Refolk
StandardProcess, data, and compliance

The Candidate Retention Standard: Hold, Re-Consent, Anonymize, or Delete

Grade every sourced or rejected candidate record Hold, Re-Consent, Anonymize, or Delete from five fields, with two graders agreeing, and prove deletion everywhere.

23 min readLast reviewed November 20, 2026Read as Markdown

Key takeaways

  • US hiring-record rules set floors (one year for private employers under 29 CFR 1602.14, four years for California applications) while EU regulators set ceilings (two years from last contact under the CNIL referential), so one candidate record can be bound by both at once.
  • A legal hold under 29 CFR 1602.14 reaches every candidate for the same position, not just the person who complained, so holds must be scoped by requisition ID.
  • Last meaningful contact must mean an inbound or two-way action by the candidate; counting automated sends or profile views is how the ICO found one tool whose retention clock restarted every time, keeping most of its database indefinitely.
  • In Greenhouse, anonymizing a candidate is permanent and removes the data from all reports, so adverse-impact and source-of-hire counts must be exported before the action runs.
  • A record with a hashed email but a kept title, employer, and city is pseudonymised, not anonymous, and is still personal data under the ICO's motivated-intruder test.
  • In Refolk's index, Germany has 7 people with recruitment-operations titles and 5 of them work at KPMG Deutschland, so outside the US the retention review usually needs a named owner from privacy or HR.

This standard is for whoever runs the monthly retention review on an ATS and a sourcing CRM: recruiting ops, revenue ops, or the person answerable for how candidate data was gathered and kept. It grades every sourced or rejected candidate record as Hold, Re-Consent, Anonymize, or Delete from five recorded fields, so two graders reach the same verdict. It ends with a checklist that proves a deletion reached every connected system. It is a working standard, not legal advice. Where the law is unsettled I say so and name the question to put to counsel.

What does this retention standard decide?

It decides, record by record, whether a candidate profile is kept as it is, kept only with fresh consent, reduced to statistics, or erased from every system. Each record gets exactly one verdict per review, plus a reason code and a next review date.

VerdictWhat happensReversible?
HoldKeep unchanged until a dated next reviewYes
Re-ConsentAsk for fresh consent; keep only if granted within 30 daysUntil day 30
AnonymizeStrip to aggregate statisticsNo
DeleteErase from ATS, CRM, outreach tool, exportsNo

Every Hold carries a reason code, so a later reviewer can see why the record survived:

  • H-LEGAL: a legal hold covers the requisition.
  • H-FLOOR: a statutory minimum period is still running.
  • H-WINDOW: the record is inside its privacy ceiling and not yet due.
  • H-COUNSEL: a floor and a ceiling conflict, and counsel has not ruled.

In scope are six kinds of record:

  • sourced profiles that never replied
  • sourced profiles that engaged
  • rejected applicants
  • finalists who were not hired
  • referrals and manual additions
  • consented talent-pool entries

Out of scope: personnel files for people you hired. Employment-record rules govern those, and this standard does not cover them.

Three terms recur throughout:

  • A floor is the minimum period a law requires you to keep a record.
  • A ceiling is the maximum period a regulator treats as justified before storage becomes excessive.
  • A clock start is the event a period is counted from, and it differs by regime.

Why do published candidate retention periods contradict each other?

The rules pull in opposite directions. US hiring-record rules set floors: how long you must keep records. EU and UK data protection regulators set ceilings: how long you may keep them. A single record can sit under both at once.

That is why published advice ranges from four weeks to four years. These sources mostly don't disagree. They answer different questions about different records. Any template that applies one period to every record will break one of the two rules for some of them.

RegimeFloorCeiling or guidanceClock start
US federal, private1 yrnonelater of record or action
US public sector2 yrsnonelater of record or action
California4 yrsnonecreation or employment action
France (CNIL)n/a2 yrs active; 5 yrs restricted archivelast contact; post filled
Netherlands (AP)n/a4 wks; 1 yr with consentend of procedure

How firm each row is, and what it rests on:

