How We Reach Conclusions

1. What "Scientific" Means for a Policy Project

This project does not run laboratory experiments. It cannot randomise Britain into a treatment and control group. When we say the approach is scientific, we mean something narrower and more honest: the same epistemic discipline that makes science work is applied to how claims are formed, checked, and revised, even though the subject matter — tax policy, healthcare, housing, trade — cannot be tested the way a hypothesis in physics can.

That discipline has a small number of actual components, and they are worth naming precisely rather than invoking "evidence-based" as a slogan.

Evidence before conclusion, not conclusion before evidence. The ordinary failure mode in political argument is to start with a preferred conclusion and select the facts that support it. The discipline this project tries to hold itself to runs the other direction: establish what the evidence actually shows first, and let the conclusion follow from it — including when that conclusion is inconvenient for the argument being made.

Claims are provisional, not settled. A published figure or argument on this site is the current best account given what has been checked so far, not a permanent verdict. It can be revised or withdrawn if better evidence emerges. This is not a weakness in the method. It is the method.

Distinguish observation from inference. A fact independently verifiable from an official source (a stat from ONS, a modelled outcome from the OBR) is a different kind of claim from a conclusion this project has drawn by connecting two or more such facts together. Both can be right. They carry different weight, and conflating them is a common way arguments quietly overstate their own certainty.

2. The Four Practices, and What Each Looks Like in Practice

Every factual claim is sourced and independently checkable

Every figure cited in a pillar links to a specific row in the Sources Register — institution, publication, and what it is used for. This is not a bibliography for its own sake. It means any reader, or any future editor of this project, can trace a number back to where it came from and check it themselves, rather than trusting it because it is written down confidently.

The strongest version of an opposing argument is presented, not the weakest

Most political writing argues against a caricature of the opposing position because caricatures are easy to beat. This project's pillars each carry a dedicated Steel Man section specifically because arguing against a strawman proves nothing except that the writer can construct one.

This is not a hypothetical standard. In practice it means going back and fixing it when we fall short of it. The Trade pillar originally characterised the Remain campaign's position on international trade in a way that, on closer research, no mainstream figure had actually argued — the real, defensible version of that argument (short-term economic forecasts that proved too severe, not a claim that trade with the world would simply stop) was weaker evidence for our side of the argument than the strawman had been, and we published the correction anyway.

Original analysis is labelled as analysis, not presented as an external finding

When this project connects two independently sourced facts into a new argument neither source made on its own, that connection is marked explicitly as this project's own synthesis, not attributed to the sources as if they had said it. The Local Government pillar's section on ageing demographics and council tax base geography is a direct example: the underlying facts are independently sourced from ONS and the Centre for Ageing Better, and the argument connecting them — that the areas facing the fastest-growing social care costs are structurally the same areas with the weakest-growing tax base — is flagged in the text itself as this project's own analytical contribution.

Claims are rejected or downgraded when they don't survive a fresh check — even after they were already drafted

A commitment to evidence is only real if it sometimes costs you something you already wrote. In the last week of work on this project alone: council tax figures for Scotland were dropped from the Local Government pillar entirely after a fresh verification pass found they remained internally contradictory even in supposedly reliable secondary sources — rather than publish a number we could not resolve. A detailed economic case for agrivoltaic farming had grown, over several editing sessions, into one of the Agriculture pillar's core arguments, driving several formal proposals — on review, the underlying technology was still trial-stage and the case did not meet the evidentiary bar the rest of the pillar was held to, so it was rebuilt down to a single honest paragraph and the full speculative case was moved to a separate working paper with an explicitly lighter evidentiary standard, clearly marked as such.

Every proposal is built through a 5-stage chain, and carries a test that could prove it wrong

The four practices above describe how individual claims are checked. This practice describes how a claim becomes a policy proposal — the step where evidence-based analysis is most often quietly smuggled into ideology, because a real fact at the top of an argument doesn't automatically justify a specific policy at the bottom of it.

