Finding art the examiner never saw. That's the whole test.
Not different art. Not better art. Better isn’t the standard. New is. An ex parte request lives or dies on whether it raises a substantial new question, which means the search starts with the prosecution record, not a keyword. We map what the examiner already saw, then search where they didn’t.
What we keep seeing before filing
Reference selection
Three references look strong on claim language. One is cumulative to what the examiner already considered. It doesn’t raise an SNQ. It just repeats the file wrapper.
NPL grounds
A conference paper covers every limitation. Proving it was publicly accessible before the priority date is a separate research problem, and it’s the one that gets skipped.
Combination grounds
“Same technical field” isn’t a reason to combine two references. The Office wants a documented engineering connection, not a category match.
Post-institution
Claims get narrowed during the proceeding. References that only covered the original broad language lose their bite the moment the amendment lands.
Lawfirms impressing clients with our help
EPR runs on a different standard than litigation.
Invalidity search asks whether art reads on the claim. EPR asks whether the Office has already seen it. Those are different questions, and they call for a different kind of search: front-loaded, prosecution-first, and built to survive the amendments that come after institution.
Invalidity / litigation search
Anticipation and obviousness against the claim as issued.
The comparison is reference-to-claim. Strength is measured by element coverage. The prosecution record is useful context, not the pass/fail line.
Ex parte reexamination search
A substantial new question against the examiner's own record.
The comparison is reference-to-prosecution-history. A reference that reads perfectly on the claim but repeats what the examiner already weighed will not institute. The file wrapper is the pass/fail line.
In the EPR requests we tracked over six months, only 16 of 227 attorneys handling the reexamination were also record counsel in the parallel district court matter for the same patent. Most EPR work is landing with a different attorney than the one running the litigation, which is exactly where a search built for the SNQ standard, not the invalidity standard, earns its keep.
Search execution
Can you point to the passage, or are you asking the examiner to read it your way?
Once the proceeding starts, no one is in the room to explain an interpretation. If an element takes a multi-step inference to map onto the claim, the reasoning may not survive review. We set a higher bar before a reference goes to claim charting: every critical limitation must be directly disclosed, textually, visually, or both, with a specific passage or figure behind it.
What we deliver
An element-by-element primary reference claim chart with specific citations to passages, figures, and tables, not paraphrased assertions. Secondary references are used only for implementation detail and minor variations, never to patch a gap the primary reference should cover.
How we execute
- 15–20 researchers per technology domain, not generalists spread across everything
- Search extends beyond patent databases into conference proceedings, standards, and vendor documentation where implementation detail actually lives
- If the best available reference still requires a stretch, we keep searching, or tell you the case is too clean for a strong search
How we've done it
The reference had "assistance data," "vector" structure, and values that changed over time. All three terms matched. The mechanism didn't.
A claim required “maintaining a persistent assistance vector with adaptive decay.” A candidate reference showed repeated bursts of assistance data with value updates over time, an easy green mark on a surface read. Looking closer: the reference’s bursts were episodic, created, used, and discarded each time, not maintained continuously. That’s repetition, not persistence. Its “updates” replaced old values outright, not reduced them by a decay function, that’s substitution, not decay. Same words, different mechanism. We don’t advance a reference on terminology overlap alone.
Mechanism-Level Reference Read
Sample output
Claim Limitation
Surface Read
Mechanism Read
"persistent…vector"
Reference shows assistance data recurring over time. Terminology matches.
YES
Data is episodic, created and discarded each burst. Repetition, not persistence. Not disclosed
"First community location register (CLR)"
Reference values change across updates. Terminology matches.
CRITICAL
Old values are overwritten by new measurements. Substitution, not an algorithmic decay function. Not disclosed
Non-Cumulative Validation Gate
Sample output
Reference
Prosecution Record Comparison
SNQ?
Centralized-repository art
Discloses a shared, centrally-managed subscriber database. This is the exact architecture the examiner already weighed when the applicant distinguished the closest prior art. Advances nothing new.
No, cumulative
Local-cache-plus-global-list art
Discloses a base station holding a local subscriber list alongside a separate global list mapping users to their home node. That specific structure was never addressed in the record the applicant argued around.
Yes
SNQ justification
The examiner already cited 20 references and reviewed a 40-reference IDS. Is your art actually new?
