Legacy System Data Migration: Cutover Plans, Go/No-Go Criteria, and Reconciliation for Government Programs

How government programs can reduce data migration risk when replacing a legacy system: mock conversions on current data, measurable go/no-go criteria, record-level reconciliation, and a funded post-installation phase.

Most discussion of legacy modernization is about code: which language, which architecture, which cloud. The replacement system can be correct and the program can still fail on the day the data moves. Decades of records, with decades of exceptions in them, have to arrive in the new system complete, mean the same thing there, and be provably the same records.

This post covers the data side of a legacy replacement: what federal guidance treats as good practice, where a recent audit found the gaps, and how a program office can write cutover criteria that can be measured.

What the Government Accountability Office looked for

In May 2026, the Government Accountability Office (GAO) published GAO-26-107863, a review of the Department of Homeland Security’s financial systems modernization programs. Part of the review compared the department’s data migration guidance, and the planning documents of individual programs, against 11 leading practices. GAO drew those practices mainly from a Joint Financial Management Improvement Program white paper on financial systems data conversion and from the General Services Administration’s Modernization and Migration Management (M3) Playbook.

The practices fall into three phases. They were written for financial systems, but they apply equally to a personnel, logistics, or case management system.

PhaseLeading practices GAO assessedWhat goes wrong when it is skipped
Pre-conversionDetailed conversion plan with risk management, data cleansing, test data and mock conversions, agreement on data governanceDefects in legacy data are discovered in production, and nobody owns the decision about how to fix them
CutoverCutover plan with a backup plan, go/no-go decision with criteria and metrics for clean data, execution of cutover tasks, reconciliation of converted dataThe go-live decision is made on schedule pressure, and there is no documented way back
Post-installationConfirming converted data works as designed, a post-conversion data cleanse, archiving legacy files and closing out legacy dataErrors that only appear under real use have no budget, owner, or procedure

Where the audit found the gaps

GAO rated the department-wide guidance as partially consistent with the leading practices overall. The guidance was partially consistent on pre-conversion activities. On cutover, it described how a go/no-go decision would be made but did not discuss the metrics for deciding that data was clean enough. On post-installation activities it was not consistent at all. GAO concluded that programs relying on that guidance “are at increased risk of data errors, time needed to resolve errors, potential delays, and potential increased costs.” The department concurred with all five of GAO’s recommendations, including one to add post-installation and operations planning to its data migration guidance.

Two details from the program level are worth a program manager’s attention. A Coast Guard official told GAO that the go-live date for its new system was “primarily schedule and calendar driven,” and Coast Guard officials reported using older data for mock testing because the schedule was compressed. Separately, a Federal Emergency Management Agency official told GAO the agency did not plan a post-conversion data cleanse because it considered its pre-conversion work sufficient.

Neither choice is unusual. Both follow from a plan that treats go-live as the finish line: rehearsals run on whatever data is available, and the period after cutover is assumed to be quiet.

Rehearse on current data

A mock conversion is a full rehearsal of extracting, transforming, and loading legacy data into the target system. Its value depends on how closely the rehearsal data matches what will be converted on cutover day.

Legacy data drifts. A field that held a two-digit year until 1999 holds four digits afterward. A status code was reused for a different meaning after a policy change. A rehearsal on a two-year-old extract will not see records created under the most recent rule changes.

Three practices keep rehearsals honest:

  1. Refresh the extract for each mock conversion, and record its date in the results.
  2. Track defects by cause (source data, mapping rule, or target system constraint), so the trend across rehearsals shows whether cleansing is working.
  3. Run at least one rehearsal at full volume and inside the planned cutover window, since a conversion that is correct but takes 70 hours does not fit a 48-hour window.

Write go/no-go criteria that can be measured

GAO’s leading practice is that management makes a go/no-go decision before go-live and sets criteria and metrics for the threshold of clean data. The metrics are the part most often missing. A decision authority under schedule pressure needs numbers that were agreed before the pressure arrived.

Useful criteria name a quantity, a threshold, and a consequence. Examples: the count of records rejected by the target system, by record type; the count and dollar value of unreconciled differences in control totals; the number of open severity-one conversion defects; and whether the fallback procedure was exercised in the last rehearsal. Each should state who can waive it and what record the waiver leaves.

Reconcile at the record level, and keep the proof

Row counts and control totals are necessary and not sufficient. Two errors can cancel in a total, and a record can load successfully with a field truncated or a code mapped to the wrong value.

A stronger approach combines three levels of check:

  • Counts and totals per record type and per period, legacy against target.
  • Field-level comparison of a statistically meaningful sample, or of every record where volume allows, using a canonical form so that formatting differences are not reported as errors.
  • Behavioral comparison: run the same query, report, or calculation in both systems against converted data and compare results. This catches cases where the data is identical but the new system interprets it differently.

Every adjustment made during reconciliation should be documented, as GAO’s practice states. Hashing the reconciled extract and the comparison results at cutover gives an auditor or authorizing official a record whose integrity does not depend on anyone’s memory. Our post on evidence for the authorization to operate covers how this kind of artifact feeds change review.

Decide in advance how long two systems will run

Keeping the legacy system available is the usual fallback, and it has a cost. In GAO-22-103636, GAO found that the Air Force had no migration plan for moving from its legacy accounting system to its replacement and had instead adopted a dual processing policy under which both systems would operate for years. The Air Force has since provided a migration timeline that targets completion in fiscal year 2029.

A fallback plan should therefore have an exit: the conditions under which the legacy system becomes read-only, the conditions under which it is archived, and a date by which someone must decide.

An illustrative example

Consider a personnel records system with 4.2 million records being moved off a mainframe. All numbers here are illustrative.

The first mock conversion, run on a current extract, rejects 61,000 records. Tracking by cause shows that 48,000 share one problem: a separation code retired in 2004 that the mapping rules do not cover. A mapping fix resolves those, and the remaining 13,000 are cleansed in the source. By the third rehearsal, rejects are down to 900, and the full-volume run finishes in 31 hours against a 48-hour window.

The go/no-go criteria, agreed two months before cutover, read: fewer than 1,000 rejected records, each with an assigned owner; zero unreconciled differences in pay-affecting fields; fallback exercised in the final rehearsal. The program meets all three and proceeds. In the first month after go-live, behavioral comparison of a retirement eligibility report finds 140 records where a date field converted correctly but is interpreted under a different time zone rule. Because the plan funded a post-installation cleanse and kept the legacy system read-only for 90 days, the fix is routine.

How QuantumWorks approaches this

PRISM analyzes a legacy codebase and maps entry points, behavior, dependencies, and must-preserve paths, each anchored to file and line, which helps a team find special-case handling of data before a mock conversion does (see our post on behavior preservation). After migration, CAM provides asset-level cryptographic evidence for assets in production.

If your program is planning a cutover and wants its go/no-go criteria and reconciliation evidence to hold up under review, we would be glad to talk. Contact QuantumWorks.

  • Data Migration
  • Cutover Planning
  • Go/No-Go Criteria
  • Data Reconciliation
  • Legacy Systems
  • Verification

Working on a hard technical problem?

We work with government organizations, research institutions, and technology teams across cybersecurity, software, and autonomous systems.