The Magenta Book, the Treasury and Cabinet Office's rulebook for judging whether government policies actually work, was rewritten in May to call registering an evaluation mandatory. The government's own page explaining how to do that, unchanged since December 2024, still calls the same rule optional.
Estimated reading time: 6 minutes
,_WHITEHALL_SW1_London_20250616_0001.jpg){kind=link}
In short. The Magenta Book, rewritten on 14 May 2026, now says every government evaluation "must be registered on the Government Evaluation Registry" and calls this "this mandatory requirement" [1] (p.66). The Evaluation Registry's own FAQ page says the identical rule "should be registered" [3], and has not been updated since 5 December 2024 [4]. The government's own April 2025 review found two-thirds of the 227-project Government Major Projects Portfolio, worth £456 billion, still has no good-quality evaluation plan [5] (p.9). Three documented mistakes: four stars.
The Magenta Book is the government's own name for its central guidance on evaluation, jointly owned by HM Treasury and the Evaluation Task Force, and it exists to answer one question: how does government know whether a policy worked? On 14 May 2026 it was substantially rewritten for the first time since 2020, with new sections on value-for-money evaluation, "test and learn" approaches and artificial intelligence added alongside a rewritten core text [2]. One change stands out because it is stated as a hard rule rather than advice: "All planned, live and completed government evaluations from 1st April 2024 onwards must be registered on the Government Evaluation Registry," the new text says, and the following paragraph refers back to "this mandatory requirement" in plain terms [1] (p.66).
The rule the rewrite did not touch
The Government Evaluation Registry is real and reasonably well used: a public, searchable GOV.UK service that currently lists over 2,000 government evaluations, planned, ongoing and complete [7]. Its own FAQ page, the document departments are pointed to for practical guidance on using it, describes the same registration duty the Magenta Book now calls mandatory in softer terms: "All planned and live evaluations and evaluation reports signed off from 1st April 2024 onwards should be registered on the Evaluation Registry" [3]. That FAQ page was last updated on 5 December 2024, before the Registry had even publicly launched, and has not been touched since [4]. A department reading the Magenta Book in May 2026 is told registration is compulsory; the same department clicking through to the practical FAQ for the same rule is told it is something evaluations "should" meet, which on any plain reading of English is advice, not a mandate.
What the mandate is meant to fix
Registration alone does not test whether an evaluation is any good, and the government's own most recent audit of the problem shows why that distinction matters. In April 2025 the Evaluation Task Force reviewed the 2023-24 Government Major Projects Portfolio, 227 projects worth £834 billion covering railways, hospitals, defence and transformation programmes [5] (p.3). It found that a third of those projects, representing £378 billion, had a robust evaluation plan in place; the remaining two thirds, representing £456 billion, did not, and of that group, 55% had no evaluation plan on record at all [5] (p.9). The minister who published the review did not soften the framing: "This represents significant progress from the past, but there is still a need for considerable further improvement... I am pleased therefore to publish this review alongside an action plan that will embed evaluation into all stages of the major project lifecycle, so that evaluation becomes a 'must do' rather than a 'nice to have'," Georgia Gould MP, then Parliamentary Secretary for the Cabinet Office, wrote in the foreword [5] (p.2). A year on, the Magenta Book has made registering an evaluation a "must do." It has not made having a good one a "must do," and registering a poor plan satisfies the new mandatory rule just as well as registering a strong one.
A strategy that repeats last year's number
The Evaluation Task Force published its own new strategy for 2026-2029 on 19 June 2026, updated 24 August, after the Magenta Book rewrite [6]. It restates the same figure from the April 2025 review rather than a fresher one: "the ETF helped raise the proportion of Major Projects with good quality evaluation plans from 8% in 2019 to over 34% in 2025" [6]. The strategy sets qualitative "outcomes" for the next three years, among them "enhanced spending, governance, and accountability frameworks where using robust evaluation evidence is a requirement," but nowhere states a target percentage, or a date, for closing the 66% gap its own figures describe [6]. A reader is given the size of the problem and a promise to keep working on it, with nothing to check progress against when 2029 arrives.
