Refolk
FrameworkMarket and talent intelligence

The Competitor Patent-Cluster Read: Live Program, Defensive, or Abandoned

Score any competitor patent cluster on six dimensions and reach a Live Program, Defensive, or Abandoned verdict that a second analyst reproduces blind.

24 min readLast reviewed November 4, 2026Read as Markdown

Key takeaways

  • Only 3% of triadic patents are purely preemptive, so Defensive is a verdict you need evidence for. It is not the default explanation for a quiet cluster.
  • A competitor's filings stay invisible for about 18 months after the priority date, and country breadth stays unknown until 30 or 31 months, so young families score U (unknown), never 0.
  • Only 41.4% of US utility patents granted in 2014 paid all three maintenance fees, which makes a paid third fee one of the strongest dated signals that an owner still values a cluster.
  • In Refolk's index, only 14 US senior research and engineering profiles mention "inventor patents" in their headline, so the inventor roster has to be built from the patent's inventor field by matching name plus employer.
  • Across 36,204 matched inventor-patent pairs, patents left behind by a departing inventor were maintained less often, but the effect weakens when the inventor joins a competitor, so always record where each leaver went.
  • The verdict comes from two layer scores out of 6: a high patent layer with a low people layer reads Defensive, both high reads Live Program, and both low reads Abandoned once revivals and live child applications are ruled out.

This guide is for the strategy, research or talent-intelligence analyst who has been handed a cluster of a competitor's patent filings in one technical area. Leadership wants a blunt answer: is the competitor building this, protecting it, or finished with it? The guide scores the cluster on six evidence dimensions, three from the patent record and three from the people behind it. It ends with one of three verdicts (Live Program, Defensive or Abandoned) and a dated confidence note that a second analyst can reproduce without seeing your work.

What do Live Program, Defensive, and Abandoned mean?

A Live Program is a cluster the competitor is still paying to protect and still staffing to build. A Defensive cluster still gets money, but no one visible is building in it. An Abandoned cluster has lost both the money and the people, once you allow for the period in which new filings cannot be seen yet.

Terms used throughout:

  • Patent cluster: a set of a competitor's filings grouped by CPC code (Cooperative Patent Classification, the shared scheme patent offices use to tag subject matter) or by a claim theme you define.
  • Patent family: every filing that claims the same original invention, across patent offices and continuations.
  • Priority date: the earliest filing date a family claims. Almost every clock in this guide starts from it.
  • Continuation: a later US application that claims the same priority as its parent, usually to pursue different claims. A divisional does the same after the patent office splits the claims.
  • Triadic patent: an invention filed at the US, European and Japanese offices. Most of the research on commercialization uses this group.

Why three verdicts and not two

Base rates show that most patents get used and that purely defensive filing is rare. In the RIETI-Georgia Tech inventor survey (Nagaoka and Walsh), about 60% of triadic patents were commercialized, at similar levels in the US and Japan. A US follow-up presented by John Walsh found 55% commercialized. It also found 17% not commercialized but held at least partly to preempt others, and only 3% held purely to preempt. A machine-learning estimate across decades of US grants put the mean probability of commercialization at 0.59 and of preemptive use at 0.19. A separate study found that a substantial share of patents, around 40%, remained unused.

3%
Share of triadic patents that are purely preemptive
Another 17% are partly preemptive, so most defensive behaviour is mixed in with real use.

This has two consequences. First, "they are just filing defensively" is a common explanation that the evidence does not support as a default. You need evidence to reach Defensive, just as you do for Live Program. Second, most clusters are mixed: some families get commercialized, some are held to block others, and some are forgotten. The verdict names the main pattern, and the confidence note records how mixed the cluster is.

One limit to state up front: no published study measures how defensive filings differ from product filings in continuation counts or claim breadth. You cannot identify a defensive filing from the patent alone. This framework therefore infers Defensive from a mismatch: the patent layer shows money still going in, and the people layer shows no one building.

When can a competitor's patent cluster be read at all?

You cannot see a filing until about 18 months after its priority date, and you cannot tell which countries it went to until 30 or 31 months. I call anything younger the blind window. Reading a cluster as quiet when its newest work is still in that window is the most common mistake in this work.

