Government go-lives fail in public. The record names five decisions.
Auditors general, legislative hearings, class actions. This sector documents its ERP failures like no other, and the same calls show up in every post-mortem. I ran cutover across build waves on a statewide Workday program for a state accounting office and am doing the same for a major American city now. Walk the five, find your seat, then run the two-minute check.
Five calls that decide a government program, from the cutover chair
Statewide Workday cutover across build waves, then a city program in flight. The evidence beside each call comes from the sector's own audit reports and court filings, which is a luxury no other industry offers. Pick a decision.
A statewide program does not get one heroic go-live weekend. It gets waves, each with its own freeze, conversion, reconciliation, and its own set of agencies convinced their wave is the important one. Wave delivery sounds safer. It moves the risk instead of removing it.
Diff every wave's cutover artifacts against the last wave's, and require any difference to be a decision somebody made on purpose.
Wave two's cutover plan quietly stops matching wave one's, because a different person built it under pressure. Unchecked, wave three relearns wave one's lessons at production prices. The comparison work is unglamorous and it is the discipline that held the statewide program's waves together. For scale on what the alternative looks like: one state's program is consolidating roughly 280 systems while about 40 agencies keep legacy applications running behind the new platform. The interface list is the program. The ERP is one line on it.
Whose job is it to prove wave three matches wave two, artifact by artifact?
First design session. Two agencies read the same statewide pay rule differently, and both are sure. Multiply by every rule, every agency, every bargaining unit. One state told a legislative hearing it was implementing against more than 50 collective bargaining agreements.
Get a name. One human being with authority to resolve a cross-agency rule dispute, and a written tie-break procedure, before design starts.
The deepest public post-mortem in the sector names this exact gap: an independent assessment of one state's failed payroll program cited disagreements among state agencies over how statewide payroll rules should be interpreted, along with design churn as new staff reviewed old decisions. Another state's program carries a 17-member steering committee plus a separate 17-member advisory council, and an audit still found a single developer underneath it all, reviewing code he wrote himself. Volume of governance and capability of governance are different things.
When two agencies read the same rule differently, who decides, and does the decision hold?
Parallel payroll testing, and the variance queue is flooding. You chase them down and a real share turn out to be the new system calculating correctly while the old one had been paying people wrong for years. A contractor on one state's program said it plainly: the old system wasn't paying people correctly.
Write the remediation policy before parallel testing starts. Signed by the CFO and the HR director: do we make people whole, how far back, out of which fund, and who tells the union.
The moment a variance is confirmed as a legacy error, you have left defect triage and entered a compliance and labor decision. Decide it at 11pm in go-live week and you get the worst available version of the answer. The variances also cluster where nobody demos: on one state program the flood came out of state police pay items, K9 handling, scuba, and weapon-carry differentials. Parallel testing does not validate the new system. It audits the old one, and somebody has to own what the audit finds.
If parallel testing proves people were underpaid for years, what is our written answer?
Starting with tax year 2026, qualified overtime is reported separately on the W-2 in Box 12 under Code TT, and only the premium portion of FLSA-required overtime counts. Contractual daily overtime, holiday premium, and sixth-day pay do not, though they feel identical to the employee.
Build two parallel overtime calculations: the FLSA-statutory one for tax reporting and the contractual one for what people are paid. In the earning codes, not in a year-end spreadsheet.
Public safety makes this hard. Police and fire run FLSA 7(k) work periods, with law enforcement overtime starting after 171 hours in 28 days and fire after 212. Court-time minimums pay hours that never count toward the FLSA threshold, and comp time accrues at time and a half only past it. A design with one overtime earning code cannot produce a compliant W-2 in January 2027, in front of every employee and every union at once, and the IRS penalty relief that covered 2025 has expired. This is configuration now or a change order later.
Show me the earning codes that feed Box 12 Code TT, for a 28-day work period, for both police and fire.
A go or no-go argument. The date is slipping, the vendor says it can hold, and nobody in the room can say what not going live costs, because nobody priced running the legacy system another year.
Price twelve more months of legacy before accepting any date. A buyer who has not priced the fallback has no position in a schedule argument.