  • US federal. 29 CFR 1602.14 requires private employers to keep records for one year from the date the record was made or the personnel action taken, whichever is later. State and local governments and apprenticeship programs must keep them for two years. ADEA rules at 29 CFR 1627.3 set the same period. The EEOC is working to rescind related recordkeeping requirements. That work is unfinished, so treat this rule as a moving target.
  • California. Government Code 12946 requires keeping applications, personnel files, and employment referral records for at least four years after they are created or received. The minimum used to be two years. A secondary source reports that the Civil Rights Council's automated-decision system regulations extend the four-year expectation to automated-decision data. Confirm that with counsel.
  • France. The CNIL's recruitment referential is guidance. An unsuccessful candidate's CV is removed from the active database two years after last contact. A restricted archive may be kept for five years from the date the post was filled (Labour Code L1134-5).
  • Netherlands. The four weeks, or one year with consent, comes from the AP's own notice to people who apply for jobs at the AP. It describes how the regulator handles its own applicants, not a rule written for other employers. It still shows how that regulator reads storage limitation.
  • UK. The ICO publishes no number. Its audit of AI recruitment tools recommended a clear retention period, after finding tools that kept data indefinitely to build large databases of potential candidates without their knowledge.

My research could not establish three things: a state-by-state table of applicant-record minimums beyond California, any EDPB period specific to recruitment, and figures from German authorities. If your team has a state table, treat it as unverified until counsel builds or signs it.

Federal contractors have one more moving part. The old 41 CFR 60-1.12 rule required two years, or one year for contractors with fewer than 150 employees or no government contract of at least $150,000. OFCCP has removed it by final rule. VEVRAA still requires covered contractors to count applicants every year under 41 CFR 60-300.44(k). Check whether the removal has taken effect on rescission docket OFCCP-2025-0001 before you change anything.

Floor and ceiling on one record

Past ceilingInside ceiling
Past ceiling, floor expired
Re-Consent if eligible, otherwise Anonymize or Delete
Past ceiling, floor running
Hold H-COUNSEL, restrict access, get a written ruling
Inside ceiling, floor expired
Hold H-WINDOW until the ceiling date
Inside ceiling, floor running
Hold H-FLOOR until the later of the two dates
Floor expired or noneFloor still running
A record's verdict depends on whether its legal minimum is still running and whether it has passed its privacy maximum, and only one quadrant needs counsel.

The five fields every candidate record must carry

Every record is graded from exactly five fields: jurisdiction, record origin, last meaningful contact date, consent state, and legal-hold flag. A record missing any one of them cannot be graded. It goes to a manual queue, never to a default verdict.

FieldRecord asWhat it provesHow it lies
JurisdictionRegime codes, e.g. US-CA + FRWhich floors and ceilings applyOffice mapped, candidate location ignored
OriginApplicant, sourced-no-reply, sourced-engaged, referral/manual, poolWhich defaults and notices applyImported applicants tagged as sourced
Last meaningful contactDate plus contact typeWhen the contact-based ceiling clocks startAutomated sends reset it
Consent stateNone, active (expiry), expired, withdrawn, requested (date)Whether consent can extend retentionCheckbox with no timestamp
Legal holdY/N, requisition ID, matter referenceThat deletion is blockedSet on the complainant only

Record every regime that plausibly applies to the jurisdiction field, covering both the hiring entity and the candidate. Greenhouse sets retention rules per office, so an office-only mapping can grade an EU-resident candidate who applied to a US office as US-only. Any candidate location that counsel has not mapped to a regime is marked UNMAPPED and goes to the manual queue.

What counts as meaningful contact

Meaningful contact is an action the candidate took toward you, or a live exchange with them. It does not include anything your systems did to the record. This definition carries the most weight in the standard. The CNIL clock runs from last contact, so a loose definition decides how long you actually keep data.

Counts as meaningful contact:

  • The candidate replied to a message.
  • The candidate applied or submitted materials.
  • A call, interview, or other live two-way conversation took place.
  • The candidate scheduled, rescheduled, or asked for an update.
  • The candidate granted, renewed, or withdrew consent.

Does not count:

  • Automated nurture or sequence sends, whether or not they were opened
  • Recruiter profile views and search hits
  • Re-imports, enrichment refreshes, and merges
  • Stage moves, tag edits, and notes added by your team
  • Outbound messages that got no reply

In one case from the ICO's audit, the retention period restarted every time, so most of that database was kept indefinitely. That is the failure this definition exists to prevent.