Finding prior art is not the hard part. Finding art the Office has not already considered is. A reference that is technically strong but cumulative to the prosecution record does not raise a substantial new question. It gets excluded, and the filing fee is spent for nothing.
What we deliver
A prosecution differentiation memo: what the examiner relied on, what the applicant argued to distinguish it, and exactly what the new reference teaches that the record does not already contain. Only references that clear this gate are advanced to claim charting. If nothing clears it, we say so before you spend money on a filing.
The three-step gate every reference has to clear
- Map the record: every IDS entry, examiner-cited reference, and related application with overlapping art
- Read the examiner’s reasoning: which limitations distinguished the art, what amendments were made and why
- Confirm new teaching: a disclosure, architecture, or combination rationale genuinely absent from the record
How we've done it
The applicant's own distinction told us exactly what a non-cumulative reference had to look like.
A patent on distributed mobile architecture overcame a prior art reference during prosecution by arguing its “Community Location Register” (CLR) was a local cache of remote subscriber data, distinguished from any shared or centralized repository, since the cited art’s nodes did not store information about subscribers registered elsewhere. A candidate reference disclosing a shared, centrally-managed subscriber database would be cumulative: the examiner already weighed that exact architecture. A reference disclosing a base station that keeps a local subscriber list plus a separate global list mapping every user to their home node, structurally close to the CLR, was never part of what the examiner or applicant addressed. That gap is the SNQ.
NPL evidence
You found the conference paper. Can you prove it was public before the priority date?
“We found it on the internet” is not proof of public accessibility. Without documented evidence of when a non-patent reference was published, indexed, and distributed, it can be excluded or discounted, and all the technical relevance in the world doesn’t matter if the date can’t be pinned down.
What we deliver
A metadata package for every NPL reference: publication timeline, archive verification (IEEE Xplore, ACM Digital Library, university repositories), and distribution evidence including dated Wayback Machine snapshots and conference attendance or circulation records: the full evidentiary chain the filing needs to survive an accessibility challenge.
How we've done it
A presentation deck predated the priority date. That alone wasn't enough.
A system covering every claimed feature surfaced in a slide deck, dated before the cutoff, but a presentation alone invites a challenge on accessibility. We went further: a pre-dated conference listing placed the system in public view, the project’s code repository showed a last-commit date consistent with the deck, and a Wayback Machine snapshot corroborated the system’s existence with the claimed features before the date. Three independent sources, one consistent timeline.
§112 Invalidity Analysis
Sample output
Priority date
Critical date line
Every evidence source below must fall before this date to count.
−14 mo
Conference proceedings listing
Session schedule and abstract archived by the conference organizer, publicly indexed.
−16 mo
Public code repository, last commit
Commit history shows the feature set implemented and visible before the cutoff.
−15 mo
Internet Archive snapshot
Timestamped capture corroborating the system’s public state and feature set.
Result: Three independent, cross-corroborating dates, not one presentation asked to carry the whole argument.
Engineering-Based Combination Logic
Sample output
Shared problem, stated in the field
SOC estimation accuracy and computational cost trade off against each other in embedded battery management hardware.
 ↓
Reference A: solves accuracy
Discloses a Kalman-filter-based SOC estimator that improves accuracy under variable load. (col. 4, ln. 12)
↓
Reference B: solves computational cost
Discloses replacing iterative computation with a precomputed lookup table to reduce processing load. (col. 5, ln. 3)
Combining them resolves both constraints at once, the direction the literature was already headed.
Obviousness grounds
Why would someone actually combine these two references, not in theory, in practice?
“Same technical field,” “addresses a design need,” “achieves a better outcome”: these are conclusions, not reasons. The Office wants a documented engineering connection: a problem stated in Reference A’s own words, a solution stated in Reference B’s own words, and a passage showing the solution was meant to travel to A’s domain.
What we deliver
A combination rationale memorandum built from specific, cited passages, not our interpretation of what the references imply. Every motivation-to-combine argument is grounded in what the references actually say, at the paragraph level, so it holds up when the patent owner challenges the combination.
What we won't submit as a rationale
- “References are in the same technical field”: a category, not a reason
- “Combination addresses a design need”: which need, documented where?
- “Combination achieves a better outcome”: better by what measure, and says whom?
