Every programme performs well against a baseline it can move

A baseline is a promise with a number attached. Everything reported afterwards is reported against it: variance, earned value, forecast at completion, the traffic light on the board pack.

Which means the baseline is load-bearing. If it can be moved, and moving it does not have to be disclosed, then every number resting on it describes the distance from wherever the programme most recently decided to stand.

This is not a hypothesis. It has been measured on public portfolios, by auditors with access to the underlying files, and the numbers are worse than the argument requires.

A builder that met its targets

Between 2014 and 2018 the US General Services Administration, the federal government's landlord, completed thirty-six major construction projects. Each cost at least twenty million dollars. Together they came to 3.2 billion.

The Government Accountability Office reviewed them and found that GSA routinely meets its cost and schedule goals.

It also found that GSA had rebaselined twenty-five of the thirty-six.

Rebaselining, in GSA's own description, reestablishes the point at which it measures on-schedule and on-budget performance. The reasons were ordinary and mostly defensible: design changes, requests funded by the tenant agencies. One modernisation took a 2.7 million dollar increase on a contract awarded at 21.8 million, to add a stairwell after the tenant decided to increase the building's occupancy.

Nothing in that is scandalous. What GAO objected to was that the annual performance reports did not say it had happened.

Simply measuring and reporting performance based on the most recent baseline may obscure how projects have performed over their entire construction time frame.

That sentence is the whole subject. Performance against the current baseline is a real measurement of a short and recent period. It is routinely read as a measurement of the project.

What appeared when the reporting changed

GAO recommended that GSA publish, for each completed project, the extent of rebaselining and the final cost. GSA did, and the first tranche covered thirteen projects completed in 2020.

Eight of the thirteen had a completion date that had been moved to a later one. Eight had a cost estimate that had been rebaselined. Measured against their original baselines, eleven of the thirteen took longer than planned; two finished sooner.

And against the final baselines, the ones already reset to reflect what had happened, ten of the thirteen still exceeded their cost.

No project changed. No new data was collected. The only thing that changed was which baseline the report subtracted from.

How common, and how often

GSA is one organisation. The scale of the practice was surveyed the year before, across the whole US federal portfolio.

GAO sampled 180 projects from the 778 major information technology projects the twenty-four major agencies planned to fund in 2008, and achieved a 99 per cent response rate. An estimated 48 per cent had been rebaselined. Of those, 51 per cent had been rebaselined at least twice, and about 11 per cent four times or more.

Four times. On a project whose whole purpose is to be finished.

The number that should stop you

The survey asked managers why. The answers are the most revealing part of the study, and they are almost never quoted.

Fifty-five per cent cited a change in requirements, objectives or scope. Forty-four per cent cited a change in funding stream. Fourteen per cent said the original baseline had been inaccurate.

Four per cent gave cost or schedule overruns due to project performance.

Rebaselining is, mechanically, what an organisation does when it can no longer report against the number it has. Asked directly, one manager in twenty-five named that as the reason. The rest attributed the reset to something arriving from outside.

Some of that is accurate. Scope does change and funding does move. But the distribution is too clean. A process triggered by variance, in which variance is almost never the stated cause, is a process describing itself in the vocabulary of things nobody could have prevented.

A legislature that had to write the second baseline into law

The strongest evidence that this is a structural problem rather than a lapse is that one government tried the obvious fix and it failed.

US defence programmes have been required to report cost growth against an approved baseline since 1982, under the provisions known as Nunn-McCurdy. The mechanism was straightforward: when unit cost grew by more than a set percentage, the department had to notify Congress.

It did not work as intended, because the baseline the growth was measured against could itself be revised. Congress amended the statute to establish an original baseline estimate alongside the current one, and required both to be used.

The thresholds now sit in law, at two levels each. A significant breach is a unit cost increase of at least 15 per cent over the current baseline estimate, or at least 30 per cent over the original. A critical breach is at least 25 per cent over the current, or at least 50 per cent over the original.

Read the structure rather than the percentages. The legislature concluded that a single threshold against a movable reference could not detect the failure it was written to detect, and that the correction was not a tighter percentage but a second, fixed reference. The doubled thresholds against the original baseline exist to price in the legitimate reasons a baseline moves, while still catching the total.

An owner does not need a statute to do this. It needs one line in the monthly report.

What a rebaselining policy has to contain

GAO assessed the twenty-four agencies' policies against five practices drawn from its own cost assessment guidance:

• Describe the reasons that make a rebaseline warranted.
• Describe the process for developing the new baseline.
• Require the new baseline to be validated, naming who can validate it.
• Require management review, naming the authority who decides and the criteria they apply.
• Require the process to be documented, including why the current plan is no longer feasible and what will prevent a recurrence.

None of the twenty-four agencies was fully consistent with all five. The weakest by a distance was the second: not one policy fully described how a new baseline should be developed, and 46 per cent did not address it at all. Validation was next, unaddressed in 54 per cent.

So the two practices that determine whether the new number is any better than the old one, how it was built and who checked it, were the two least likely to be written down. What agencies did specify was who approves. Approval without a described method is a signature on an estimate nobody has tested.

The companion failure, in the forecast rather than the baseline

, read: A forecast that never moves is not stable, it is unexaminedA forecast that never moves is not stable, it is unexamined

What to examine

None of this requires access to a cost system.

Ask for the sanctioned figure and the current figure, and for the list of approved changes between them. If the chain cannot be assembled, that is the finding, and it is more common than it should be.

Then ask three things about each reset. Who approved it, and did that person also carry accountability for the performance being re-based. What method produced the new number, and who validated it. Whether scope was removed to hold the cost, because descoping presented as underspend is the quietest version of this failure and the hardest to see later.

Finally, ask for the variance against the original. On a well-run programme that number is unsurprising and somebody already knows it. On a programme in trouble it will take three weeks to produce, which is itself the answer.

Why the body approving the reset is usually the wrong body

, read: A gate that has never been failed is not a controlA gate that has never been failed is not a control

---

Sources. US Government Accountability Office, Federal Buildings: GSA Can Improve Its Communication about and Assessment of Major Construction Projects, GAO-20-144, December 2019, for the thirty-six projects, the twenty-five rebaselines and the subsequent reporting of thirteen completed projects. US Government Accountability Office, Information Technology: Agencies Need to Establish Comprehensive Policies to Address Changes to Projects' Cost, Schedule, and Performance Goals, GAO-08-925, July 2008, for the survey of 180 projects, the frequency of rebaselining, the reasons given at Table 2 and the policy assessment at Table 4. The five practices are drawn from GAO's Cost Assessment Guide, GAO-07-1134SP. Cost growth thresholds and the definitions of the current and original Baseline Estimate are at 10 U.S.C. 4371, with the reporting duty at 10 U.S.C. 4374. ISO 21502:2020, Clause 7, and the PMBOK Guide, Eighth Edition, November 2025, are cited without a link because both are sold rather than published.