Heading into the election.
What they promised.
What they did.
Before you vote, a record of what the last two governments said they would do, and what they actually did. Every election and coalition promise from five parties, across Labour’s term from 2017 to 2023 and the National-led coalition since. Each one scored on whether it was delivered, not on whether it was a good idea.
The short version
If you read nothing else here, read this. It sets out what the numbers below show, and what they do not.
On delivery alone, there is very little between these two governments. Labour kept 55% of the promises it could be judged on, over six years. The National-led coalition has kept 58% in two years and nine months, and has 11 promises still running that could lift it to 64%. Across 347 promises, three points apart is a tie.
What really separates them is what they promised in the first place, and the conditions they were working under.
- Time in office
- 6 years, 1 month
- Promises tracked
- 152 · 25 per year
- Promised before polling day
- 57% manifesto pledges
- Kept most of
- Foreign affairs & defence 83%, from 6
- Kept least of
- Transport & infrastructure 20%, from 10
- Biggest constraint
- Covid-19
- Time in office
- 2 years, 9 months unfinished
- Promises tracked
- 195 · 71 per year
- Promised before polling day
- 35% manifesto pledges
- Kept most of
- Law & order 83%, from 12
- Kept least of
- Cost of living & tax 38%, from 8
- Biggest constraint
- Three-party coalition
1Labour had three times as long, and lost much of it to Covid
Six years sounds like a big advantage over two years and nine months, and it is. But Covid-19 arrived in March 2020, near the end of Labour’s first term, and it dominated the second. The border stayed shut for two and a half years. Parliament’s agenda, the Budget and most of the public service’s capacity went into the response. Promises with nothing to do with the pandemic were put off, then quietly dropped. Light rail, the rental warrant of fitness, the capital gains tax and social unemployment insurance all went this way.
This cuts both ways, and this site does not settle it. Covid is a real reason many promises were not kept. It is also true that a government is judged on the term it had, and that Labour won a larger majority in 2020, in the middle of the pandemic, on a fresh set of promises it also underdelivered. Both views are fair. What is not fair is applying the excuse to one side only.
2National has promised more, faster, and its promises were easier
195 promises in 33 months, against Labour’s 152 in 73 months. That is close to three times the rate. But the volume is not the interesting part. The composition is. 65% of National’s commitments come from signed coalition agreements, written after the election by people who already knew what could be delivered. Only 43% of Labour’s do. Across this dataset, coalition agreement commitments are kept 61% of the time and manifesto pledges 51% of the time.
The same pattern shows up in the kind of promise each party made. Promises to do a thing, like passing a law or abolishing a body, are kept 70% of the time by anyone. Promises to hit a number, like cutting hardship by so much or building so many houses, are kept 36% of the time by anyone. Put both governments on the same mix of promise types and the gap between them gets smaller, not bigger.
3National’s term is not over, and that shapes the score
Every single unfinished promise in this dataset belongs to the current government, because the other one ended in 2023. That creates a trap. Leave unfinished promises out of the score and National gets an exemption Labour cannot use. So an unfinished promise counts in the denominator, but not as a success. Labour’s incomplete long-term targets are treated exactly the same way.
Some commitments still described as under way are, by the record’s own account, moving away from their target. Jobseeker numbers are up against a pledge to cut them by 50,000. Material hardship is up against a pledge to reduce it. Those are scored as partly delivered rather than left open, which is why this section matters more than it looks.
4The real difference is what they delivered on, not how much
Both governments kept roughly three promises in five. They kept different ones. Labour delivered on trade and foreign policy, on housing rules and on climate law. It failed almost completely on transport and on the health system it had just restructured. National has delivered on law and order and on energy. It has fallen shortest on cost of living and tax, which is the ground it won the election on, and on foreign affairs and defence.
Read the area-by-area table rather than the headline. A government can keep 60% of its promises and still have kept none of the ones you voted for.
Two governments, side by side
Each bar is every promise a party made in that term, sorted by what became of it. The big number is the share of judgeable promises it kept. An unfinished promise counts against the score. That is the same rule applied to Labour’s unfinished targets when it left office in 2023.
Who kept what
Junior coalition partners are scored on their own promises, not their government’s. That is why a support party can beat the party leading the government, or fall well behind it.
Every promise
Filter, search, and open any promise to see what happened and where that comes from. Grouped by policy area.
Head to head, area by area
The same eighteen policy areas, both terms. Holding the subject steady is the fairest way to compare, but it also thins the sample out. Most cells rest on five to ten promises. A figure marked * comes from fewer than ten. n/a means fewer than six, where a percentage would be noise. Read the bars, not just the numbers.
