Quick Takeaway: Search results can shift hidden reconstruction work onto attorneys. Buy verifiable evidence handoffs, not hit counts. There is no universal review-time benchmark; in the hypothetical example below, removing 20 reconstruction hours saves $8,000. A strong validity challenge still does not establish freedom to operate.
jurisdiction · procedural deadline · claim version · relevant date · product exposure · evidence provenance · review capacity
Attorney review takes longer when a search report lacks exact passages, defensible dates, or stable claim mappings. The useful question is not simply, “How many hours will counsel spend?” It is, “Which hours advance the legal assessment, and which repair the handoff?”
Measure reconstruction, technical assessment, and legal analysis separately. Without that distinction, a faster search can look economical while creating more work downstream.
The Executive Decision: Buy Evidence Quality, Not Hit Counts
Why engineering relevance is not legal sufficiency
An opposition search locates and evaluates evidence for challenging granted claims under an applicable procedure. Technical similarity is only a starting point. It does not establish disclosure of every limitation, an eligible prior-art date, or an available ground of challenge.
The handoff often breaks between disciplines. Engineers identify similar mechanisms. Searchers return documents. Attorneys then reconstruct what each document supposedly establishes.
The operating objective should be time to a defensible evidence package, not time to a long results list.
This article uses “opposition search” as a workflow label. EPO opposition, US post-grant proceedings, and other challenge routes have distinct rules; the label does not make them interchangeable.
The seven variables that determine the workflow
Put the seven variables above in the scope memorandum. Each controls a different failure boundary:
- Jurisdiction: Which procedures and grounds are available?
- Procedural deadline: How much verification time remains?
- Claim version: Exactly which claims are being challenged?
- Relevant date: Which dates constrain the proposed use of evidence?
- Product exposure: What business risk makes the matter a priority?
- Evidence provenance: Can reviewers recover and scrutinize the sources?
- Review capacity: What can the team assess before a decision is required?
Keep validity challenges and freedom-to-operate analysis separate:
| Lane | Question | Accountable owner | Output |
|---|---|---|---|
| Validity challenge | Can specified granted claims be challenged on available grounds? | Challenge counsel | Evidence-backed legal assessment |
| Product clearance | Could relevant live claims cover the product in target markets? | Clearance counsel | Product-specific risk assessment |
What the final review packet must contain
Require a frozen claim set, claim chart, exact source passages, chronology, family relationships, search log, missing limitations, and unresolved questions.
Acceptance means counsel can begin substantive review without reconstructing the searcher's reasoning. It does not mean counsel agrees with the proposed challenge.
That distinction makes the handoff measurable without pretending to automate legal judgment.
Qualify the Matter Before Choosing a Search Workflow
Freeze jurisdiction, claim set, and procedural clock
Start with the official patent record, not a semantic query.
Under EPC Article 99, notice of opposition may be filed within nine months of publication of the mention of grant. The notice is not deemed filed until the opposition fee has been paid. Counsel should verify the applicable procedural requirements and current fee schedule.
US procedures differ. USPTO inter partes review guidance describes IPR challenges under Sections 102 and 103 based on patents or printed publications. Post-grant review generally permits broader grounds but has distinct eligibility and timing requirements. Ex parte reexamination follows a separate statutory framework.
Do not transfer a deadline, evidence rule, or ground of challenge from one procedure to another. Official records and governing law control the procedural decision.
Separate granted-patent challenge from clearance
A validity search tests specified claims against potentially relevant evidence. A prior-art search supports that inquiry; it does not determine infringement.
| Decision | Appropriate treatment |
|---|---|
| Use | Run opposition search against identified granted claims and a counsel-approved scope. |
| Supplement | Maintain freedom-to-operate analysis for the product, markets, and launch timeline. |
| Escalate | Ask counsel about alternative procedures when a deadline or evidence restriction changes the route. |
| Do not infer | Treat neither missing results nor one vulnerable patent as clearance. |
Expiration of an opposition period does not, by itself, establish that every other challenge route is unavailable.