RouteEarliest visibility
US normal~18 months from priority
US nonpublicationAt grant only
PCT publication~18 months from priority
PCT to US, CN, JPBreadth known at 30 months
PCT to EPO, KR, IN, UKBreadth known at 31 months

How the timing works:

  • US publication. Most US utility applications are published promptly after 18 months from the earliest priority date they claim (MPEP 1120).
  • Nonpublication requests. An applicant can keep a US application unpublished only by certifying that the invention has not been and will not be the subject of a foreign or PCT filing that publishes at 18 months. The request must be made when the application is filed. A US-only filing on this route stays invisible until grant. No published figure shows how many competitor filings use it, so you cannot correct for it. You can only record it as a possibility in the confidence note.
  • PCT national phase. The PCT treaty sets national-phase entry at 30 months from priority (Articles 22 and 39), and national law may allow longer. In WIPO's table, 76 offices use 30 months and 51 use 31. The EPO allows 31 months under Rule 159(1) EPC. The US deadline is 30 months and cannot be extended.

When each part of a family becomes readable

  1. Priority filing
    Clock starts; nothing is public yet
  2. Publication, ~18 months
    Family becomes visible unless a nonpublication request was filed
  3. National phase, 30 months
    US, CN and JP entries known; the US deadline cannot be extended
  4. National phase, 31 months
    EPO, KR, IN and UK entries known; breadth can now be scored
  5. Grant and fees
    Fees due at 3.5, 7.5 and 11.5 years each record a dated decision by the owner
A patent family moves from invisible to fully readable in stages, and each stage unlocks a different scoring dimension.

What are the six evidence dimensions?

The six dimensions sit in two layers. Three patent-layer dimensions show whether the competitor is still spending on the area, and three people-layer dimensions show whether anyone is still building in it. Score each one 0, 1 or 2, or U when the evidence cannot exist yet.

DimensionWhat it provesHow it lies
1. Filing cadence and continuationsOngoing drafting spendBlind window or nonpublication hides new filings
2. Prosecution and fee statusOwner still values the rightsLapses get revived or moved into a child application
3. Family and national-phase breadthWillingness to pay for marketsProcedural splitting inflates member counts
4. Named-inventor retentionKnow-how is still in the companyName collisions; leavers who went to a rival
5. Inventor-team growthHeadcount is being addedQuery drift between snapshots
6. Off-patent corroborationWork is happening outside the legal recordA single posting; unproven vendor lead times

Dimensions 1 to 3 add up to the patent layer score (0 to 6). Dimensions 4 to 6 add up to the people layer score (0 to 6). Score U as 1, the midpoint, so that an unknown does not push the verdict either way, and count every U, because each one lowers your confidence.

The rubric below is the one I use. "Most recent visible year" means families with priority dates between 18 and 30 months before the as-of date. That is the newest work you can actually see.

Six-dimension scoring sheet
CLUSTER: <CPC code or claim theme>   AS-OF DATE: <YYYY-MM-DD>   ANALYST: <initials>

PATENT LAYER
1. Filing cadence and continuations
   2 = new families, continuations or divisionals in the most recent visible year
   1 = families present, but no new filings or children in the most recent visible year
   0 = no new families or children since an earlier burst, blind window cleared
2. Prosecution and fee status
   2 = fees paid across the cluster, including any second or third fee; applications actively prosecuted
   1 = mixed: some lapses or abandoned national phases, some live
   0 = lapses or abandonments across most families, no revival petition, no live child
3. Family and national-phase breadth (count distinct offices, not members)
   2 = national phase entered at several distinct offices
   1 = priority country only, but several priority filings or a long first-to-last timespan
   0 = single office and single filing, or national phases abandoned at 30/31 months
PATENT LAYER TOTAL: __ / 6    U count: __

PEOPLE LAYER
4. Named-inventor retention (core inventor = named on 2 or more families)
   2 = most core inventors still at the competitor
   1 = roughly half stayed, or most leavers went to a rival
   0 = most core inventors left, and not to a rival
5. Inventor-team growth (identical query, prior snapshot)
   2 = matched headcount up
   1 = flat
   0 = down
6. Off-patent corroboration
   2 = several roles across functions in one quarter, or a posting plus a repo or paper naming the technique
   1 = one isolated signal
   0 = nothing after a full search
PEOPLE LAYER TOTAL: __ / 6    U count: __

Score at the cluster level after reviewing every family; write U where the blind window rules out evidence, and count U as 1.

Turning two layer scores into a verdict

The verdict comes from where the two layer scores land on a simple grid. For each layer, 0 to 3 counts as low and 4 to 6 counts as high.

