DataKeeper · Pension data quality

    Find the error in the database before the calculation, and prove what was corrected

    For RPPS (Brazil's public-sector pension schemes) and closed pension funds (EFPC), DataKeeper brings together quality tests, member re-registration, census and proof of life campaigns, tracking of inconsistencies and a history of validations in one place. Your team understands the database, prioritizes corrections and documents each check, with personal data protected by pseudonymization and encryption.

    The demo is public, requires no sign-up and runs on a fictitious database. No file from your institution is sent. The demo and the DataKeeper website are in Portuguese.

    Findings queue

    Illustrative example, fictitious data

    Hire date after retirement date

    DB plan · retireesForwarded

    Dependent with no linked public servant

    Survivor pensionersCorrection reported

    Possible duplicate document ID across employment links

    Import file 03/2026Resolved after re-run

    Run recorded: database analyzed, rule version, coverage and history of each finding

    The problem

    Four ways a pension database fails silently

    An inconsistent database rarely breaks anything right away. The error moves on to the actuarial valuation, the benefit award, the benefit payroll and the rendering of accounts.

    The error shows up late

    A hire date after the retirement date, a dependent with no linked public servant, the same document ID in two records. Nobody notices until the actuarial valuation, the benefit award or an oversight body asks.

    Nobody knows who is handling it

    The list of open issues lives in a spreadsheet that each department edits. There is no way to tell what was forwarded, what was corrected and what simply disappeared from the list.

    No proof is left of what was checked

    Which database was analyzed, with which rules, over which period? Without a history of the runs, database quality becomes a matter of opinion, and it does not hold up when someone asks.

    The benefit keeps being paid after death

    Without up-to-date member re-registration and proof of life, the database only learns of a death when someone reports it. Auditing the civil registry's death records, the TCU (Federal Court of Accounts) estimated R$ 4.4 billion paid between 2016 and February 2025 to 275,800 people who had already died, adding up welfare, social security and labor benefits and the payrolls of active and retired public servants and survivor pensioners.

    Source: TCU (Federal Court of Accounts), news release on Acórdão 1.606/2025-Plenário (opens in a new tab).

    Six questions every pension database needs to answer

    Each DataTest checks one quality dimension. The result is observable: which records, which rule, which severity.

    01

    Completeness

    Are the required fields filled in?

    02

    Validity

    Do the values follow the expected format and criteria?

    03

    Consistency

    Does related data tell the same story?

    04

    Uniqueness

    Are there signs of duplicate people or employment links?

    05

    Integrity

    Do the expected relationships, such as plan, employment link and dependent, exist?

    06

    Timeliness

    Does the data meet the time reference of the analysis?

    From the database received to the proven correction

    Four steps. The last one closes the cycle: what was corrected is tested again.

    01

    Organize the database

    Identify the origin, period and scope of what will be analyzed. Field mapping links the file received to the DataKeeper data dictionary.

    02

    Run the DataTests

    Apply the rules that fit the database and check the coverage: which rules ran, on which records and what was left out.

    03

    Prioritize the findings

    Separate what is critical, identify the affected records and route the handling to whoever is responsible for it.

    04

    Verify the correction

    Run the tests again. A finding is only considered resolved after the re-run, and the link between finding, handling and result is kept.

    The platform

    What DataKeeper does

    Preparation, verification, handling, protection and evidence, in the same system, over the same database.

    Member re-registration, census and proof of life

    Campaigns with invitations, in-person service, proof of life by video or at the counter, a review desk and possible death cases. Reviewed responses return to the database as a data load and go through the same DataTests.

    See how it works

    Preparation and lineage

    Field mapping, deterministic standardization of dates, percentages and codes, and an immutable history of each import: records received, accepted and rejected, with origin and version.

    DataTests and rule catalog

    Parameterized rules for pension databases, such as a missing required field, date sequence, plan reference and possible duplicate, with the coverage of each run in view.

    Findings queue

    Each inconsistency with the affected record, the rule that flagged it and its severity. Three handling states: forwarded, correction reported, resolved after re-run.

    Groups and labels

    Groups by plan, sponsor or government body, age range, position, status, period, source system or import file. Open issues in a group never drop out of the overall count.

    Protection and sharing

    Vault-based pseudonymization, field-level encryption and blind indexing. Outputs for analysis or sharing are prepared with masking, pseudonymization or anonymization according to the purpose.

    Evidence of each run

    Run identifier, database analyzed, rule version, coverage, period, findings history and comparison between runs, traceable by participant and by field.

    Member re-registration, census and proof of life

    The campaign summons members, receives the responses and takes them to the review desk. What the desk approves returns to the database as a data load, goes through the DataTests and stays in each person's history.

    What the rules require

    Campaign from a template

    Ready-made templates for the pension census (RPPS), the annual proof of life of retirees and survivor pensioners, member re-registration (EFPC) and simple re-registration. The fund adjusts the audience, deadline, fields and required documents. The audience is frozen when the campaign is published and becomes the denominator for coverage.

    Response by phone, at the counter or in the field

    The participant opens the invitation link with their document ID and date of birth, checks the data and photographs the documents. Staff record the response at the counter, by phone, during a visit or from the paper form. A field census carried out by a contracted company comes in by file.