A loose definition of contact turns a two-year cap into indefinite retention, one automated email at a time.
Retention review log row
record_id:
review_date:
jurisdiction: (e.g. US-CA + FR, or UNMAPPED)
origin: applicant | sourced-no-reply | sourced-engaged | referral/manual | pool
last_meaningful_contact: (date) | type: reply | application | interview | consent
consent_state: none | active (expires ...) | expired | withdrawn | requested (sent ...)
legal_hold: Y/N | requisition_id: | matter_ref:
floor_end_date:
ceiling_end_date:
verdict: Hold | Re-Consent | Anonymize | Delete
reason_code: H-LEGAL | H-FLOOR | H-WINDOW | H-COUNSEL | n/a
grader_1:
grader_2 (sample only):
disagreement_note:
next_review_date:

One row per record per review; use the ATS record ID so the log itself holds no personal data.

Verdict rules, in precedence order

Apply the rules from the top down, and the first one that matches decides the verdict. Legal hold comes first, then field completeness, the statutory floor, the privacy ceiling, and last the consent and origin defaults that choose between Re-Consent, Anonymize, and Delete.

Verdict precedence, first match wins

  1. Legal hold
    Any hold on the requisition means Hold H-LEGAL; nothing below is read
  2. Field completeness
    A missing or unmapped field sends the record to the manual queue with no verdict
  3. Statutory floor
    Floor running means Hold H-FLOOR, or H-COUNSEL if also past a ceiling
  4. Privacy ceiling
    Inside the ceiling means Hold H-WINDOW, reviewed again on the ceiling date
  5. Consent and origin
    Past the ceiling means Re-Consent if eligible, otherwise Anonymize or Delete by origin
A record is graded from the outermost layer inward, and a match at any layer ends the grading for that record.

When a record is eligible for Re-Consent

All of these must be true. If any one fails, the record goes to the origin default.

  • Origin is sourced-engaged, finalist, referral with contact, or consented pool.
  • At least one meaningful contact is on record.
  • Consent has not been withdrawn.
  • The regime allows retention to be extended with consent. The Dutch AP's own practice is one year with consent.
  • A named recruiter wants to keep this person for a stated role family. "Might be useful" does not pass.

Default verdict by origin, once past the ceiling

OriginDefault verdictWhy
Sourced, never repliedDeleteNo relationship and no applicant statistics to keep
Sourced, engagedRe-Consent if eligible, else DeleteA relationship exists but there was no application
Rejected applicantAnonymizeFeeds adverse-impact counts
Finalist, not hiredRe-Consent if eligible, else AnonymizeHigh re-hire value; also in counts
Consented pool entryRe-Consent before expiryConsent is the only basis for keeping it

Referrals and manual additions take the row that matches their engagement. Also check that they received your privacy notice (see failure mode 5).

Sourced profiles that never replied

This is where most published policies stay vague, and the law is genuinely thin here. The only explicit US rule on database-sourced candidates was OFCCP's 60-1.12(a). It covered resume database searches, recording the position searched for, the criteria used, and the date, "regardless of whether the individual qualifies as an Internet Applicant." That rule falls away with the removal of Part 60-1. I found no public EEOC position on whether a never-contacted sourced profile counts as a 1602.14 record that is "made or kept". Littler notes that the Internet Applicant definition was aligned with the applicant definition used in Title VII hiring discrimination cases, which is a reason not to assume the answer either way.

So ask counsel once and write the answer into the policy as a named setting, such as sourced_no_reply_is_1602_record: yes/no. If yes, these profiles take H-FLOOR for one year from record creation. If no, they sit only under privacy ceilings. In either case they default to Delete once past the ceiling. I never send them re-consent requests: emailing someone who never answered you to ask permission to keep their data is cold outreach.

US-only records have no regulator ceiling in the table above. Set a policy ceiling in writing anyway. The ICO's audit recommendation was a clear retention period, and a record with no ceiling ends up kept indefinitely.

Who should own and grade the review

You need one named owner who runs the review every month and a second grader who independently grades a 10 to 20% sample. In the US that owner usually holds a recruiting-ops title. Outside the US, that title is rare enough that the owner usually comes from privacy or HR.