Patent spend against people activity

People layer high (4 to 6)People layer low (0 to 3)
Live Program, hidden or early
Check the blind window and nonpublication; confidence no higher than medium
Live Program
Funded and staffed; brief leadership with the dated note
Abandoned
Rule out revival petitions and live child applications before calling it
Defensive
Money still going in, no builders; check where the inventors went
Patent layer low (0 to 3)Patent layer high (4 to 6)
The verdict comes from comparing what the competitor pays for with who is still building, not from either layer alone.

The rules in words:

  • People high, patent high: Live Program.
  • People high, patent low: Live Program with confidence no higher than medium. The newest filings may be in the blind window or under a nonpublication request.
  • People low, patent high: Defensive.
  • People low, patent low: Abandoned, but only after revival petitions and live child applications have been ruled out. Until then, confidence is low.

Confidence is set separately:

  • High: no U scores, neither layer sits at 3 or 4 (the boundary), and the agreement check below passes.
  • Medium: one U, or one layer at the boundary.
  • Low: two or more U scores, or the two layers disagree and you cannot explain why.

The agreement check comes from the research on inventor departure covered below. When inventors leave, their patents are maintained less often, so the fee record and the inventor roster should normally point the same way. Inventors all staying while fees lapse, or inventors all leaving while every fee is paid, is a mismatch worth writing down. Sometimes it has an explanation, such as inventors moving to a rival, which makes it rational for the owner to keep the patents. If you cannot explain it, it lowers your confidence.

How do you read prosecution, fees, and family breadth?

Treat every fee payment and every national-phase entry as a dated decision by the owner that the rights are still worth money. The later and more expensive the decision, the more it tells you.

Maintenance fees

US maintenance fees fall due at 3.5, 7.5 and 11.5 years after grant, each with a six-month grace period. If a fee is not paid by the end of the grace period, the patent expires at that point. USPTO Patent Center states whether a patent has expired and gives the expiration date. Fees rise at each stage: in the large-entity schedule I reviewed, the third fee was nearly four times the first. Fee amounts change, so check the current USPTO schedule rather than relying on a figure you remember.

Grant yearPaid 1st feePaid 2nd feeFull term
200085.7%68.7%51.1%
200587.5%69.3%46.0%
201086.6%65.9%43.1%
201485.8%64.6%41.4%

These survival rates come from a secondary source. Check them against USPTO data before quoting them to leadership.

How a US grant cohort thins at each fee

  1. Granted
    100%

    2014 US utility cohort

  2. Paid 1st fee
    85.8%

    Due at 3.5 years

  3. Paid 2nd fee
    64.6%

    Due at 7.5 years

  4. Paid full term
    41.4%

    23.2 points lost at the third fee

For the 2014 US grant cohort, the largest drop comes at the third fee, which is why that payment is the most informative.

The same source reports that patents which have already paid two fees pay the third at a 63% rate. That conditional rate is well above the 41.4% rate for the cohort as a whole, because each payment filters out patents the owner has stopped valuing. In practice, a competitor that keeps paying second and third fees across a cluster is making repeated, expensive, dated decisions that the area matters. A cluster that drops at the first fee tells you something else.

Lapses are not final

A lapse is not the same as walking away. The USPTO Director can accept a maintenance fee after expiration if the owner shows the delay was unintentional. The owner may also have moved the claims into a continuation that is still pending. At the PCT stage, a US national entry is considered abandoned once the 30-month limit passes without the required copy and fee. Before you score dimension 2 as 0, check every lapsed family for a revival petition and a live child application.

My sources did not establish two things: the specific status codes for US abandonment after a final rejection, and how EPO renewal-fee lapses appear in the European Patent Register. Until someone on the team has read the field definitions in Patent Center and the EPO Register, record the raw status text alongside your coding, so a second analyst can check how you read it.

Family breadth

The OECD's Measuring Patent Quality treats family size as an established proxy for value, and family size for EPO patents mostly ranges between 1 and 9. But Dechezleprêtre and colleagues found that inventions never patented abroad make up 80% of patented inventions worldwide. A family filed only in its priority country is normal and is not, on its own, a sign of neglect.

Three rules for scoring breadth:

  • Count distinct offices, not family members. Procedural splitting or a patent thicket (a dense web of overlapping filings) can inflate member counts without any intent to enter more markets.
  • For young clusters, use the priority-country count and the family timespan. The number of filings in the priority country and the time between a family's first and last filings are both strongly correlated with other common measures of patent value, and you can read both before 30 months.
  • Do not read breadth as a launch date. No published case study shows that continuation counts or jurisdiction counts predict a launch. They are value signals, not launch signals.

Are the competitor's named inventors still there?