    Proof of life by video or in person

    A video of up to 30 seconds recorded on a phone, in which the person reads a sentence and a code that changes with every attempt. At the counter or during a visit, staff record an in-person proof after checking a photo ID. An AI analysis checks the speech (whether the sentence and the code were read) and the face (whether a single person appears, the same one from start to finish and the one in the ID photo) and gives an opinion to the reviewer, who decides. The video used in the analysis is deleted right afterwards, and no face template is stored. Each proof is marked as pending, received, confirmed, inconclusive or possible death.

    A possible death becomes a case for the fund

    The review desk flags the possible death, the administrator and the campaign manager are notified, and the fund records the outcome: death confirmed, person alive with a new proof, or other. DataKeeper does not suspend a benefit or change the registration on its own. A confirmed death leaves the coverage denominator.

    Review desk

    Previous database and response side by side, field by field, with a decision per item and per document and return to the participant with the reason. With the second reviewer turned on, changes to name, date of birth, sex, dependent or beneficiary, responses given by a representative and possible deaths go through another person on the team.

    Back to the database, tested

    When the campaign closes, each plan receives a data load containing only the reviewed responses. It goes through the DataTests, approval and publication like any other load, and every changed field enters the history with its origin. Responses not yet reviewed stay out of the tests, comparisons and database reports.

    What each campaign leaves on record

    • Coverage report by plan and by category: summoned, responded, reviewed, no response and confirmed deaths. In an RPPS census, active members, retirees and survivor pensioners appear separately, with the Pró-Gestão RPPS reference of 80% per group, shown as an indicator and without attesting the certification.

      Source: Manual do Pró-Gestão RPPS, version 4.2, item 3.1.6, subitem 2.2 (opens in a new tab).

    • Final campaign report, with the response funnel, the review desk decisions and the data loads generated per plan.

    • Possible death cases, open and resolved, with who flagged them, when, and the recorded outcome.

    • Invitation by email, automatic reminder before the deadline, SMS reminder for people without email (within the configured sending hours, up to two per person in each campaign) and a letter with a QR code; a receipt with a QR code for the participant.

    • History of every team action in the campaign and a draft data protection impact report (RIPD, the LGPD equivalent of a DPIA), built from the campaign's own data.

    • Documents and videos encrypted and scanned by antivirus, delivered to the fund as a package and deleted 30 days after the campaign closes.

    Validating a document ID does not require exposing it

    Personal data stays protected throughout the analysis, including from those who operate the platform.

    A separate database per client

    Each client's current database sits in a database separate from the platform's control database.

    Column encryption, one key per client

    Personal data such as document ID and name are encrypted field by field, with each client's own key.

    Search by blind index

    Equality searches work without decrypting the data. The plain value does not circulate in order to be found.

    Centralized identity

    Sign-in goes through Datacoder's identity service. DataKeeper never receives a password and reloads profiles at every session.

    Append-only trail

    Chained access log and a history of every change by participant, field and origin.

    Recorded outputs

    Each export records who requested it, when, where it went and why. Preparing an output never changes or deletes the original database.

    The platform supports the data protection required by the LGPD, but it does not replace the institution's responsibilities. Technical details on environment, backup, retention and disposal are provided on request.

    What DataKeeper does not do, by design

    It does not replace the management system

    DataKeeper is a quality and monitoring layer over the data. Records continue to be maintained in the source system.

    It does not correct the database by itself

    No automatic change without approval. The correction is reported by whoever is responsible for it and confirmed by re-running the tests.

    It does not hide the database behind a score

    Instead of an opaque score, direct indicators: rule coverage, people with open issues, findings by severity.

    Who it is for

    Whoever looks after a pension database

    Pension funds (EFPC)

    Member re-registration of participants and beneficiaries, record clean-up, benefits and actuarial databases as part of the routine, before inconsistencies pile up.

    RPPS

    Pension census, proof of life of retirees and survivor pensioners and verification of the database before the actuarial valuation and the rendering of accounts.

    Consultancies and administrators

    Organized validation of the databases received from clients, with a history of what was checked in each delivery.

    Actuarial preparation

    A verified database with open issues documented before it goes to the calculation, with scope, history and references.

    What the subscription includes and what is contracted separately

    An explicit split, so the commercial conversation starts in the right place.

    Available in the platform

    • Import with field mapping and a history of each data load
    • DataTests catalog and coverage tracking
    • Findings queue with severity and handling states
    • Groups and labels to follow populations separately
    • Member re-registration, census and proof of life campaigns, with video proof of life analyzed by AI, email and SMS reminders, in-person service, review desk, possible death cases and coverage reports
    • Pseudonymization, field-level encryption, blind indexing and purpose-based access control
    • Audit trail and evidence of each run
    • Preparation of masked, pseudonymized or anonymized outputs

    Services contracted per project

    • Initial database assessment, with a fixed scope
    • Recurring specialist support, with analysis and review
    • Rules exclusive to the institution
    • Special integrations with source systems
    • Support for LGPD compliance projects

    Subscription

    from R$ 489.00 per month

    For up to 500 unique people monitored, with 6 months of history retention. The price follows the monitored population, without multiplying by employment link or entity within the contracted scope. New accounts get a 30-day free trial, with full access and no credit card. Above 50,000 people, the proposal is tailored.

    Estimate the price for your database (in Portuguese)

    History retention

    6 months

    included

    12 months

    +5%

    24 months

    +10%

    36 months

    +15%

    48 months

    +20%

    60 months

    +25%

    120 months

    +30%

    Before you talk to us

    Part of Datacoder's Data, Security and Governance Engineering.

    Want to see your database go through DataKeeper?

    Tell us the size and origin of the database. We set up an assisted evaluation with your team.

    Schedule a demo