Identify evidence classes the selected procedure permits
Store three separate assessments:
- Technical relevance: What does the material disclose?
- Prior-art eligibility: What date and legal basis support its use?
- Procedural admissibility: Can it support the proposed ground in this proceeding?
A result can pass one assessment and fail another. Preserve those distinctions rather than compressing them into a single confidence score.
Price the Evidence Handoff, Not Just the Search
Define acceptance before measuring cost
Count a package as accepted only when its sources, date rationale, claim mappings, and unresolved issues meet the agreed rubric.
A usable claim chart identifies exact passages and missing limitations. A document list with generated summaries is a different deliverable.
Source normalization also takes time. Family-navigation tools can assist discovery, but an analyst must still establish which publication contains the disclosure being relied upon.
Calculate cost per accepted package
Cost per accepted package
Cost per accepted package = (License + Analyst + Counsel + Verification + Rework costs) / Packages accepted for legal review
Use non-overlapping accounting categories. If verification is already included in analyst time, do not charge it twice. Define what counts as one package so the denominator cannot be inflated by splitting the same work into smaller deliverables.
Example Scenario: This hypothetical comparison is not market pricing. Both workflows produce four packages accepted under the same rubric, and all inputs are held constant except attorney reconstruction time.
| Cost input | Workflow A | Workflow B |
|---|---|---|
| Tool allocation | $5,000 | $5,000 |
| Analyst and source-verification work | $3,000 | $3,000 |
| Attorney reconstruction at an assumed $400/hour | 30 hours: $12,000 | 10 hours: $4,000 |
| Total measured cost | $20,000 | $12,000 |
| Cost per accepted package | $5,000 | $3,000 |
Removing 20 reconstruction hours saves $8,000 under these assumptions. It does not establish a typical attorney rate, predict total matter cost, or imply a better legal outcome.
The comparison isolates reconstruction. Substantive legal analysis remains separate.
Track marginal yield without claiming completeness
Review yield
Review yield = Mappings passing the handoff rubric / (Analyst hours + Counsel hours)Marginal review yield
Marginal yield for iteration k = New accepted mappings in iteration k / Additional review hours in iteration k
Contrarian operating rule: More retrieval can hurt a deadline-bound project when additional hits consume more verification capacity than they contribute usable evidence.
Keep broad discovery separate from prioritized review. A falling marginal yield can support resource allocation; it cannot establish exhaustive coverage.
Use consistent mapping units, too. Otherwise a team can appear more productive merely by dividing claim elements more finely.
Include translation, export, and source-recovery overhead
Include document recovery, translation review, export cleanup, and necessary legal-status checks in the cost model. A low-priced search can become expensive when citations cannot be reopened or exported passages lose their locations.
Measure handoff time from package receipt to substantive review readiness. Record both elapsed time and working hours: a two-day wait for a source document is not the same as two days of attorney effort.
Diagnose Failures Before They Reach Counsel
Family duplicates that masquerade as corroboration
Ten related publications are not ten independent disclosures. Normalize families while retaining publication-level differences, including added material, languages, and dates.
| Symptom | Root cause | Verification test | Owner |
|---|---|---|---|
| Apparent corroboration across many hits | Family duplication | Compare relied-upon passages and publication relationships | Analyst |
| Strong mapping disappears after amendment | Claim-version drift | Diff frozen claims against the current record | Patent operations |
| Early family date supports later disclosure | Chronology collapse | Check the exact passage in each dated document | Analyst and counsel |
| Translation supplies a missing relationship | Unchecked machine rendering | Review original wording and technical context | Qualified reviewer |
Ranking quality cannot repair these failures. The underlying record must be corrected.
Claim-version drift and context decay
A validity search becomes unreliable when a claim identifier remains unchanged but its text changes. Use immutable claim versions and flag dependent mappings for revalidation whenever a limitation changes.
Context can disappear even when the quotation survives. A spreadsheet may retain a sentence but lose the embodiment, figure, or antecedent needed to interpret it. Preserve surrounding paragraphs and explain whether the relied-upon features occur together.