Every major proposal on this site is built through five stages, shown directly on the page rather than left as an internal drafting note: Evidence (what the data actually shows, tagged for how solid it is — FACT, SYNTHESIS, CONTESTED, or UNVERIFIED), Causation (what mechanism is being claimed, and how strong the causal case is versus merely correlational), Options (the range of interventions the causation would justify, not just the one preferred), Values (what's doing the selection work between those options — stated explicitly, not smuggled in as if it were also evidence), and Proposal + Test (the specific reform, plus a falsification test attached to it).

That falsification test is the load-bearing addition: a prediction of what should happen if the proposal is right, how big the effect should be, over what time horizon, against what counterfactual, and — most importantly — what result would make this project abandon or revise the proposal. A policy argument that cannot say what would prove it wrong is advocacy wearing evidence-based language, not evidence-based policy.

This is new, and retrofitting it across every existing pillar is a large undertaking not worth stalling active work for. It is being applied going forward, piloted first on Economic Renewal's flagship proposal — see the chain and falsification test attached to the 95% inheritance tax proposal — with two further retrofits since, on Wealth Tax Comparison and Housing, and wider retrofit across the rest of the site continuing opportunistically rather than all at once.

3. Where the Method Has Limits

Not every question this project could cover is a question evidence can answer on its own, and pretending otherwise would be its own kind of dishonesty.

Some public debates are primarily questions of value, not questions of fact — where reasonable people fully informed of the same evidence still land in different places because they weigh competing goods differently. Assisted dying legislation is a clear example, and it is deliberately not covered by this project for exactly that reason: the empirical questions involved (safeguarding efficacy, international outcomes data) are real and checkable, but the core disagreement is about the value of autonomy against the value of protecting the vulnerable, and no amount of additional evidence resolves that trade-off. Forcing an "evidence-based" verdict onto a question that is not primarily about evidence would misrepresent what evidence can do.

More broadly, even the pillars this project does cover are not value-free simply because they are evidence-based. A conclusion that a given trade-off is worth making still rests on a judgement about whose costs and benefits should weigh more heavily — the scientific discipline here is about getting the underlying facts right and reasoning honestly and transparently from them to a conclusion, not about eliminating judgement from the process altogether. Readers who weigh the same evidence differently are not thereby wrong; they may simply hold different values, openly disclosed rather than smuggled in as if they were facts.

4. What This Project Is Not

This is not primary research. No claim on this site is the product of an original study, survey, or experiment run by this project. Every factual claim traces back to existing published data, government statistics, or academic research — this project's own contribution is synthesis, argument, and policy design built on top of that evidence, not the generation of new evidence itself. Where that synthesis goes beyond what any individual source claims, Section 2 above describes how that is meant to be flagged.

This is not a neutral aggregator either. A project that simply summarised "both sides" without ever reaching a conclusion would be abdicating the actual work of policy analysis, not practising objectivity. Non-partisan means not aligned with a political party and not selecting evidence to flatter one — it does not mean declining to reach a view.

This is also not the only project doing work like this, and it should not be read as if it were. Comparable Projects sets out the closest UK equivalents honestly, including where they disagree with the conclusions reached here.

5. How to Hold Us to This

The claims in this document are themselves checkable, using the same public record every pillar on this site is checkable against.

The Sources Register shows exactly what every factual claim across the project traces back to.

The Gaps Register is where this project logs its own unresolved limitations publicly, rather than letting them go unmentioned — currently 26 open items (of 31 logged in total; 5 fully resolved), ranging from figures needing better primary sourcing to structural questions a pillar has not yet fully resolved.

The Update Register records what changed in any published document, and why — including the corrections described in Section 2 above.

If you believe a claim on this site has not actually been held to the standard described here — a source that does not check out, a strawman that survived, an original argument presented as if it were an external finding — cite the specific claim and the specific evidence, and it will be checked against the same standard applied everywhere else.