Find out by matching every named inventor on the cluster to a current role on name plus employer, marking each one stayed, left or unknown, and recording where each leaver went. Do not expect inventors to label themselves. Keyword search will not find them.

14
US senior research and engineering profiles with "inventor patents" in the headline
Across Research Scientist, Principal Engineer, Staff Engineer and Principal Scientist titles in Refolk's index; Qualcomm leads with 2.

The roster has to start from the inventor field on each patent. Work through it in this order:

  1. Deduplicate the names. The same person often appears with and without a middle initial, or with a transliterated spelling.
  2. Match on three things. Require name plus prior employer plus a matching location or technical field. A name alone produces false matches.
  3. Mark status. Mark each inventor stayed, left or unknown. Unknown is an honest answer and counts toward U.
  4. Record the destination. For every leaver, note where they went and whether that employer competes in the same area.

Why the people layer matters: the "Nest without birds" study in Research Policy used 36,204 matched inventor-patent pairs. It found that when an inventor leaves, their patent is less likely to be maintained than a comparable patent whose inventor stayed. The suggested mechanism is lost tacit knowledge, the practical know-how that never makes it into the claims, and the effect is stronger for complex inventions. It weakens when the inventor joins a competitor, which is consistent with firms keeping patents they might need to use against the rival. Hence the destination rule: "inventor left" only points to Abandoned when the inventor did not go to a rival.

One thing nobody has measured: there is no published lead time for how long related filings take to go quiet after a departure. Do not promise leadership that the filings will stop within some number of quarters.

The matching step usually takes a talent-intelligence analyst one to two days, because each inventor needs a three-part match. Refolk cuts this down: you describe the people in plain English, by employer, patent history and whether they stayed or left, and get matched profiles from LinkedIn, GitHub and the open web that you can check against the three-part rule.

Measuring team growth

Count current employees at the competitor whose skills match the cluster's theme, then compare that count with an earlier snapshot. The count only means something if both snapshots used the same definition. Keep the query text identical, keep the depth of the search identical, and record the date. In Refolk, that means saving the plain-English query and re-running it word for word each quarter, so any change in the count reflects hiring and not a reworded search.

What off-patent evidence corroborates a filing cluster?

Job postings, public repos, conference papers and regulatory filings that name the same technique can corroborate a cluster. None of them has a measured lead time, so treat them as confirmation that work is happening, not as a countdown to launch.

What the evidence supports:

  • Job postings. The only lead times in circulation are vendor claims with no published method: hiring signals showing up 6 to 18 months before a product or go-to-market launch, or 4 to 8 months before a feature ships. Use postings to confirm activity, not to time it.
  • Public repos. Engineers at the competitor publishing code on the technique the cluster claims, for example solid-state battery anode coatings for a battery cluster, are strong corroboration when the authors match names on your roster.
  • Papers. Conference and journal papers that list the cluster's inventors as authors tie the legal record to ongoing research.
  • Regulatory filings. These can name a technique before launch, but no measured lead time exists for them either.

How to score it: a 2 needs several roles across different functions in the same quarter, or a posting plus a repo or paper that names the technique. One isolated signal scores 1. Nothing found after a real search scores 0.

Sources disagree on the order of work. Competitive-intelligence vendors put postings first, while the IP literature puts patents first. I start with the patents. Patent events are dated legal acts with fees attached, while a posting costs little to publish and nothing to leave up.

A filing-count chart tells you a competitor is paying lawyers. It does not tell you anyone is still building.

The procedure: from a list of filings to a defended verdict

Follow these eight steps in order. The patent layer comes before the people layer because the blind-window tags from steps 1 and 2 decide which people-layer results you can trust.

Competitor patent-cluster read

  1. Define the cluster and freeze an as-of date
    Group the filings by CPC code or claim theme into a list of family IDs, and write down the date the read is valid for. Every clock below runs from that date.
  2. Mark the blind window
    Tag each family visible yes/no (18 months from priority) and breadth assessable yes/no (30 or 31 months). Anything that cannot be assessed yet scores U, never 0.
  3. Map families and continuations
    In Patent Center, the EPO Register and PATENTSCOPE, count continuations, divisionals, distinct offices and the first-to-last filing timespan for each family. Score dimensions 1 and 3.
  4. Check prosecution and fee status
    Record every lapse, missed fee and abandoned national phase with its date, then search for revival petitions and live child applications. Score dimension 2.
  5. Build the inventor roster
    Deduplicate the names, match each on name plus prior employer plus location or field, and mark stayed, left or unknown with each leaver's destination. Score dimension 4.
  6. Measure team growth
    Count current employees whose skills match the cluster theme and compare with a prior snapshot taken with the identical query. Score dimension 5.
  7. Look for off-patent corroboration
    Search postings, public repos, papers and regulatory filings that name the same technique, and note the quarter each appeared. Score dimension 6.
  8. Score and write the dated confidence note
    Add up each layer, apply the verdict rules and fill in the note. A second analyst re-scores without seeing your score; you are done when both reach the same verdict.