Better provenance reduces reconstruction. It does not remove the need for substantive legal judgment.
Publication dates mistaken for effective prior-art dates
An earliest family priority date is not automatically the date applicable to every disclosed feature.
Verify the source document, public availability where required, the relevant disclosure, and the proposed legal basis. Earlier-filed, later-published material may receive different treatment depending on jurisdiction and ground.
The chronology therefore needs a documented rationale, not just a date filter.
G 2/21: why evidence about a technical effect needs its own lane
The EPO Enlarged Board's G 2/21 decision, dated 23 March 2023, illustrates why evidence roles matter.
The Board held that evidence supporting a technical effect for inventive-step analysis may not be disregarded solely because it was generated after the filing date. It also held that an applicant or proprietor may rely on a technical effect where the skilled person, using common general knowledge and the application as originally filed, would derive that effect as encompassed by the technical teaching and embodied by the originally disclosed invention.
The workflow error would be treating later experimental evidence as either ordinary prior art or automatically unusable.
Maintain separate roles for prior-art disclosure, support for an asserted technical effect, and counter-evidence. G 2/21 establishes neither a blanket ban on later evidence nor a universal answer to sufficiency questions.
Compare Workflow Alternatives on the Same Test Set
Public discovery interfaces and authoritative records
Evaluate alternatives using the same claims, date assumptions, time allowance, and acceptance rubric.
Public interfaces support discovery. Official registers support verification of recorded legal status and procedural events. Interface convenience does not establish authority, and an official record still requires interpretation.
Google Patents and Espacenet can support discovery and family navigation. Verify the source documents, dates, and relevant status before relying on their results.
Commercial retrieval and evidence-management tools
Do not award a platform credit because a demonstration produces plausible summaries. Test whether a reviewer can reconstruct the search and mappings from the exported record.
| Test dimension | Public tools | Commercial platform | Specialist firm | Counsel-led hybrid |
|---|---|---|---|---|
| Dates and families | Analyst must verify | Verify in pilot | Require documented treatment | Assign verification owner |
| Claim-element alignment | Usually separate work | Inspect exported mappings | Specify deliverable | Agree mapping rubric |
| Non-patent literature | Source-dependent | Verify actual coverage | Specify search scope | Allocate specialist lanes |
| Reproducibility and translation | Preserve queries and originals | Test logs and translation handling | Require methods and sources | Maintain shared record |
| Export and confidentiality | Assess each interface | Test exports and contract | Review transfer controls | Define system of record |
| Analyst and attorney effort | Measure | Measure | Measure | Measure |
Specialist search firms and counsel-led teams
Specify evidence classes, languages, exclusions, and whether the engagement includes chronology and claim mapping or only retrieval.
Ask specialists to report rejected candidates and unresolved limitations. Those records help distinguish missing disclosure from insufficient review time or excluded sources.
Also identify who resolves disagreements. A searcher's technical mapping and counsel's legal interpretation may differ without either party having made a clerical error.
Hybrid workflows with explicit handoff ownership
A hybrid search can combine internal technical knowledge, commercial retrieval, external language expertise, and counsel-led interpretation.
Assign one evidence-record owner. Otherwise, each participant may assume someone else verified dates or checked translations.
When reporting benchmark performance, use:
Benchmark recall
Benchmark recall = Relevant benchmark references retrieved / Relevant references in the adjudicated benchmark
Benchmark recall measures recovery within that benchmark. It does not establish recall across unknown prior art.
Define who adjudicates relevance and how disagreements are resolved. Include a held-out matter in the pilot to reduce overfitting to familiar examples.
Run TRACE-Loop from Claim Freeze to Review Packet
Target: freeze claims, dates, and jurisdiction
1. Freeze the target. Record the official claim text, dependencies, jurisdiction, procedural clock, and product exposure.
Give every claim version an immutable identifier. Decompose limitations while preserving relationships, sequence, conditions, and dependencies. “Sensor,” “controller,” and “threshold” are not an adequate mapping if the claim requires a particular feedback relationship.
