The score at a glance
The PatentIndex Score is a percentile. A score of 78 means: among comparable patents — same technology area, same filing era — this one is estimated to sit around the 78th percentile of significance. It is a single number blended from three internal measurements (“lenses”), each answering a different question about the patent.
Because it is a percentile, the score is always relative. A 55 is not a failing grade; it means “ahead of most of its peers.” By construction, the median patent in a cohort scores near 50 — and the scale runs 1–99 because a statistical estimate should not pretend to a perfect 100.
Reading the seal
- 94
90–99Field-defining
The top decile of its cohort — the one brass moment on the site.
- 72
50–89Strong
Clearly ahead of its field-and-era peers.
- 38
1–49Emerging or niche
Below its cohort's midline — often simply young.
The same seal appears wherever a score does — search results, patent pages, leaderboards.
Three lenses, one number
Internally, the score is assembled from three lenses. They are ingredients, not public sub-scores — the headline number is the only score we publish — but you should know exactly what is inside it.
Lens 1 · Impact
How much has this patent mattered?
- Forward citations — How often later patents cite this one, across the global citation graph. Counted on a logarithmic scale, so the difference between 0 and 100 citations matters far more than the difference between 10,000 and 10,100.
- Family size — How many jurisdictions the invention was filed in. Filing broadly is slow and expensive — owners do it for inventions they believe in. Log-damped, and weighted at half the citation signal.
- Claim breadth — The length of the first claim. In patent drafting, a shorter first claim generally stakes out broader protection. A rough proxy, and weighted like one — it is the smallest input.
Lens 2 · Momentum
Is it moving right now?
- Citation velocity — Citations per year of life, so a five-year-old patent gathering citations quickly can outrank a monument the field stopped citing a decade ago.
- Freshness — A term that decays smoothly with age from the filing date. Recent filings get a modest lift while the field catches up to them.
Lens 3 · Pedigree
Who is behind it?
- Inventor track record — Every named inventor's granted patents are valued with the same impact signals and rolled up into a 1–99 profile. A patent inherits the average of its named inventors' profiles.
- Granted patents only — Pending applications contribute nothing to a track record, so a reputation cannot be manufactured by bulk-filing applications.
- Why it exists: cold start — A six-month-old patent has no citations no matter how good it is. Its inventors have a history on day one. Pedigree is the signal that lets genuinely new work surface.
How they blend
The lenses are combined as a weighted sum and clamped to 1–99. Impact carries the largest share — about half the blend. Pedigree is deliberately capped at roughly a quarter to a third, for an honest reason: track records favor established, prolific inventors. That bias is useful signal — it is exactly what freedom-to-operate work cares about — but left uncapped it would quietly bury newcomers. Momentum takes the remainder. The exact weights are tuned empirically against real rankings whenever the corpus is re-scored, but those constraints hold: Impact leads, Pedigree stays capped.
One more honesty note: the lenses are not independent. Pedigree is, by construction, the rolled-up Impact of an inventor’s other patents — which is another reason it is capped rather than trusted as a separate witness.
In plain termsWhat the patent has actually done counts most. Who filed it helps a brand-new patent get seen — but can never outweigh the record.
Ranked within a cohort — the fairness mechanism
Percentiles are only meaningful against the right comparison group. Raw citation counts are grossly unfair across fields and across time: fields cite at very different rates, and citations accumulate for decades. A semiconductor-device patent filed in 1998 has had nearly thirty years to gather citations; a solid-state-battery patent filed in 2024 has had months. Rank them in one pool and every recent patent loses by default — regardless of merit.
So no score on this site is computed in a single global pool. Impact and Momentum are turned into percentiles within a cohort: patents in the same CPC subclass, filed in the same five-year window. CPC — the Cooperative Patent Classification — is the taxonomy patent offices themselves use. Its top-level sections are the colors you see on every row and card across this site:
Cohort example: H01M (batteries) × filed 2020–2024 — a 2024 battery patent is measured against its own generation of battery patents, and nothing else.
When a cohort is too thin
Some subclass × era combinations contain only a handful of patents — too few to support a stable percentile. When a cohort falls below a minimum population (a few hundred patents), it is widened step by step: from the subclass to its parent CPC section, and as a last resort to the filing era alone. The same widening applies when a cohort is degenerate — for example, a cluster of very young filings where almost nothing has any citations yet, so there is no spread to rank against.
Widening keeps every patent scorable, but the cost should be stated plainly: a score computed in a widened cohort is a lower-confidence estimate — the comparison group is coarser than we would like. The mechanism trades precision for stability, and it is the exception, not the rule.
In plain termsEvery patent is ranked against its own field and its own generation — never against a thirty-year-old patent from another industry.
A worked example
Here is the whole pipeline on one invented patent. The numbers below are made up so the arithmetic can be printed — this is not a real patent record.
Worked example — illustrative record
US 12,987,654 B2
Solid-state electrolyte separator with ceramic interlayer
Meridian Cell Systems (fictional) · H01M — batteries · filed Mar 2019 · granted Nov 2022
Collect the raw signals
412 forward citations · 9-jurisdiction family · concise first claim
Score Impact within the cohort
The log-damped signals are summed, then ranked against every H01M patent filed 2015–2019. This one lands in the 91st percentile → Impact 91.
