SPARKNETRequest a demo

Know when a patent deserves attention again.

SPARKNET continuously connects patent claims with newly emerging scientific, patent, legal, funding and ownership evidence and produces an evidence-backed review brief when meaningful change occurs.

Request a demoSee how SPARKNET works

For technology-transfer offices and patent portfolio teams. A person makes every decision.

Fictional asset EX-2019-0417Since the last review
  1. Apr 2025Scientific publication Same field, different sensing principleSet aside
  2. Jul 2025Public funding award Shares keywords, not the claimed methodSet aside
  3. Nov 2025Later patent filing Same record as one already consideredAlready seen
  4. Feb 2026Assignment recorded Transfer between known partiesChecked
  5. Jun 2026Later patent granted Describes the arrangement in claim 1, element (b)Possible connection
  6. Sep 2026Legal-status event Maintenance-fee window opensReview trigger
Review Brief issued. Review state: Needs human review
Illustrative example, fictional asset. Most new records are set aside with a reason. The line breaks only when evidence that bears on a claim survives verification.

Decisions age. Evidence changes. Attention is scarce.

Technology-transfer decisions are made at discrete points, but evidence keeps changing. SPARKNET maintains continuity between the last human decision and what has changed since, so teams know which assets deserve attention now, and why.

A patent is reviewed at one point in time. After that day, the world keeps moving, and nothing updates the decision on file by itself:

  • New publications
  • Later patents
  • Assignments
  • Legal-status changes
  • Public funding
  • Technical developments
  • Maintenance-fee events

The challenge is not retrieval. Finding records is easy. Knowing which of them should reopen a decision, and which should not, is the hard part.

The questions every review cycle has to answer, in order
  1. Last human decision
  2. What changed since?
  3. Does it matter to this asset?
  4. What argues against it?
  5. Does it deserve human attention now?
  6. Human decision
  7. New decision baseline and the cycle starts again

Why current monitoring and review fall short

Alerts report everything.
Keyword and citation alerts return volume, not judgement. Someone still reads every hit to find the one that matters.
A one-time review goes stale.
Nothing tells a portfolio team when the reasoning behind an earlier decision no longer holds.
Same topic is not significance.
A record about the same technology is not, by itself, a reason to reopen a decision. Most of them are not.

How SPARKNET works: decision continuity

One loop, from the decision on file to the next decision. Each layer keeps its inputs, so every conclusion can be retraced, and the loop closes back into the record the office already keeps.

  • Record and context
  • External world
  • Evidence intelligence
  • Challenge
  • Review state
  • Human decision
  • Decision memory

  1. Record and context

    System of record

    The case as the office already holds it in its IP management system: case state, the patent family, the previous decision and the knowledge cutoff that decision was based on.

    • Previous decision
    • Knowledge cutoff
    • Patent family
    • Review triggers: schedule, fee window

  2. External world

    • Later patents
    • Scientific papers
    • Company activity
    • Legal events
    • Public funding

    New evidence arrives between reviews and can itself start a cycle.

    Evidence intelligence

    Evidence delta

    What changed since the last review. Public sources are queried and compared with the knowledge cutoff; only records that are new and not already considered go forward. A source that cannot be checked is recorded as unchecked, never filled in.

    • New since the baseline
    • Not already considered
    • Unchecked sources stated

  3. Evidence intelligence

    Asset relevance

    Does the new record bear on this asset? It is compared with the claims, and the parties and dates are checked against the asset's own record. Relevance is a signal, not a verdict.

    • Claim
    • Entity
    • Time

  4. Challenge

    Challenge

    A deliberate attempt to show the record does not matter. Only what survives moves on, and the reasons against it travel with it into the brief.

    • Counter-evidence
    • Duplicate
    • Stale
    • Mismatch

  5. Review state

    Review priority and review state

    Deterministic, recorded rules assign the state. Calibrated AI signals are inputs to those rules, never the decision. The result is an evidence-backed Review Brief.

  6. Human decision

    Human review

    A person reads the brief and decides. This is the final decision. SPARKNET recommends a state; it never acts on a patent by itself.

    • A person decides

  7. Decision memory

    Decision memory and new baseline

    The decision, its reasons and the evidence it considered are recorded and written back as the new baseline. The next cycle starts from what a person last decided, so evidence already considered is not raised again.

    • New baseline
    • Review history

The loop closes: the new baseline goes back into the system of record, and the cycle repeats whenever the world changes: on a schedule, when a fee window opens, or when a new record appears.

Example scenario — fictional asset

Dormant patent Wake-Up

Wake-Up is one use case of the broader question: which patent deserves attention now, and what changed since the last decision? Here the same loop runs on an asset that has been quiet since its last review.

  1. System of recordPreviously reviewed, quiet assetAsset EX-2016-0291, a coating process. Reviewed in 2023 and left to monitor. No activity since.
  2. External worldNew external evidence appearsA later patent, granted to a company with no recorded tie to the holder, describes the claimed coating step.
  3. Evidence delta · Asset relevanceA meaningful delta is detectedDated after the last review, not considered before, and it bears on claim 4.
  4. ChallengeThe challenge survivesNot a duplicate, not stale, not a known licensee. The counter-evidence found is listed in the brief.
  5. Review state · Human reviewBack in the Review QueueNeeds human review A Review Brief is issued. A person decides what happens next.

Evidence, not answers

SPARKNET reads five kinds of public evidence. Language models help read and compare it, but every statement in a brief points to a stored source record. A statement without one does not appear.

Scientific literature
New publications that describe the claimed technology, or something close to it.
Later patent records
Filings and grants after the baseline, including the patent's own family.
Legal-status information
Status changes, lapses and maintenance-fee events.
Assignment and ownership data
Recorded transfers that change who holds or uses the technology.
Public research funding
Awards that indicate new activity in the claimed area.

