestamp on each firm page reflects when that page was last rendered against the current baselines.

Caveats

Firms that prosecute only a handful of cases per examiner / art unit will show wide confidence bands. The page exposes the underlying sample size (n=) on every tile — read the numbers with that n in mind. Allow rates near 100% or 0% are real but extremely high-variance for small n; treat them as directional, not definitive.

← Back to firms