Score Momentum the same way
412 citations over ~7 years is a strong arrival rate, though the freshness term has largely decayed. Against the same cohort → Momentum 68.
Inherit Pedigree from the inventors
Four named inventors; their granted-portfolio profiles average to Pedigree 62. Had this patent been three months old and citation-free, this is the lens that would keep it visible.
Blend and clamp
78Weighted sum — Impact counting most, Pedigree capped — clamped to 1–99: PatentIndex Score 78, “strong” tier — ahead of roughly three-quarters of its field-and-era peers.
On the live site, every input above is read from the corpus, and the score recomputes as the data refreshes.
Inventor & assignee identity
Scores about people require knowing who is who, and inventor names are genuinely ambiguous: many people share a name, and one person’s name is often spelled several ways across twenty years of filings. PatentIndex resolves inventor mentions to disambiguated person records built on USPTO PatentsView’s statistical disambiguation of the inventor record.
“Statistical” is the honest word. The overwhelming majority of patent credits link to a disambiguated person; the small remainder — typically a few percent, measured per leaderboard — fall back to conservative name-based grouping. And even good disambiguation makes mistakes: occasionally two people are merged, or one person’s work is split. When an inventor profile looks wrong, this is usually why.
Assignees — the organizations patents are assigned to — are grouped by normalized name from the patent record. That is deliberately simpler than corporate-structure resolution: subsidiaries, renames, and spelling variants of one organization may appear separately, and distinct organizations with similar names can collide. It also means the assignee shown is the assignee of record at publication — patents are bought, sold, and reassigned, and PatentIndex does not track chain of title.
How leaderboards are ranked
Leaderboards exist per CPC subclass, in three time windows: the last two years, the last five, and all-time. Windows run on the date a document first became publicly visible — its pre-grant publication, or its grant if it never pre-published. Boards cover granted patents.
- Patent boards — The all-time board is dominated by accumulated citations, as it should be. Recent boards cannot be — young patents have few citations yet — so they rank on citation velocity where any citations exist, falling back to family breadth and inventor pedigree where a patent is too young to have either. Recent boards also allow one slot per patent family, so a chain of continuation filings cannot occupy the whole board.
- Inventor & assignee boards — An identity's board score is the summed impact of its best five patents in that topic. Best-5 rewards peak work: five landmark patents beat five hundred middling ones. The recent windows show who is active now — appearing requires a recent patent in the topic — ranked by all-time standing in it.
- Full co-inventor credit — Every named inventor inherits full credit for a patent. Patents carry no first-author/last-author convention, so any splitting rule has to guess at seniority — and tested against real boards, splitting punished genuine co-inventors far more than it filtered passengers.
- Board numbers are within-board percentiles — The 1–99 on a leaderboard row is that row's standing within the displayed board (top entry = 99). It is the same scale as the site-wide score but a different comparison group, so the two will not always match.
Each board also carries its own provenance, measured at generation time: the share of its identities resolved by person records versus name fallback, and its assignee-coverage rate.
What these numbers are not
Trusting a score requires knowing its edges. These are the claims PatentIndex does not make.
- 01
Not legal advice, and not legal status. Scores, status labels, and every figure on this site are statistical estimates computed from bulk data. They are not opinions on validity, enforceability, or infringement, and they are no substitute for a patent attorney or the official USPTO record.
- 02
Term and expiry dates are estimates. Expiry is projected from the nominal 20-year term from the earliest effective filing date. It does not account for patent-term adjustment or extension, terminal disclaimers, or early lapse from unpaid maintenance fees — real terms run both longer and shorter. Verify against official records before relying on a date.
- 03
Pending applications score low as a class. Most citations arrive years after a document publishes, so pending applications carry structurally fewer of them. Cohort normalization softens this; it does not erase it. A modest score on a young application is a property of the data, not a verdict on the invention. Pending applications also contribute nothing to inventor track records (§ 02).
- 04
Identity is statistical. Inventor disambiguation and assignee name-grouping are estimates (§ 05). A profile can merge two people or split one person's work; assignee groupings can lag renames and corporate structure; and the assignee of record may no longer be the owner.
- 05
Citations are a lagging, partial proxy. Influence that never gets cited is invisible to us. Citation practices differ across fields — cohorts compensate, within limits. And counts reflect the most recent snapshot of the global citation graph, so the newest citations take time to appear.
- 06
Not market value. The score does not estimate licensing revenue, litigation odds, or commercial success. Signals that would speak to those — maintenance-fee events, litigation and PTAB history — are not yet inputs.
- 07
Scores move. A percentile is a position among peers. As the corpus grows and refreshes, a patent's score can shift without anything about the patent changing. Treat a score as a current estimate, not a permanent grade.
Nothing on PatentIndex is legal advice.
Data sources & freshness
- Patent records, citations & families
- USPTO bulk data — granted patents and pre-grant applications — with the global citation graph and patent-family data as published in Google's Patents Public Data.
- Inventor identity
- Person records built on USPTO PatentsView inventor disambiguation, with conservative name-based fallback where no person link exists.
- Refresh cadence
- Scores are recomputed as the corpus refreshes; leaderboards are regenerated in batches and stamped with their generation date.