Typical effort: half a day to define the cluster, an hour for the blind window, one to two days for family mapping, half a day for fee status, one to two days for the roster, half a day for team growth, and a day for corroboration. Plan on roughly a working week per cluster the first time, and less on re-reads, because the roster and the saved query carry over.

The note below is what leadership sees. Write it so that the strongest evidence against your verdict sits next to the strongest evidence for it.

Dated confidence note
VERDICT: <Live Program | Defensive | Abandoned>
CONFIDENCE: <High | Medium | Low>
AS-OF DATE: <YYYY-MM-DD>
CLUSTER: <CPC code(s) or claim theme>; <n> families; priority dates <earliest> to <latest>
BLIND WINDOW: <n> families under 18 months; <n> under 30/31 months (breadth scored U)
SCORES: Patent <x>/6 (U: <n>)   People <y>/6 (U: <n>)
AGREEMENT CHECK: <fee layer and people layer agree | disagree because ...>
STRONGEST EVIDENCE FOR: <one dated fact>
STRONGEST EVIDENCE AGAINST: <one dated fact>
WHAT WOULD CHANGE THIS: <the event, e.g. third-fee decision on family X, due YYYY-MM-DD>
RE-CHECK DATE: <YYYY-MM-DD>, triggered by <18-month publication | 30/31-month entry | fee due | quarterly snapshot>
SECOND ANALYST: <initials>; blind re-score Patent <x>/6, People <y>/6; verdict <...>; match <yes/no>

One note per cluster; the second analyst fills in the last line without seeing the scores above it.

How does this read go wrong?

Most wrong verdicts come from reading one signal as proof when it only shows timing, procedure or coincidence. Each failure below has a specific check, and a verdict that skips any of these checks should not go to leadership.

  • Low filing cadence read as abandonment. What it looks like: no new families in the latest year of the data. Why it lies: the newest work is inside the 18-month blind window, or sits under a nonpublication request until grant. Check: the priority dates against the as-of date, before you score anything 0.
  • A lapse read as walking away. What it looks like: Patent Center shows a patent expired for non-payment. Why it lies: the owner can petition for reinstatement on unintentional delay, or may have moved the claims into a continuation. Check: look for a revival petition and any live child application in the family.
  • Family breadth read as a funded program. What it looks like: a large family with many members. Why it lies: family size can come from procedural splitting or a patent thicket rather than market intent. Check: count distinct offices, not family members.
  • "Inventor left" read as a dead program. What it looks like: most core inventors have new employers. Why it lies: firms keep patents when the inventor moves to a rival, and the maintenance effect weakens in exactly that case. Check: record where each inventor went before scoring dimension 4.
  • Name collisions. What it looks like: a confident match on a common name. Why it lies: common inventor names produce false matches. Check: require name plus prior employer plus a matching location or field, and mark anything weaker as unknown.
  • Job postings treated as proof. What it looks like: one posting that names the technique. Why it lies: every published lead time for postings is a vendor claim with no method behind it. Check: require several roles across functions in the same quarter, or a posting backed by a repo or paper.
  • Index counts treated as the whole market. What it looks like: a team-size or analyst-supply figure quoted as a census. Why it lies: counts depend on how strictly titles and headlines match. Check: quote counts as headline matches from a named source, with the query saved.

The mismatch flag deserves its own line. When the fee record and the inventor roster point in opposite directions and you cannot explain why, the right answer is a lower confidence grade and a re-check date, not a forced verdict. Given that only 3% of triadic patents are purely preemptive and 17% are partly preemptive, a mixed reading is often the accurate one.

How do you keep the read current and staff it?

Every verdict carries a re-check date set by the next clock that will change the evidence. That might be a family crossing 18 months, a national-phase deadline at 30 or 31 months, a fee due date, or the next quarterly people snapshot. A verdict with no re-check date becomes stale quietly.

