The Affordable Ten

The first paper established that London runs more than thirty planning systems for small sites, that the gap between the most and least permissive borough is about forty-three percentage points once caseload mix is controlled for, and that the gap tracks house prices. It stopped short of the question every reader actually wants answered. When an affordable borough sits twenty points below expectation, is that a legitimate local choice written into an adopted plan, or an unexplained posture the plan does not account for?

Section 7.2 of that paper named the two categories. We run the test here on the ten boroughs where the answer matters most, and it returns a clear result: the shape of a borough's restrictiveness predicts whether the plan can account for it, and the audit confirms the prediction.

§1 Why these ten: the natural experiment

The price–approval gradient in the first paper is a correlation across all thirty-two boroughs, and a sceptic can always say price is doing the work. The clean way to kill that objection is to hold price roughly constant and see whether outcomes still fan out. London's ten cheapest boroughs by median house price do exactly that. They sit inside a £107k band, at the bottom of a distribution that runs to more than a million pounds, and span almost the entire differential range in the dataset.

Forty points of outcome spread across roughly a hundred thousand pounds of median price. Affordability cannot be the mechanism inside this band, because it is near-fixed against London's full range and the outcomes are not. Whatever separates Bexley (+18.9) from Havering (−21.4) at near-identical prices is plan, politics, practice, or applicant quality. This essay reads the plans to find out which.

§2 What the borough number hides

The headline differential is one number per borough. Disaggregating it to site type (the same mix-adjusted residual computed inside each borough-by-type cell, with any cell below thirty decisions suppressed) changes the picture in a way that turns out to be the whole story. A borough that is hard on everything is exercising a general posture. A borough that is hard on one type the plan explicitly constrains is applying a policy. The shape of the restrictiveness is itself the evidence.

Uniform restrictiveness. Croydon is the clearest case in the dataset. It sits below expectation on every type with enough data to measure: conversions −20, demolish-and-rebuild −22, backland −22, extensions −19, end-of-terrace −32, mid-terrace −25. No single typology carries the borough; the whole surface is depressed. Havering is the same shape on the two types it has volume in. Restrictiveness this evenly spread is hard to read off a development plan, because plans constrain types of development, not development in general.

Concentrated restrictiveness. Newham is hard on conversions (−20) but only mildly below on demolish-and-rebuild (−2) and extensions (−10). Barking & Dagenham's one reliable cell is conversions at −29. The negative differential is a conversions story, which a loss-of-housing-stock or space-standards policy could explain.

The split. Hounslow's headline is −13.1. Underneath it, conversions run −22 while demolish-and-rebuild runs −2, with mid-terrace at −18. A developer reading "Hounslow, below average" would never guess the borough is hostile to subdivision while treating rebuild as ordinary.

And two boroughs sit on the line for different reasons. Greenwich decides at close to the expected rate type by type; its small negative headline is genuinely noise. Lewisham's near-zero headline hides more: a generous +14 on the open backland plot beside a −6 on conversions, a cancelling split this series reads in The Middle Ten rather than a general restriction. Neither is materially restrictive. The working hypothesis from this cut was simple: uniform restrictiveness is the pattern least likely to be explained by a plan; concentrated restrictiveness is the pattern a plan can own. Section 4 tests it. It holds.

§3 The language does not settle it

Before reading ten plans, the natural test is whether the words boroughs use when they refuse already give the answer. Scoring every refusal reason for discretionary character-and-harm language across the twenty-seven boroughs with enough refusals, the borough-level intensity of that language barely correlates with the differential. The decision corpus is too contaminated to carry the argument. Refusal text is written to justify a decision already taken, so it follows the outcome rather than revealing the stance behind it. Coverage is uneven and refusal-only. And boroughs simply write differently: Croydon codifies its restrictiveness into design-standards language, while verbose authorities pile on character language regardless of how permissive they actually are.

