Two scores, one snapshot
Patentability (0–100) estimates how likely your invention is to survive novelty and non-obviousness scrutiny against known prior art. Filing Readiness (0–100) estimates how well your documentation and human-inventorship record would hold up at filing time.
Four pillars
- Subject matter (§101) — Is this more than an abstract idea under the Alice/Mayo framework?
- Novelty (§102) — Is there prior art that anticipates your claims?
- Non-obviousness (§103) — Would a skilled person combine references to reach your solution?
- Documentation & enablement (§112) — Does your disclosure teach how to make and use the invention, with a strong human-inventorship record when AI assisted?
Prior art index
Each analysis compares your invention against a continuously updated index of USPTO filings, PTAB decisions, abandoned applications, examiner rejection excerpts, academic research, and open-source prior art. The index refreshes on a daily ingest cycle so scores reflect current filings — not a static snapshot from years ago. Paid Interactive Code Review adds live USPTO ODP search and a deeper 200-match overlap pass plus an extended deep report (up to 500 references).
Crucially, we don’t only learn from patents that were granted. The index includes the applications that were rejected or abandoned and the examiner office actions that explain why — down to the specific claim language and phrasing that drew §101, §102, §103, and §112 rejections. That lets us flag where your wording echoes language that has historically met resistance, not just whether a similar invention exists.
As of our May 2026 engine update, the free results page shows the closest indexed references (title, similarity %, overlap theme, and likely statute). That is the same class of output enterprise novelty tools provide — integrated into your score, not a separate PDF weeks later.
Calibration (why we avoid false confidence)
A large language model first scores each pillar. A deterministic calibration layer then applies caps when:
- Embedding similarity to prior art is high (dense §102/§103 field),
- Your code matches well-known industry patterns (e.g., LRU caches, token buckets, adaptive rate control), or
- “Human-authored” comments appear without structural evidence of conception.
You see a patentability range and a filing prep verdict (e.g., strengthen before filing) — not a single inflated number.
How we test our scores
We don’t guess and hope. Before we ship changes, we run your-style checks on real U.S. applications where we already know the outcome (granted or not). We also measure our prior-art search against references examiners actually cited — the same kind of art that drives rejections.
We tune for honest, conservative scores. If your idea looks too much like existing art, we say so. That way the number you see is one you can trust before you spend filing fees. We do not promise you will get a patent.
How we measure prior art search
Most tools fail for a simple reason: the reference isn’t in their library at all. We track how often examiner-cited art appears in our index and in your top results, then improve coverage (having the right documents) before we fuss about ranking tweaks.
Human conception signal
We weigh structure, self-reported tool use, and language patterns. Inventor narrative in comments is not treated as proof of human conception — aligned with how examiners evaluate AI-assisted applications under current USPTO guidance.
What scores are not
Scores are not legal advice, not a guarantee of allowance, and not a substitute for a freedom-to-operate or invalidity opinion from counsel. They are designed to answer: Where should I focus before I spend filing fees?