CountryPeople with these titlesDominant title (share)Top-employer concentration
United States160Recruiting Operations (72% of a 25-person sample)8% (2 of 25)
United Kingdom19Recruitment Operations (63%, 12 of 19)5% (1 of 19)
Germany7Recruitment Operations (57%, 4 of 7)71% (5 of 7, KPMG Deutschland)

These are counts from Refolk's index of professional profiles. The US title share comes from a 25-person sample, not all 160. The US has 8.4 times as many of these people as the UK, and the UK 2.7 times as many as Germany (both ratios derived).

7
People in Germany holding recruitment-operations titles in Refolk's index
Five of the seven work at KPMG Deutschland, so at most German employers the retention review will fall to someone without that title.

My inference, which is not established fact: in EU markets the review usually lands with a DPO or an HR generalist, and they need this standard spelled out more than a dedicated ops lead would. The title split matters too. US profiles say "Recruiting Operations" and UK profiles say "Recruitment Operations", so a title-keyword search for an owner or a second grader quietly drops half the market. Describing the person in plain English in Refolk avoids that. The search runs on the role you describe rather than one spelling of the title.

The two-grader agreement rule

The second grader works blind: they see the five fields, not the first grader's verdict. The target is 100% agreement on the sample. Every disagreement is logged with the rule that caused it. Then the rule is amended so the same case can't be read two ways next month. Don't settle a disagreement by majority or seniority. That fixes one record and leaves the ambiguity in place.

The monthly retention review procedure

The review runs in eight steps. They run in the order shown because two of them are irreversible. Freeze holds first, grade before acting, export statistics before anonymizing, and prove deletion after executing.

Monthly retention review

  1. Run the freeze check
    Pull every open charge, complaint, and litigation hold from counsel and set the legal-hold flag on every candidate for the same requisition. Done when the flags match counsel's list one-to-one.
  2. Complete the five fields
    Confirm every record due for review carries all five fields. Records missing any field go to a named manual queue, never to a default verdict.
  3. Assign the clocks
    Start each regime's clock at its own event and write a floor-end date and a ceiling-end date on every record.
  4. Grade verdicts with two graders
    Apply the precedence rules, have a second grader blind-grade a 10 to 20% sample, and amend any rule that produced a disagreement.
  5. Send re-consent requests and wait 30 days
    Send the request to every Re-Consent record. No reply after 30 days converts the verdict to Anonymize or Delete by origin.
  6. Export aggregates before anonymizing
    Export and date adverse-impact and source-of-hire counts for every requisition with an Anonymize verdict, before any anonymize action runs.
  7. Execute and propagate
    Run the action in the ATS, then the sourcing CRM, the outreach tool, and every export or shared file.
  8. Prove it
    Search every system by email or hash, log zero results, get a second signature, and confirm the automated retention jobs actually ran.

Clock starts by platform and regime

Each clock starts at a different event, and getting the event wrong moves every date that follows from it:

  • Greenhouse: the retention timer starts when a candidate has been rejected on all job applications.
  • CNIL: runs from last contact, which is why the definition of meaningful contact is load-bearing.
  • California: runs from when the record is created or received.
  • US federal: runs from the later of when the record was made or the personnel action was taken.

What your ATS will and will not do

CapabilityGreenhouseTeamtailor
Retention by statusRejected and hired, per officeRejected, inactivity, permission
Sourced-specific expirySend policy separately to non-job-post entriesYes, sourced and referred
When a period endsNotifies users, then manual (secondary source)Deletes automatically by rule
Default periodNot publishedCustomer sets

Neither platform publishes a default period. Teamtailor says plainly that it is always up to you to set the retention times. Greenhouse sends consent extension emails automatically where consent is the lawful basis, which handles step 5 for those records. Greenhouse's Talent Matching caches parsed resumes for 30 days, so a record deleted today may still sit in that cache. Allow for it when you schedule proof searches. The "notifies, then manual" row comes from a reproduction of Greenhouse's documentation. Check it against your own instance before relying on it.

Sources disagree on whether the aggregate export belongs before or after grading. My standard only fixes the order that matters: the export must exist before any Anonymize action executes.

Re-consent request
Subject: Should we keep your details on file?

Hi {first name},

We last spoke on {date of last meaningful contact} about {role family or team}. Our retention policy is to remove candidate details after {period} unless you ask us to keep them.

If you would like us to keep your profile so we can contact you about similar roles, reply "keep" or use this link: {consent link}. We will keep it for {extension period} and then ask you again.

If we have not heard from you by {send date + 30 days}, we will delete your details or keep only anonymous statistics. You do not need to do anything.