How we've done it
Not "it would have been obvious to combine X and Y." A stated tradeoff, and two references that resolve it.
A battery management claim recited determining state of charge from a Kalman-filter estimate combined with a lookup-table correction. Rather than assert the combination as obvious, we grounded it in the shared engineering problem: a battery engineer at the priority date faced a known tradeoff between estimation accuracy and computational cost. One reference solved accuracy with a Kalman filter; a second reduced computation using lookup tables. Combining them addressed both constraints at once, a path the literature was already converging on, not an inference we supplied.
Post-institution strength
Patent owners narrow claims during reexamination. Will your reference still matter after they do?
A reference that covers only the broad, original claim language loses effectiveness the moment the patent owner amends to add a protocol, a parameter range, or a structural detail. Searching for references that match today’s claim is not the same as searching for references that survive tomorrow’s amendment.
What we deliver
An amendment resilience assessment: the likely narrowing strategies for this claim family, and whether the reference set already discloses the implementation-level detail, such as protocols, parameter ranges, and structural configurations, that a narrowing amendment would introduce.
How we execute
- 15–20 researchers per technology domain, not generalists spread across everything
- Search extends beyond patent databases into conference proceedings, standards, and vendor documentation where implementation detail actually lives
- If the best available reference still requires a stretch, we keep searching, or tell you the case is too clean for a strong search
How we've done it
A limitation missing from the granted claims sat in the specification, waiting.
The notice of allowance quoted a clause, a specific network-security mitigation step, that never made it into the issued claims. We treated that gap as a plausible amendment, not a dead end, and searched for art disclosing that exact limitation while the primary search was still underway. When we anticipated the patent owner might add it back during reexamination, the reference was already in hand.
Amendment Resilience Assessment
Sample output
Amendment Pattern
Example
Protocol-specific
“network communication” → “network communication using TCP”
Parameter range
“temperature control” → “temperature control between 50–75°C”
Added stage
“data encryption” → “data encryption followed by authentication verification”
Structural detail
“sensor array” → “sensor array arranged in concentric rings”
Built around the five ways EPR requests fail.
01
Primary reference strength
Element-by-element coverage with a citable passage behind every limitation. No stretches asked of the examiner.
02
Non-cumulative validation
Every reference is checked against the prosecution record before it advances. Cumulative art doesn’t get charted.
03
Public accessibility proof
A full metadata package for every NPL reference, so accessibility challenges have an answer waiting.
04
Engineering-based combination logic
Motivation to combine grounded in the references’ own words, not a category-level assertion.
05
Amendment-resilient references
References selected to survive the narrowing the patent owner hasn’t filed yet, based on the pattern in the claim family.
06
What we tell you before you spend money
If the prosecution record is too clean, or the best art we find is cumulative to it, we say so before a full search, not after a filing.
619 EPR filings in six months. Here's who's filing them.
We maintain our own tracker of EPR filing activity, covering law firms, requesters, and attorneys, refreshed against USPTO data. A few things stood out.
619
EPR filings, Jan–Jul 2026
Across 149 distinct requester law firms and 271 distinct patents or patent owners targeted.
227
Distinct attorneys handling EPR work
Scoring is done by analysts isolated from client-facing engagements. A minimum of three independent assessments precedes any recognition decision.
16 / 227
Also record counsel in a parallel district court matter
Recognition is tied to measurable, reproducible signals in patent data. Where evidence is ambiguous, no recognition is issued. There are no participation tiers.
Fastest-growing requester firms, H1 → H2 2026
Fish & Richardson PC
2 → 19 (+17)
Quinn Emanuel Urquhart & Sullivan LLP
2 → 18 (+16)
Erise IP PA
11 → 23 (+12)
Orrick, Herrington & Sutcliffe LLP
2 → 14 (+12)
Fastest-growing requesters, H1 → H2 2026
Samsung
2 → 19 (+17)
Apple Inc
2 → 18 (+16)
Heartflow Inc
11 → 23 (+12)
Micron / SK Hynix
2 → 14 (+12)
Two searches, in practice.
Anonymized. Technology domain and mechanics are accurate; identifying details are withheld.
Non-infringement
Software / Cloud
Public accessibility package
A presentation deck alone wasn't proof. Three sources together were.