NZ First: the same party, two governments
NZ First is the only party here to have governed in both terms. It was with Labour from 2017 to 2020, and has been with National since 2023. It is the closest thing to a controlled experiment New Zealand politics offers.
What actually moved
Promises are one thing; outcomes are another. 17 official series covering both terms. The tinted band behind each line shows who was in office. The hatched band marks Covid-19, from March 2020 to the full border reopening in August 2022. Under each chart is the change during each government’s own stretch, so you can read the two terms separately.
Method, and what to distrust
A scorecard is only as good as its rules. Here are the ones used, including where they are weak.
The seven outcomes
What counts as a promise
Two kinds, tagged separately throughout. Manifesto promises are what a party told voters at the election. Coalition agreement commitments are what it actually signed up to in government: the 2017 Labour and NZ First deal, the 2017 and 2020 Labour and Greens agreements, and the 2023 National agreements with ACT and with NZ First.
Where a party’s own manifesto policy was traded away in coalition talks, both are recorded and the manifesto one is marked superseded. Specific, checkable promises were preferred over aspirations.
How the score is calculated
The percentage is kept ÷ (kept + broken + in progress). Kept means delivered or largely delivered; broken means partly, not delivered, or abandoned.
An in-progress promise counts in the denominator, but not as a success. This matters more than it sounds. Every in-progress promise in the dataset belongs to the current term, so leaving them out would have given National a break Labour could not use. Promises superseded in coalition talks are left out completely, because the party never got to attempt them.
A percentage is not shown at all where fewer than 6 promises are scored. Small samples produce confident-looking numbers that mean nothing.
Three things that can skew a scorecard like this
Delivery scores are easy to tilt without meaning to. These three effects were tested for, and handled as follows.
- Unfinished promises can only help one side. Every in-progress promise belongs to the current term, so excluding them would flatter it and nothing else. They count in the denominator instead. Commitments the record itself shows moving away from target are scored as partly delivered, not left open.
- The kind of promise matters more than the party. Promises to do a thing are kept 70% of the time across the whole dataset. Promises to hit a number are kept 36% of the time. A party that campaigned mostly on one kind will score differently for reasons that have nothing to do with competence, so the like-for-like column reweights every party to the same mix.
- Coalition commitments are co-signed but recorded against one party. All 19 broken signed-agreement commitments sit with ACT or NZ First, while National’s own agreement list shows no outright failure. That is a bias by role rather than by politics, and it flatters whichever party leads a government. It is not corrected for, so read partner scores with it in mind.
The risk in a scorecard like this is rarely loaded language. It is almost always the counting rules.
The comparison is not symmetrical
- Labour governed for six years; National has governed for two years and nine months.
- Labour’s term included Covid-19, which took over the agenda from 2020 to 2022 and killed off promises unrelated to it.
- Labour held a single-party majority from 2020. National governs with two partners whose agreement it needs.
- National’s term is unfinished. Some promises scored here will change status before the election.
On the Treaty section
This is the most contested material on the site, and it was handled with extra care. Policies are described by what the law actually did, in the parties’ own neutral terms.
Where the significance of a change is disputed, and it often is, both readings are set out in the promise’s contested note and the site does not pick one. Delivery is scored on whether the stated commitment happened. That is a narrower question than whether it was right.
Known weaknesses
- Several manifesto PDFs from 2017 and 2020 could not be retrieved. Those promises cite the strongest available record of the commitment instead of the manifesto itself.
- The Greens’ number is the least trustworthy on the site. It rests on 33 promises, 42% of them targets for things outside the party’s control, and five of them pledges in portfolios the Greens never held in Cabinet. Several cooperation agreement clauses have nothing measurable in them and are still counted. Use the like-for-like column, and treat even that with caution.
- Treaty and Māori policy is the hardest area to score fairly and the one to read most sceptically. It carries 17 promises from the Labour years and 13 from the current term.
- 0 promises are flagged contested, meaning the status call is genuinely arguable. Open them and judge for yourself.
- Promise counts differ by party, so a percentage drawn from a small sample is less reliable than one drawn from a large sample. Counts are shown everywhere for that reason.
- This is one round of research, not a peer-reviewed audit. Treat it as a starting point with its sources attached, not a final verdict.
Sources
Primary sources were used wherever possible: the signed coalition agreements, legislation.govt.nz for what actually passed, Beehive releases, Treasury Budget documents and departmental data. New Zealand news outlets, including RNZ, the NZ Herald, The Post, Newsroom, The Spinoff and BusinessDesk, were used to check them.
Every promise carries at least one link to the promise and at least one to the evidence of what happened. 0 distinct source links in total.
Check the work
Everything behind this page is published. The full dataset, every source link, the outcome series and the scripts that build the site are all in one repository. If you think a call is wrong, the evidence is there to argue with.