Skip to content

Platform

Data migration validation you can show an auditor

17 checks on every table, checks on every code object, and one verdict computed by code.

2,141,876,402 = 2,141,876,402

rows in Oracle Database 19c and PostgreSQL 16.4, all 1,036 tables

Example run · Validation Report / ERP-Oracle to ERP-PG

1,036 tables validated, 17,612 table check runs in 42 min

Passed with warnings
SALES, 312 tables
FIN, 248 tables
INV, 196 tables
HR, 142 tables
COMMON, 138 tables

17,605 passed, 0 failed, 7 warnings. Verdict: Passed with warnings

Check groups

What are the six groups of checks?

Five groups decide the verdict; Performance readiness is advisory.

Check groups and results, ERP-Oracle to ERP-PG
GroupWhat it provesExample run result
CompletenessEvery row and object arrived2,141,876,402 rows on both sides; 1,036 of 1,036 tables
AccuracyValues match, not just countsChecksums on all rows, 0 differ; 1,036,000 sample rows compared; 7 warnings
StructureColumns, types and defaults as planned71,940 columns; 11 mapping rules applied
IntegrityConstraints hold7,969 constraints hold
Code objectsPresent, compiled, same results540 compiled; 12,200 function calls, results equal; 88 triggers enabled
Performance readiness (advisory)Indexes and statistics in place4,212 indexes; statistics on 1,036 tables

How it works

How does validation work?

Four steps run on every migration, and counts and checksums cover every row, not a sample.

  1. 1. Count and checksum every table

    Row counts compared table by table, and checksums on every chunk of rows: one changed value changes the checksum.
    • 2,141,876,402 rows
    • 17 checks per table
  2. 2. Compare sample rows

    In the example run, 1,000 rows per table were compared value by value, shown side by side in the report.

    • 1,000 rows per table
  3. 3. Check every code object

    Every object must compile. Deterministic read-only functions run on the same inputs on both databases and the results are compared.
    • 540 compiled
    • 12,200 function calls
  4. 4. Compute the verdict

    Code computes one verdict: Passed, Passed with warnings, Not verified or Failed. Warnings need a named approver.
    • Passed with warnings

Evidence

What does the Validation Report show?

Each result links to its evidence, with source and target values side by side.

  • EqualSource and target hold the same value.
  • MappedThe value changed type by a mapping rule.

In the example run, Sarah Mitchell accepted 7 FIN rounding warnings on 3 Oct 2026, with a note.

Example run · Evidence sample / Accuracy / SALES.ORDER_LINES
Source and target rows side by side
ColumnSourceTargetResult
ORDER_LINE_ID88214501, 88214502, 9033017788214501, 88214502, 90330177Equal
IS_TAXABLE'Y'trueMapped
LINE_AMOUNT1249.50001249.5000Equal
SHIPPED_ATDATEtimestampMapped

Related

Questions

Frequently asked questions

How do you validate a data migration?

Compare the target with the source at every level: row counts, checksums on all rows, sample values, structure, constraints and code behaviour. Datalobe runs 17 such checks per table and keeps the evidence.

Are row counts enough?

No. Matching counts can hide changed values. Datalobe also compares checksums by chunk on all rows and compares sample rows value by value.

What happens to a warning?

A named approver must review it and can accept it with a note. The report records who and when. In the example run, the certificate came only after all 7 were accepted.

What does a failed validation look like?

In a rehearsal of the example migration, validation found 1,250 missing rows after an interrupted batch and a checksum mismatch on 3 rows. Verdict: Failed, no certificate.

Sample reports

Read a full Validation Report

See every check, every warning and the evidence behind the verdict.