The claims, tested
| The document's own words | What the evidence actually shows | Verdict |
|---|---|---|
| The Magenta Book: evaluation registration is "this mandatory requirement" [1] (p.66) | The Evaluation Registry's own FAQ page for the identical rule says it "should be registered," unchanged since 5 December 2024 [3] [4] | A newly mandatory rule and its own practical guidance disagree on whether it is mandatory |
| Evaluation is meant to ensure good decisions are made about "the most complex and strategically significant projects" [5] (p.3) | 66% of the GMPP by project count, £456 billion by value, has no good-quality evaluation plan [5] (p.9) | The registration mandate does not test or require plan quality |
| The ETF's 2026-2029 strategy sets "outcomes" including evaluation evidence being "a requirement" in spending decisions [6] | No quantified target or date is stated anywhere in the strategy for the 66%-without-a-good-plan baseline it reports [6] | A goal with no number and no date to be checked against |
The mistakes, counted
The Magenta Book's new mandatory registration rule contradicts the Evaluation Registry's own practical guidance for the same rule (1). The Magenta Book calls registration "this mandatory requirement" [1] (p.66); the Registry's own FAQ page for the identical obligation says it "should be registered" [3], and has not been updated to match since 5 December 2024 [4].
Mandatory registration does not touch the problem the government's own review found (2). Two thirds of the Government Major Projects Portfolio, £456 billion by value, has no good-quality evaluation plan [5] (p.9); registering a weak plan meets the new rule exactly as well as registering a strong one.
The Evaluation Task Force's new three-year strategy sets no measurable target for the gap it reports (3). The strategy repeats the 34% good-quality figure from April 2025 [6] without a fresher measurement, a target percentage, or a date to check against.
Credit where due
The improvement the government cites is real and independently documented, not asserted: good-quality evaluation plans on major projects rose from 8% in 2019 to 34% in 2024, verified by the Evaluation Task Force's own review rather than self-reported by departments [5] (p.3). Coverage is also better where it matters most: the 34% of projects with robust plans account for 45% of the portfolio's total cost, meaning the government's biggest, highest-risk projects are evaluated better than the average [5] (p.9). The Evaluation Registry itself is a genuine, working public database, not a paper commitment: over 2,000 real entries that anyone can search, not a promised service still waiting to launch [7]. And the Magenta Book's other new material, an artificial intelligence evaluation chapter and a dedicated transparency annex among them, addresses gaps this site has found elsewhere in government's AI and data guidance.
Verdict
Four stars, from three documented mistakes. The Magenta Book's shift from advisory to mandatory language on evaluation registration is a real, dated policy change, not a rebrand: the text is unambiguous, and the improvement in evaluation quality it builds on, from 8% to 34% of major projects, is genuine. But the government's own guidance disagrees with itself about whether the new rule is actually compulsory, the rule itself cannot fix the two-thirds-by-value gap in evaluation quality it sits alongside, and the strategy meant to close that gap over the next three years does not say by how much, or by when. A reader can now check, in public, that a plan for a project exists. Whether it is a good one is still, on the government's own most recent figures, a coin flip weighted against them.
Sources
- Magenta Book: Central Government guidance on evaluation, HM Treasury and Evaluation Task Force, May 2026
- The Magenta Book, GOV.UK publication page, HM Treasury and Evaluation Task Force, published 27 April 2011, updated 15 May 2026
- Evaluation Registry FAQs: HTML version, GOV.UK, Evaluation Task Force
- Evaluation Registry FAQs, GOV.UK publication page, published 15 May 2024, updated 5 December 2024
- Government Major Projects Evaluation Review, Evaluation Task Force, April 2025
- Evaluation Task Force Strategy 2026-2029, GOV.UK, published 19 June 2026, updated 24 August 2026
- Evaluation Registry search, Cabinet Office and HM Treasury Evaluation Task Force