{Recruiter name}
{Company}

Send only to records graded Re-Consent; set the extension period to the ceiling for that record's jurisdiction.

How this standard goes wrong

Most retention failures look compliant on the dashboard. The clock reads current, the report says statistics were preserved, the ATS says the record was deleted. Each check below tests whether the signal is true.

  1. Non-meaningful touches reset the clock. An automated nurture email or a profile view gets logged as contact, so the last-contact date keeps moving and the record never ages out. This is the restarting retention period the ICO found. Check: count records whose last-contact date is recent but which have had no inbound reply in the last 24 months. Any count above zero means your contact field is being fed by automation.
  2. Anonymizing destroys the adverse-impact evidence. The false positive is "statistics preserved" when the counts have actually left the reports along with the personal data. Check: confirm a dated aggregate export exists for every requisition before its records are anonymized.
  3. Pseudonymization labelled as anonymization. A record with a hashed email but a kept title, employer, and city fails the motivated-intruder test, and pseudonymous information is still personal data. Check: pick five anonymized records and try to re-identify each with a web search. One success fails the batch.
  4. Hold scoped to the complainant only. 1602.14 reaches all candidates for the same position. A person-level hold lets the rest of the slate be deleted while the charge is open. Check: query holds by requisition ID, not by person, and compare the counts.
  5. Non-job-post entries miss the notice. Greenhouse says referrals and manual additions should be sent your GDPR policy separately. Records added outside a job post may never have received it. Check: filter by origin and look for missing notice timestamps. A missing notice is a manual-queue item, not a grading input.
  6. An ATS delete treated as a global delete. The ATS record is gone, but the CRM, the sequencer, and CSV exports still hold the person. Check: keep a lookup log for each system, signed by a second person (next section).
  7. Federal contractors drop all applicant tracking. Removing Part 60-1 does not remove VEVRAA applicant counts under 41 CFR 60-300.44(k). Check: get counsel's sign-off before retiring any applicant-count report or field.
  8. One period applied to all records. A California applicant who is also covered by EU rules faces a four-year floor that runs longer than the CNIL's two-year active-base ceiling. Check: run a query for records carrying both a floor regime and a ceiling regime. Every hit is H-COUNSEL until there is a written ruling. One question to put to counsel is whether moving the record to the CNIL's restricted archive meets both rules.

Proving the deletion reached every system

A deletion is proven when a search by email or hash returns zero records in every system on your register, and a second person has signed the log. Neither Greenhouse nor Teamtailor documents deletion reaching outside tools. Greenhouse says its retention rules only anonymize personal information as it appears on the candidate profile. So assume nothing propagates on its own.

Deletion propagation and proof

  1. ATS
    Run the anonymize or delete action and note the job or batch ID
  2. Sourcing CRM
    Delete the matching person record and any attached notes or files
  3. Outreach tool
    Remove the contact from sequences, suppression aside, and delete the record
  4. Exports and shared files
    Purge CSVs, spreadsheets, and drive copies listed on the export register
  5. Proof log
    Search each system by email or hash, record zero results, second reviewer signs
An erasure is only complete when the last system on the register returns zero results and a second reviewer has signed the log.

Keep a system register: one line for every place candidate data can live, with an owner per line. If an export is not on the register, you cannot prove it was deleted. So the register comes first and the exports second.

What anonymized has to mean

I found no regulator-published field list for candidate records. The list below is my own, derived from the ICO's test of whether there are means reasonably likely to be used to identify someone, with the motivated intruder as the starting point. Treat it as derived, not official. Remember too that the anonymization step itself counts as processing.

  • Remove: name, email, phone, address, profile URLs and handles, photo, CV file, free-text notes, and exact dates.
  • Remove the combinations: exact title plus employer plus city, and school plus graduation year. Together these re-identify people even when no single field does.
  • Keep: requisition ID, job family, stage reached, outcome, source channel category, month or quarter, coarse region.
  • Prefer aggregates to rows: in a small requisition, even coarse row-level fields can point to one person. Where the count is small, keep only the exported totals.