Source record

Later patent record, granted June 2026

Stored as
Snapshot of a public record
Retrieved
14 Sep 2026
Fingerprint
sha256 4f1c9a07…e07b
Cited passage
Column 6, lines 12 to 19

“…interleaved electrode pairs sampled at a reduced duty cycle to extend battery life…”

Automatic check: the passage is present in the stored snapshot

Illustrative record. Each source is kept as a snapshot with its retrieval date and a content fingerprint, so a statement can be traced to the exact text it relies on, even if the public page later changes.

The Review Brief

When meaningful change occurs, SPARKNET produces one brief for the asset. The reasons for renewed attention and the reasons for caution sit on the same page, what could not be verified is stated rather than guessed, and the decision is left to a person.

Illustrative example — fictional asset and organisations

Review brief

EX-2019-0417 Capacitive soil-moisture sensor array with low-power polling

Holder
Northmere Institute of Technology (fictional)
Why now
A later patent granted to a third party describes the claimed electrode arrangement, and a maintenance-fee window has opened.
Last reviewed
14 Mar 2025 · decision on file: keep and monitor

What changed

Two new records since the last review: a patent granted to Corvel Instruments (fictional), filed after this patent's priority date, and a university publication on field-scale soil sensing. R1 R2

Why it may matter

The granted patent describes interleaved electrodes polled at a low duty cycle, the arrangement recited in claim 1, element (b). R1

“…interleaved electrode pairs sampled at a reduced duty cycle to extend battery life…”

Found in the source R1 column 6

What argues against it

  • The publication measures resistance, not capacitance: same field, different mechanism. R2
  • Corvel may be a licensee of the holder. The records do not establish this either way.
  • AI suggestion, not a finding: the similarity may come from shared background in low-power sensing.

Already considered

  • The inventors' 2018 conference paper, considered at the last review.
  • One family member in another jurisdiction, already on file.

What cannot be verified

  • Whether any Corvel product practises the claim. No product documentation is public.
  • Legal status in one family jurisdiction. The register did not respond. R3

Suggested review

Needs human review

  • Confirm whether Corvel is a known licensee.
  • Decide whether the new grant changes the licensing plan before the fee is paid.

Human action

Pending reviewer decision

No decision is recorded yet. SPARKNET suggests a review; a person records the outcome, and that record becomes the new baseline.

Evidence and provenance

RefRecordRetrievedFingerprint
R1Later patent record, granted Jun 202614 Sep 20264f1c9a07…e07b
R2Scientific publication, May 202614 Sep 2026b82d3e51…91a0
R3Legal-status register, one jurisdictionCould not checkNone stored

How to read a Review Brief

  1. It starts from the last decision. Why now and Last reviewed say what the brief is measured against.
  2. Every statement cites a stored record. References point to snapshots, not to live pages.
  3. Caution gets the same space as support. What argues against the change is collected deliberately.
  4. Gaps are stated, not filled. A source that did not answer is shown as unchecked.
  5. Machine readings are marked. An AI suggestion is labelled as such and never presented as a finding.
  6. A person decides. The brief ends with a suggested review and a pending human action, never a conclusion.

How SPARKNET is evaluated

Two lines of evidence: an external public benchmark for the relevance step, and an expert evaluation of the review decision itself, now in preparation.

Public benchmark: PatentMatch

Derived from European Patent Office search reports

PatentMatch pairs patent claims with passages that examiners cited against them. On it, PaECTER, a language model trained on patent citations, is a useful technical-relevance signal: it separates passages about the claimed technology from passages that are not.

The same evaluation shows its limit. Neither PaECTER nor any other system tested reliably tells a novelty-destroying passage from background on the same topic. Technical relevance is not novelty, and it is not, on its own, a reason for review.

That is why SPARKNET keeps relevance, verification and the human decision as separate steps. PaECTER informs relevance only. It does not predict novelty and does not decide what is reviewed.

Expert evaluation

Expert validation study in preparation

Whether a change justifies renewed attention is a professional judgement. SPARKNET will be evaluated with qualified technology-transfer professionals on a small, representative set of real cases, comparing decisions made with a SPARKNET Review Brief against simpler workflows.

  • Qualified technology-transfer reviewers
  • Comparison with conventional and simpler AI-assisted workflows
  • Decision quality, time, evidence reliability and handling of uncertainty
  • Frozen evidence packets, so every judgement refers to the same material
  • Blinded, independent judgements with auditable records

Results will be reported when the study is complete.

What already works

Prototype capabilities, demonstrated end to end on real public patent cases. These describe a working prototype, not a production service.

Scheduled review triggers
Reviews start on a schedule and when a fee window opens, computed from the recorded rights.
Event-triggered review
A new public record can open a review between scheduled cycles.
New-evidence detection
Records are measured against the last review. What was already considered is not raised again.
Technical relevance
Each new record is compared with the patent's claims as a relevance signal.
Hard-negative suppression
Records on the same topic but a different mechanism, and duplicates, are set aside with their reason.
Evidence-backed Review Brief generation
Every statement in the brief carries a reference to a stored source.
Persistent review history
Decisions are recorded, and the next cycle starts from the new baseline.

An evidence-grounded review system

SPARKNET is an evidence-grounded patent intelligence and review system. Its relevance step is evaluated on an external benchmark, an expert evaluation with technology-transfer professionals is in preparation, and its outputs are assembled from stored, citable records rather than free-form model answers.

What SPARKNET does not do. SPARKNET does not determine patent validity, novelty, infringement, licensing value or commercial success. It supports the people who make those judgements.

Request a SPARKNET demo

See a full review cycle on a real public patent, from trigger to Review Brief.

Request a SPARKNET demo

Or write to [email protected]

SPARKNET demo