At the level of the marquee pair the signal is sharp. Havering's refusals are roughly three times as character-laden as Bexley's. The stance is real where it is strongest, but a single text index cannot separate the middle of the field. The honest conclusion is the one the first paper anticipated. The decision corpus describes how a borough talks. Only the upstream corpus (the adopted plan, the politics, the appeal record) can establish whether the talk is licensed. That is the audit, and it cannot be skipped.

§4 The verdicts

The test is a conjunction. A borough holds a Category 1 (legitimate) position when its adopted plan contains a materially restrictive small-sites policy that goes beyond the London Plan's Policy H2 duty to pro-actively support well-designed small homes, its refusals cite that policy and fall on the typologies it constrains, and the Planning Inspectorate broadly upholds them. It holds a Category 2 (unexplained) position when the plan is broadly H2-conformant, the refusals nonetheless cluster on discretionary character grounds the plan does not compel, and the appeal record, where there is enough of it to read, shows those refusals resting on policies that do not themselves support them.

Havering and Croydon are the unexplained pair. Havering defends its restraint openly: its examination statement called suburban intensification "ambitious to the point of absurdity". But that defence is a town-centre growth strategy, not a small-sites policy. Its development-management policies (7, 10, 26) are the ordinary character, design and amenity tests almost every borough carries; there is no density cap, no garden-development bar, no conversion restriction in the adopted text. And when an applicant appealed a backland refusal at 20 Parsonage Road, the Inspector found the scheme complied with Havering's own Policies 10, 7 and 26 and allowed it. Restrictiveness applied uniformly across types, through discretionary character judgement, on policies that do not themselves support refusal, is the definition of Category 2. Croydon is the same verdict by a different route: its restrictive instrument, the SPD2 suburban design guide, was revoked in July 2022, yet approvals stayed low and uniform across all six measurable types. The plan that remains is broadly conformant and even contains an intensification policy. The restrictiveness is political and cultural, not written down.

Barking & Dagenham, Newham and Hounslow are the legitimate hard boroughs, and they are legitimate for the same reason. Each is hard on conversions specifically, and in each case a policy, adopted or emerging with growing weight, deliberately resists the subdivision of family-sized homes into flats. Newham's family-housing protection, carried in its emerging Policy H4 on a longstanding priority, supports keeping family homes whole and turning flats back into houses. Barking & Dagenham's DMH5 and SP3, adopted in September 2024 and cited in draft through the data window, resist the loss of dwellings capable of housing a family with children. Hounslow's SC6 sets a 130 m² original-floor-area threshold, explicitly refuses conversions of properties extended to reach it, and requires a ground-floor family unit. All three policies have been upheld at appeal. And all three boroughs treat demolish-and-rebuild as ordinary new housing, which is exactly why their differentials concentrate on conversions and leave rebuild alone. Hounslow proves the mechanism cleanest: the −22 on conversions and the −2 on rebuild are not a contradiction, they are two different policies (SC6 versus SC4) doing what they say.

The appeal record: a refusal is, in practice, the end of the road

What the Inspectorate adds to this picture is precise: the instinct on reading a low number is to assume the refusals are bad decisions that an appeal would correct. They are not, and the correction rarely comes. Across London's matched small-site appeals only about six in every hundred refusals are challenged at all, and of those that are, the boroughs win more often than they lose. The pattern runs the counter-intuitive way: the hardest boroughs are overturned at roughly sixteen per cent of decided appeals and the most permissive at around twenty-four, so restrictive boroughs are upheld more often, not less (r = +0.40 between difficulty gap and overturn rate). Whatever a persistent negative differential is, it is not a borough making capricious refusals that the Inspectorate routinely reverses. The differential is settled at first instance, which is exactly why it matters, because for the overwhelming majority of refused schemes the borough's first decision is the only decision there will ever be.

Inside the affordable ten the same pattern holds, and at the legitimate end it holds cleanly. Newham is the sharpest case: of the small-site appeals the Inspectorate has decided it upheld every one, none overturned, including all of the conversion refusals, which is as close as this record comes to certifying that the family-housing policy is doing lawful work. Hounslow is the honest wrinkle in the other direction: its refusals are overturned about a third of the time, so the SC6 gate is contested at the margin even as it survives in the main, a reminder that legitimate is a verdict on the policy and not a guarantee on every case.