Before calling the review done

  • Every legal-hold flag matches counsel's list and is scoped by requisition ID.
  • The manual queue is empty, or every record in it has a named owner and a due date.
  • Every graded record carries a floor-end date, a ceiling-end date, and a next review date.
  • Two blind graders agreed on 100% of a 10 to 20% sample, or the amended rule is logged.
  • Every Re-Consent record without a reply after 30 days has been converted to Anonymize or Delete.
  • A dated aggregate export of adverse-impact and source-of-hire counts exists for every requisition with an Anonymize verdict, created before the action ran.
  • A search by email or hash returns zero results in the ATS, sourcing CRM, outreach tool, and every register-listed export.
  • A second reviewer has signed the lookup log for every system.
  • Five anonymized records were tested against a web search and none could be re-identified.
  • The ATS's automated retention jobs for this cycle were confirmed to have deleted or anonymized what they were set to.
  • For federal contractors, VEVRAA applicant counts were preserved and counsel signed off on any retired report.

Keeping the standard current

Re-check the inputs on a fixed schedule and whenever one of these sources moves. The verdict rules stay stable. The periods and clock starts that feed them do not. Put a named owner and a quarterly date against each item:

  • EEOC recordkeeping. The rescission of the Guidelines and related recordkeeping requirements is unfinished. When it lands, re-check the one-year floor and whether the 1602.14 hold language survives.
  • OFCCP and VEVRAA. Confirm the removal of Part 60-1 has taken effect on docket OFCCP-2025-0001. Then check that the VEVRAA counting duty under 41 CFR 60-300.44(k) is still in force.
  • California. Have counsel confirm the scope of the automated-decision data rule, which so far rests on a secondary source.
  • State minimums. Commission a state-by-state applicant-record table from counsel. Until it exists, unmapped states go to the manual queue.
  • CNIL and the Dutch AP. Re-read the CNIL referential and the AP's applicant notice for changes to periods or clock starts.
  • ATS settings. Export your retention rules from the ATS and compare them with this policy. The ICO's audit is a reminder that a rule set correctly in configuration can still fail to delete.

When any input changes, update the precedence rules first, then re-grade the next month's sample with two graders before the change goes live. If the graders disagree after the change, the new wording is ambiguous. Fix it before it reaches the whole database.

Questions practitioners ask

How long can recruiters keep candidate data?

It depends on the record, because two kinds of rule apply. US rules set minimums: one year for private employers under 29 CFR 1602.14, two years for the public sector, four years for applications in California. EU regulators set maximums: the CNIL's referential allows two years from last contact for an unsuccessful candidate's CV. Find each record's floor and ceiling and grade it against both. A single period for every record will break one rule or the other.

Can we keep sourced candidates who never replied?

Only for as long as your policy and the applicable privacy ceiling allow. The explicit US rule on resume-database searches sat in OFCCP's Part 60-1, which is being removed, and I found no public EEOC position on whether a never-contacted sourced profile counts as a 1602.14 record. Ask counsel once, write the answer into policy, then delete these profiles when they pass their ceiling. Don't send them re-consent requests.

What is a reasonable talent pool re-consent period?

No regulator gives one period for everyone, so use the ceiling for each record's jurisdiction. The Dutch data protection authority keeps its own applicants' data for up to four weeks after a procedure ends, or up to one year with consent. That makes one year with consent a defensible benchmark for EU talent pools. Send the request before the period runs out, wait 30 days, and treat no reply as a no.

Does deleting a candidate in the ATS delete them everywhere?

Assume it does not. Neither Greenhouse nor Teamtailor documents deletion reaching outside tools, and Greenhouse says its retention rules only anonymize personal information as it appears on the candidate profile. Delete in the sourcing CRM, the outreach tool, and every export as well. Then search each system by email or hash and log zero results, with a second person signing the log.

Is pseudonymized candidate data good enough instead of deleting?

No. Under the ICO's anonymisation guidance, pseudonymous information is still personal data, so it stays under every retention ceiling. A record with a hashed email but a kept title, employer, and city usually fails the motivated-intruder test, because a web search can re-identify the person. Either strip the record to aggregate statistics that no one can trace back to a person, or delete it.

What happens when US and EU retention rules conflict on one record?

Hold the record under a counsel reason code, restrict who can access it, and get a written ruling. Don't settle it in a template. For example, a California applicant also covered by CNIL guidance faces a four-year floor that runs past a two-year active-base ceiling. Ask counsel whether moving the record to a restricted archive, which the CNIL recognizes, meets both rules.

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