A system covering every claimed feature appeared in a conference presentation dated before the critical date. A presentation alone invites a challenge, so we corroborated it: a pre-dated conference listing, a code repository with a consistent last-commit date, and an Internet Archive snapshot of the live system. Three independent timelines, all pointing the same direction.
Accessibility challenge had nowhere to land. The date was established three separate ways.
Amendment resilience
Networking / Security
Anticipated claim amendment
A clause in the notice of allowance never made it into the issued claims. Until it might.
The notice of allowance quoted a specific mitigation step involving a virtual private network that wasn’t present in the granted claims. Rather than treat that as irrelevant, we searched for art disclosing that exact limitation while the primary search was underway, on the reasoning that the patent owner could reintroduce it during reexamination.
Reference was already in hand for the exact limitation the claim family was most likely to gain.
Send a patent number. We tell you if we can help.
No long discovery process, no extended back-and-forth before you know what you’re getting.
1
EPR filings, Jan–Jul 2026
We pull the file wrapper and review the full prosecution history.
2
3–5 business days
We tell you whether the record looks searchable, what domains we’d search, and a cost and timeline estimate.
3
2–4 weeks to deliverables
Prosecution mapping and search in week one, validation and claim charting in weeks two–three, final review after.
4
You decide, either way
If the prosecution record is too clean to raise an SNQ, we tell you that before a full engagement begins.
Answered directly.
Will you find art that's merely better than what the examiner saw?
No. GreyB takes the investigation work at technical rates. The associate owns every legal decision. GreyB builds the technical record they argue from. The model is the Associate’s Associate, not the associate’s replacement.
Do you write the reexamination request itself?
No. GreyB provides technical analysis only, never legal opinions. The service sits in the same category as expert technical consulting that has been embedded in litigation for decades. E-billing language that has worked: “Technical advisory support” or “Associate-equivalent technical support.”
Can we get a preliminary read before committing to a full search?
GreyB draws exclusively from publicly available sources: USPTO records, PACER filings, PTAB proceedings, published standards, and archived public materials. Client materials provided for a specific engagement are used only for that matter. Standard NDAs are executed at the start of every engagement.
How is this priced?
You pay only for what was delivered. Scope and hours are agreed per module at the start. If the case settles before a stage is reached, that stage is not billed. Pro-rated on delivery. No retainer model. No minimum commitment.
What's the timeline?
Yes. Every module is available independently. If you need a reference triage before contentions and nothing else, that is a complete engagement. The full-lifecycle model is available for firms that want continuous technical support across the matter, but it is never required.
What technology domains do you cover?
The team has search and analysis depth in:
Wireless / 5G / 6G
Wi-Fi (802.11)
Bluetooth / IoT
Semiconductors
Consumer electronics
HEVC / AV1 codecs
Software / cloud architecture
Medical devices
Life sciences
Automotive / ADAS
Chemical / metallurgy
For domains outside these, we say so directly rather than take a matter we cannot do well.
How fast can you start?
Send the patent number and we review the prosecution history within 3–5 business days before committing to a scope and cost estimate. If we’ve already searched the same patent family in litigation, that turnaround is often faster.
What if the search comes back thin?
We tell you. Finding art is easy; finding non-cumulative art is the actual work, and not every prosecution record leaves room for it. We’d rather flag that at the feasibility stage than after a full engagement.
Questions we hear often.
We already worked with GreyB on the district court litigation for this patent family. Is EPR a separate engagement?
Yes, and that’s an advantage, not a reset. The team already knows the technology and the prior art landscape from the litigation search. EPR analysis begins with the prosecution record, so there’s real overlap in the groundwork. No cold handoff.
Our internal team already runs invalidity searches. Why is EPR different?
Invalidity search and SNQ search ask different questions of the same reference. A search built to beat a claim doesn’t automatically clear the cumulative-art bar the Office applies. It’s a distinct skill, not a bigger version of the same one.
Send us a patent number.
The fastest way to know whether an EPR search is worth pursuing is to let us look at the actual file wrapper. We’ll tell you honestly what we find.
- Response within 3–5 business days with a scope and cost estimate
- Flat fee, quoted after prosecution history review
- If the record is too clean for a strong SNQ, we'll say so before you pay for a search
- NDA executed at the start of every engagement
Start with one patent
Share the patent number and we’ll take it from there.