And here the appeal record runs into the same wall the refusal language did in §3. For the two Category 2 boroughs it cannot speak: Havering and Croydon each have too few matched small-site appeals to compute a rate, so the aggregate falls silent precisely where the question is hardest. What it offers instead is a single decisive case. When an applicant appealed a backland refusal at 20 Parsonage Road, the Inspector did not merely allow the scheme: he found it complied with Havering's own Policies 10, 7 and 26, the very tests the refusal had invoked. One case cannot convict a borough, but it can show the shape of the problem: a refusal resting on policies that, read by an independent tribunal, do not support it. The aggregate overturn rate cannot separate the legitimate hard boroughs from the unexplained ones: only the upstream plan audit can. The Inspectorate confirms the floor; the plan draws the line.

Appeal figures: LSSPD matched-appeal cohort, 458 decided small-site appeals, 2022–2026; London overturn 21.0%. Only boroughs with ≥5 decided appeals carry a rate; ~6% of refusals are appealed, so the sample is selected by applicant judgement. Havering appeal: APP/B5480/W/24/3355037, allowed.

§5 What "restrictive" turns out to mean

The audit reframes the moral content of a negative differential. The instinct from the first paper is that a low affordable-borough number is a regressive outcome. For Havering and Croydon, that instinct survives: they suppress small-site housing across the board with no policy account, and the homes that go unbuilt are a real loss. But for Newham, Barking & Dagenham and Hounslow, the low number is the visible cost of a deliberate, defensible choice to protect family-sized homes from being carved into flats in some of London's most overcrowded boroughs. That is not capricious gatekeeping. It is a housing-mix policy with a distributional logic of its own, and it is exactly the kind of legitimate local choice the first paper said the differential must leave room for.

So the differential alone cannot tell you whether a borough is wrong. It tells you where to look. The type cut tells you what kind of restrictiveness you are looking at. The plan audit tells you whether it is licensed. Only at the end of that chain does a verdict appear, and when it does, two superficially identical numbers, Havering at −21 and Newham at −15, turn out to mean opposite things.

§6 The permissive end is not one thing either

The three affordable boroughs that sit above expectation get there by three different routes, and the distinction matters for anyone trying to copy them. Bexley is permissive by plan design: its 2023 plan is among London's newest and tells applicants outright that backland and infill "will generally be encouraged", with small-site windfall built into its housing maths. It says yes because it has decided to, and written it down. Enfield is permissive by political-delivery posture: its adopted plan is a decade old and silent on small sites, but a Labour administration that took control in 2022, carries a Housing Delivery Test buffer, and is steering a new plan through examination has every incentive to approve borderline schemes and defend them. Its permissiveness is administrative, not policied. Sutton is permissive by structural necessity: its 2018 plan is actually cautious about intensification, but the borough depends on small sites for roughly 79 per cent of its housing target, the highest reliance in London, and an ageing plan carries less weight against the small-sites presumption with every year. It says yes because it has to.

Only Bexley's permissiveness is durable, because only Bexley's is written into an adopted plan. Enfield's depends on who runs the council; Sutton's depends on a plan that is about to be replaced. The lesson for the hard boroughs, and for anyone recommending they soften, is that the permissive examples worth emulating are the ones with a policy behind them.

§7 The spine: Bexley versus Havering

Everything converges on the marquee pair. Fifteen thousand pounds apart on price, forty points apart on outcome, and now we can say why. Bexley's 2023 plan encourages the exact small-site development that Havering's discretionary character judgement resists, and Bexley has written its encouragement down while Havering's resistance lives nowhere in its plan. The decision-culture reading and the applicant-quality reading both remain live for the residual, but the first-order explanation is now visible and it is policy posture: one borough adopted a permissive small-sites policy and the other did not, at the same price, under the same London Plan. Bexley is the standing proof that affordability does not oblige a borough to sit below expectation, and the productive question it puts to Havering is not an accusation but an enquiry the first paper already framed: what are you doing that Bexley is not?