State disputed interpretations in the scope rather than silently selecting one.
Retrieve: generate complementary search lanes
2. Retrieve across independent channels. Use lexical queries, CPC classification, backward and forward citations, semantic retrieval, and non-patent literature.
Retain lane-specific query logs. For Boolean searches, preserve field restrictions, proximity operators, truncation rules, and database syntax. The same string may behave differently across interfaces.
Separate broad discovery from the verification queue. Preserve lower-priority candidates without allowing them to displace deadline-critical checks.
Align: preserve passages and missing limitations
3. Align exact evidence. For each claim-chart row, retain the source passage, location, original language, disclosure type, and missing limitations.
Distinguish express disclosure from an asserted implicit or inherent disclosure requiring review under the applicable legal standard.
Keep novelty mappings separate from inventive-step or obviousness combinations. A combination assembled from multiple references is not the same analytical object as a single-reference novelty mapping.
Challenge: run an adversarial counter-read
4. Challenge the mapping. Assign a reviewer to test its weakest dependency:
- Does the passage disclose the claimed relationship?
- Are the features present together?
- Is the date supported for this exact disclosure?
- Does translation change a necessary condition?
- Is the proposed use permitted in the selected procedure?
Record failure codes: disclosure, chronology, translation, interpretation, or procedure.
Retrieval rank is not legal confidence.
Export: assemble the review packet
5. Export a versioned record. Use the following evidence contract as a minimum:
matter_id:
jurisdiction:
claim_version:
claim_element_id:
reference_id:
family_id:
source_url:
retrieved_at:
document_hash:
publication_date:
asserted_relevant_date:
date_basis:
exact_passage:
page_or_paragraph:
translation_source:
disclosure_type:
missing_limitations:
review_status:
reviewer:
uncertainty_notes:
Extend it with the procedural ground, evidence role, and separate relevance, eligibility, and admissibility assessments.
A document hash supports artifact identity, not historical public availability. A repository timestamp or archived copy also requires contextual evaluation.
Include the claim chart, original documents, chronology, search log, rejected mappings, and unresolved questions. Provenance must survive export, not remain accessible only inside a vendor interface.
Re-enter: search the unresolved evidence frontier
6. Convert rejection into a search delta. TRACE-Loop is a working framework, not a claim of proprietary technology.
Build a dependency graph with claims, elements, passages, documents, and date assertions as nodes. Mark reviewed relationships as supported, rejected, or unresolved. A structured table can implement the same logic; a graph database is not required.
When a source date fails, flag every dependent mapping for revalidation. When only one limitation fails, preserve the supported work and generate new queries for the unresolved relationship.
Close each iteration with:
failure reason → affected mappings → revised query → new evidence → independent review
An iteration ends when its evidence state is documented, not when the result counter reaches a target. Counsel owns the legal-review gate.
Copy the schema into the matter's system of record and assign ownership before the next retrieval cycle.
Benchmark PatentScan Against the Handoff Contract
Demand reproducible query and source records
Treat PatentScan as a candidate component, not as a substitute for the acceptance rubric.
No product capabilities, trial terms, or security controls are assumed here. Require dated documentation or observed pilot evidence for each procurement claim.
| Requirement | Pass condition | Fail or unresolved condition | Evidence |
|---|---|---|---|
| Reproducibility | Reviewer can reconstruct discovery and source selection | Query or source history missing | Logs and source records |
| Mapping | Exact passages support reviewable element mappings | Summary without traceable passage | Exported claim chart |
| Chronology | Date basis is explicit and inspectable | Family date substitutes for disclosure date | Chronology record |
| Export | Relationships and citations survive export | Manual reconstruction required | Independent reopen test |
| Confidentiality | Terms meet approved requirements | Retention or model-data use unclear | Contract and security documents |
Test reviewer reconstruction time
Give a reviewer the exported package without the operator's narrative. Time source recovery, date checking, mapping reconstruction, and correction separately.
Compare equivalent deliverables. A polished demonstration against an unprepared baseline does not establish a productivity gain.