One county budgeted roughly $1.1 million just to keep its legacy system breathing through a three-year conversion overlap, a line almost no business case carries. A state auditor examining one of the country's largest programs found no finalized contingency plan for project failure at all. The double-pay window arrives as a surprise only because nobody wrote it down. Once the fallback has a price, a go decision becomes a comparison instead of a leap, and the schedule conversation with the integrator changes tone immediately.
What does it cost to run the legacy system for another twelve months, and who decides?
What those five mean for the chair you sit in
Government programs answer to more chairs than most. This is the screen to forward: each seat's sharpest exposure, the early sign it is live, and the one question worth asking this quarter.
The interface list is the program
One state's scope reached roughly 280 systems, with about 40 agencies keeping legacy applications that need modification to interoperate. Underneath the legacy stack sit single points of human failure: one state's 40-year-old payroll ran on a language one employee knew, and an audit elsewhere found one developer reviewing code he had written himself. PeopleSoft, meanwhile, is not a burning platform. Oracle's support runs past 2036, and saying otherwise burns credibility in one sentence. The hard date is Infor Lawson on premise, end of support 2030-12-31, and that move is a reimplementation.
Nobody can produce the full interface and legacy-retention list with an owner per line.
Which legacy systems stay after go-live, for how long, and who owns each interface?
Control that is statutory, not preference
An appropriation is a legal ceiling, not a plan. Whether the system hard-stops an unappropriated spend, who can override, and whether the override is logged decides if finance trusts the platform. GASB 103 lands in FY2026, the first reporting model overhaul since 1999, on the same three-person accounting team doing the conversion. And the slip risk is measured in years: one state's first phase went from $144M to roughly $291M while its go-live moved about four years.
Budgetary control is being discussed as a configuration preference, with hard stop, warning, and reporting-only all still on the table late in design.
Is the appropriation check a hard stop, who can override it, and is the override logged?
Fifty rulebooks and a pension that reads as earned
One state put the number on the record: 50-plus bargaining agreements and payroll rules for a single implementation. Large retirement systems want payroll reported as earned rather than as paid, with pay rate and special compensation broken out, which commercial engines do not natively do. One city absorbed $534,000 correcting retirement contributions missed for 4,341 employees on a single cost-of-living payment. And after go-lives elsewhere, the class actions came from employees, pleading the city's own wage ordinances.
The pension interface is being described as a file. It is an accounting obligation transmitted as one.
Does the retirement interface report as earned or as paid, and what happens on the first retro settlement?
What a statewide program costs by the time it is honest
One state's program, from its own budget record. The first phase alone doubled while the go-live moved roughly four years, from mid-2022 to a July 2026 through January 2027 window. Another state's replacement program was created in 2014 and now targets January 2027.
The lesson is not that this state is unusual. It is that funding runs through legislative sessions, go-lives cross elections, and any date more than a budget cycle away is a hypothesis. Decision five exists because of charts like this one.
Four items already on the calendar for 2026 and 2027
Legislative sessions move, appropriations move, go-lives move. These do not. Every one of them lands on payroll or the ledger while the program is still in flight.
The obligation deadline already passed. Anything unspent returns to Treasury, and any program funded that way has a hard stop on a fixed date.
The overtime split either holds or it does not, in front of every employee and every union at once. The IRS penalty relief that covered 2025 has expired.
The first financial reporting model overhaul since 1999, landing on the same small accounting team running the conversion.
The real vendor deadline in this sector. Far enough out to plan properly, close enough that doing nothing is a decision. PeopleSoft is not on this list, and Oracle support runs past 2036.
Five questions the audit reports keep answering too late
Answerable from memory, scored on this page, nothing captured and nothing emailed. Every one of these has appeared as a finding in a published state audit or a court filing. Cheaper to answer here.
These five are the start of the instrument, not the whole of it. A full review also covers appropriation control, encumbrance for filled and vacant positions, the grants and SEFA scope, and the wave artifact discipline. Or skip the tooling and book the program review.
Which of the five is your program missing?
Pre-procurement, mid-build, or stabilizing after a rough payroll go-live. I sell no software and staff no builds, and in a sector where failures end in an auditor's report, that independence is the point. Tell me where the program is and I will tell you what I see.