§8 What this means, and what would settle the residual

For the boroughs, the audit hands each of the ten a finished version of the one-page account the first paper asked them to write. Three of them (Newham, Barking & Dagenham, Hounslow) can point to a named, appeal-tested policy and say the differential is the price of protecting family housing. Two of them (Havering, Croydon) cannot, and on this evidence their restrictiveness belongs in the transparency regime the first paper proposed, not in a plan-examination defence. Two more (Lewisham, Greenwich) do not belong in the restrictive narrative at all: Greenwich decides at close to the expected rate across the board, and Lewisham's near-zero headline conceals a backland-versus-conversion split rather than a general restriction, the case The Middle Ten takes up. And the three permissive boroughs are the worked examples, with the caveat that only one of the three has made its permissiveness durable.

For the open question the first paper could not close (how much of the differential is decision-culture and how much is applicant quality), the residual now sits inside the two Category 2 boroughs, because everywhere else the differential is either explained by policy or absent. That is a far smaller and sharper target than "the affordable boroughs", and it is the place to run the experiment the first paper named. If under-prepared applications are part of why Havering sits at −21, then raising preparation quality should move its number. A small-sites design code with a fast-track for compliant schemes, or a productised preparation service pitched at this price band, run as a prospective trial in Havering and Croydon, would put a number on the share of the gap that is preparation versus posture. It is the one thing this essay cannot do from the record, and the one thing that would finish the argument.

§9 Conclusion

The first paper asked each borough a fair question: given what you received, what outcome did you produce? This essay narrowed it to the ten boroughs where price cannot be the excuse, and then did the work the first paper deferred: it read the plans. The result is not a league table. It is a map with the moral content put back in. Two boroughs are hard for reasons their plans do not contain and their Inspectors do not uphold. Three are hard for a reason their plans state plainly and defend successfully on appeal. Two are not really hard at all. Three are permissive by three different mechanisms, only one of them durable. The differential found the variation; the type cut diagnosed its shape; the audit decided its legitimacy. None of it tells a borough what to choose. All of it tells each borough, and the city that depends on the homes, exactly which question it now has to answer.

Sources & confidence

Outcome differentials and type-disaggregated cells: London small-sites planning dataset (LSSPD), build 4 June 2026 (type_differential_cells.csv). Refusal-stance index: coded refusal-reason corpus (refusal_stance_index.csv, 27 boroughs with ≥30 reasons). Plan audit: each borough's adopted (and emerging) Local Plan, cross-referenced with Housing Delivery Test results and Planning Inspectorate decisions. Appeals: Havering APP/B5480/W/24/3355037 (20 Parsonage Road, allowed); Barking & Dagenham APP/Z5060/W/25/3370953 (17 Campsey Road, dismissed); Newham flat-to-HMO (house-in-multiple-occupation) appeal (dismissed, 2024).

Verified against primary sources (June 2026): Bexley adopted Local Plan (26 April 2023): Policy DP2 confirmed verbatim ("Proposals for new dwellings on small (less than 0.25ha) brownfield sites, including infill and backland sites, will generally be encouraged…"); windfall 3,050 over ten years (≈305/yr) under Policy DP1. Enfield windfall 281 dwellings/yr over 2012–2022 (≈48% of delivery), Housing Topic Paper 2024 para 3.27; adopted plan Core Strategy 2010 + DMD 2014, new plan at examination (submitted 6 Aug 2024). Housing Delivery Test 2023 measurement: Havering 61% (Presumption), Croydon 160% (Pass), Barking & Dagenham 66% (Presumption).

Correction: there is no Newham Local Plan adopted December 2023. The adopted plan is the 2018 Newham Local Plan; family-housing Policy H4 sits in the emerging plan (submitted for examination July 2025, not yet adopted as of June 2026). Newham's Category 1 verdict is therefore provisional, resting on the borough's longstanding family-housing priority and the 2024 appeal rather than on adopted Policy H4.