For a small team, reducing attorney reconstruction may matter more than shortening the search session. Measure both before deciding.
Verify exports and confidentiality controls
Test whether evidence remains usable after platform access ends, subject to the applicable license and retention terms. Inspect permissions, deletion provisions, retention, model-data use, and subcontractor access where relevant.
Complete these checks before entering confidential product details. Undocumented controls remain unresolved questions, not presumed safeguards.
Require a documented failure report
The pilot report should identify missed benchmark references, unsupported mappings, date errors, export losses, and unresolved limitations.
Evaluate visible failure handling alongside successful retrieval. A vendor that exposes uncertainty gives the reviewer something to act on; an unexplained confidence score does not.
Request a scoped PatentScan benchmark using one representative matter, a fixed review budget, and the agreed handoff contract. Confirm availability and terms directly rather than assuming a free trial exists.
Commercial FAQs and the Portfolio Decision Gate
Can this replace outside counsel?
No. Search can support retrieval and evidence organization. Legal interpretation, procedural strategy, and filing decisions remain distinct responsibilities.
A better-scoped evidence package can make counsel's work more efficient. It does not turn search output into a legal opinion.
How much should we budget beyond the license?
Budget for source recovery, translation, training, security review, export cleanup, and attorney verification. Estimate these from a representative pilot rather than treating the license price as the matter budget.
Separate implementation costs from recurring matter costs, and define how unused subscription capacity is allocated.
Does a strong reference mean we can launch?
No. Freedom-to-operate analysis evaluates potentially relevant live claims against a product, jurisdiction, and commercial timeline.
Invalidating one claim or patent does not resolve other rights, related family claims, or different product configurations.
When should we stop searching and escalate?
Escalate when the procedural clock, material uncertainty, or business exposure requires a legal decision.
Falling marginal yield can justify reallocating effort, but it cannot prove completeness. Preserve the unresolved queue and identify evidence classes that remain unsearched.
Is a platform economical for a small IP team?
Compare total cost per accepted package against the existing workflow. Low matter volume may favor specialist services; repeated matters may justify tooling.
Test the actual workload. One unusually easy patent is a poor basis for an annual procurement decision.
How should semantic AI compare with manual search?
Use complementary search lanes. Semantic retrieval can surface different terminology; lexical, classification, citation, and literature searches test other paths.
Measure accepted-mapping yield and failure patterns, not similarity scores alone. Review generated summaries against the underlying source.
What must procurement require before approval?
Require reproducible queries, exact passages, date bases, family handling, usable exports, confidentiality terms, and a failure report.
A polished interface without inspectable artifacts has not demonstrated a lower-cost handoff.
For portfolio management, connect evidence maturity to exposure:
| Evidence state and exposure | Authorized next action | Accountable owner |
|---|---|---|
| Reviewable challenge evidence; high exposure | Obtain legal assessment of challenge options | IP counsel |
| Material gaps; high exposure | Target further search and parallel clearance work | IP counsel and R&D |
| Weak challenge evidence; high exposure | Evaluate licensing or redesign | Business lead with counsel |
| Limited exposure; incomplete evidence | Monitor with a defined escalation trigger | Portfolio owner |
| Low marginal value; no near-term dependency | Consider declining further spend | Portfolio committee |
Every decision needs an owner, budget ceiling, deadline, and unresolved-risk statement. Leadership should authorize a specific next action, not label the portfolio “safe.”
References & External Sources
- EPC Article 99: Opposition: Establishes the opposition period, its triggering event, and the fee requirement.
- USPTO: Inter Partes Review: Explains the IPR framework and its patent-or-printed-publication evidence restrictions.
- USPTO: Post Grant Review: Describes PGR as a distinct proceeding with its own scope and timing.
- EPO Enlarged Board: G 2/21: Provides the decision underlying the discussion of post-published evidence and technical effects.
Considering PatentScan? Use the handoff contract above to structure a pilot, verify current capabilities and terms, and measure how much reconstruction remains for counsel.