Re-check triggers to put on the calendar:

  • 18 months after the newest priority date you know of: families in the blind window start to publish. Re-score dimension 1.
  • 30 and 31 months after priority: breadth becomes readable for PCT families. Replace U with a score on dimension 3.
  • Each fee due date plus the six-month grace period: a paid or missed fee is the clearest dated decision you will get. Re-score dimension 2.
  • Quarterly: re-run the saved people query and re-check leavers. Re-score dimensions 4 to 6.

Staffing is a build-versus-buy question, and the labour pool for the patent layer is not where most strategy teams sit. In Refolk's index, the patent-analyst pool is mostly offshore:

MarketProfilesTop employer share of sample
United States16576% (Cardinal IP, 19/25)
India87816% (Honeywell, 4/25)
India:US ratio5.3x (derived)

These are headline matches on Patent Analyst, IP Analyst, Patent Search Analyst and Patent Intelligence titles, not a census, and the employer shares come from 25-profile samples. Two conclusions still hold. Steps 3 and 4 draw on a pool that is about five times larger in India than in the US. And the US pool is concentrated: one employer accounted for 19 of the 25 US profiles sampled. Steps 5 and 6 belong with talent intelligence wherever the patent work sits, because the roster and the saved query carry over from one read to the next.

Before the verdict goes to leadership

  • Every family carries visible and breadth-assessable tags against a written as-of date.
  • No dimension is scored 0 for a family still inside the 18-month or 30/31-month window.
  • Family breadth counts distinct offices, not family members.
  • Every lapse has been checked for a revival petition and a live continuation or divisional.
  • Raw status text is recorded next to any coded abandonment or EPO lapse.
  • Every inventor match rests on name plus prior employer plus location or field.
  • Every leaver has a recorded destination or is marked unknown.
  • Team growth compares two snapshots taken with the identical saved query.
  • Off-patent corroboration rests on more than one posting.
  • The confidence note names its strongest evidence against and a dated re-check trigger.
  • A second analyst re-scored blind and reached the same verdict.

Questions practitioners ask

How do you tell defensive patents from real product plans?

You cannot tell from the patent record alone. No published study shows defensive filings differ from product filings in continuation counts or claim breadth. The working test is a mismatch between two layers: the competitor keeps paying fees and filing in the area, but the named inventors have left, the matching team is not growing, and no postings, repos or papers back up the filings. Fees paid with nobody building reads Defensive.

How long after filing can I see a competitor's patent application?

Usually about 18 months after the earliest priority date, when US and PCT applications publish. A US-only filing made with a nonpublication request stays invisible until grant. The applicant must make that request when filing and certify that there will be no foreign filing. Which countries a PCT family entered is not known until 30 months for the US, China and Japan, and 31 months for the EPO, Korea, India and the UK.

Does a lapsed patent mean the competitor walked away from the technology?

Not on its own. The USPTO Director can accept a late maintenance fee if the delay is shown to be unintentional, so a lapse can be reversed. The owner may also have moved the claims into a continuation or divisional that is still pending. Before scoring a lapse as abandonment, check for a revival petition and for any live child application in the same family.

How do I check whether a competitor's patent inventors still work there?

Start from the inventor field on each patent, not a keyword search. In Refolk's index, only 14 senior US research and engineering profiles mention "inventor patents" in their headline, so inventors rarely label themselves. Match each name to a current profile on name plus prior employer plus a matching location or field. Then mark each person stayed, left or unknown, and record where each leaver went.

Do competitor job postings predict a product launch?

Not reliably. The lead times in circulation, such as hiring showing up 6 to 18 months before a launch or 4 to 8 months before a feature ships, are vendor claims with no published method. A single posting proves nothing. Treat postings as corroboration only when several roles across different functions appear in the same quarter and name the technique the cluster claims.

Does national-phase entry in many countries show commercial intent?

It shows the owner is willing to pay for those markets. Family size is an established proxy for patent value. That is not proof of a launch, and no published case study shows that jurisdiction counts predict one. Since about 80% of inventions are never patented abroad, a single-country family is normal. Count distinct offices rather than family members, because procedural splitting inflates member counts.

Try it on the search you came here for

Stop building boolean strings. Just describe the person.

Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.

  1. 01Describe them

    One plain sentence. Role, city, stack, stage, whatever matters to you.

  2. 02I read the web live

    GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.

  3. 03You read the shortlist

    Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.

  • No boolean, no filters, no seat to buy. One box.
  • Read at search time, so a profile updated yesterday counts today.
  • Every step visible as it runs, every name with its reason.

500 free credits on sign-up. No card, no demo call. See real searches.

Read next