The Middle Ten

The two earlier notes read the ends of the table, where the borough number is broadly honest. A borough that sits twenty points below expectation is restrictive on nearly everything; a borough twenty points above is generous on nearly everything. This note reads the middle: the ten boroughs whose differential is closest to zero. It is here that the headline figure is least trustworthy. A number near zero can mean a borough decides every kind of small site at roughly the expected rate. It can also mean a borough is generous on one typology and hostile to another, the two cancelling to a quiet average that describes none of its actual decisions.

I will argue that the middle is the one part of the table where the borough figure must not be read on its own. The same near-zero headline can sit on an internal range of around twenty points or on a range of barely two, and an applicant who reads only the headline will, in some of these boroughs, be precisely wrong about the site in front of them.

I Why the middle: the boroughs whose headline says least

The ten boroughs nearest zero on the differential run from Brent at −7.6 to Merton at +8.7, a band so narrow that, read as borough figures, they look interchangeably ordinary. None is materially restrictive in the way Havering or Croydon is; none is materially permissive in the way Bexley or Richmond is. On the league table they are the unremarkable middle. The argument of this note is that the middle is where the league table fails, because a small net number is the sum of larger gross ones, and the gross numbers are what decide a given application.

Read down the table and the middle splits into two kinds of row. Hillingdon and Greenwich are flat: every reliable cell sits on or beside the zero line, and the borough number means what it says. Lewisham and Redbridge are the opposite: a strong cell above expectation and a strong cell below it in the same row, a borough doing two contradictory things and reporting their average. Lewisham's +14 on backland beside its −6 on conversions is the cleanest split among the ten. The misleading headline is not a failure of the extremes, where the borough's posture is broadly uniform. It lives here, where opposite postures cancel.

II What the number hides: the range inside the borough

The heatmap shows the postures; the next chart shows the cost of averaging them. For each borough it draws the full span of its reliable type-level differentials, the distance from its harshest typology to its most generous, against a headline that, for all ten, sits close to zero. The bars are the borough as an applicant actually meets it. The labels are the borough as the table reports it.

Lewisham's bar runs from −6 to +14, a twenty-point internal range behind a headline of under two. Harrow's visible range is more than twenty as well, and understates the true figure: its extensions cell is generous (+27) but rests on 28 decisions, just below the thirty-decision bar, so it is suppressed. Brent's range is fifteen points, Redbridge's thirteen. At the other end Greenwich's bar is a point wide and Hillingdon's and Merton's only a little more. In the middle of the table, the size of the borough number tells you nothing about the size of the variation it conceals.

III A small number is the most dangerous number

At the extremes of the table the borough figure is a usable shorthand. An applicant in Havering can reasonably expect a hard time on most things; an applicant in Bexley can reasonably expect a fair hearing on most things; the single number ports across typologies because the borough's posture is broadly uniform. In the middle that shorthand inverts. The number is small because it is an average of opposites, so the one quantity it cannot describe is the typology in front of you. A developer who reads "Lewisham, roughly average" and proposes a conversion has walked into the −6 cell; the same developer proposing a backland scheme has walked into the +14. The headline gave the same advice for both, and it was wrong for both.

The differential is a real and well-behaved measurement, but it is a borough-level measurement, and a borough is not the unit at which permission is granted: a site of a particular type is. At the extremes the borough is a fair proxy for the site. In the middle it is not, and the middle is large: a third of London sits in this band. The correct response is not to distrust the number but to refuse to stop at it: to treat a near-zero headline as a flag that the borough must be read one typology at a time before it means anything at all.

IV Three things a near-zero headline can mean

Reading the ten internal structures against the ten plans resolves the small number into three situations. They are not always exclusive (a flat borough can also be a moving one), but each borough has a dominant reading, and the reading is what tells an applicant whether the headline can be trusted.

The terracotta rows are the boroughs where the headline actively misleads, because opposite postures cancel inside them. The gold rows are boroughs whose headline is honest today but is the wrong question, because the borough is in motion. Only the green rows are boroughs where a near-zero number can be taken more or less at face value, and even there, as Greenwich shows, "flat now" and "flat next year" are not the same claim.

V Cancelling extremes: Lewisham's twenty points

The cancelling boroughs are the ones the league table fails most completely, and Lewisham is the clearest case. Its adopted plan supports housing delivery and backland intensification in general terms, then gates conversions and mid-terrace subdivisions through amenity and garden-land tests that fall hardest on exactly those typologies. The same plan barely touches the open backland and corner plots that come with their own space and access, and these run well above expectation. The result is a borough that says yes to the open plot and no to the awkward subdivision, with a twenty-point gulf between them, reported as a single mild minus. Nothing in the headline tells an applicant which side of that gulf their site sits on, and the gulf is the whole decision.

Harrow is the same phenomenon with one arm hidden below the measurement threshold. Its conversion cell is on expectation and its end-terrace cell is firmly negative (−20, n=43), while its generosity on extensions cannot be drawn. Brent shows the pattern at the restrictive edge of the band: an affordable-housing contribution on schemes of five to nine homes presses down on conversions (−11, n=242) and backland (−13, n=80), while demolish-and-rebuild sits a touch above expectation (+2, n=111). Redbridge repeats the structure more mildly: a conversion brake holding subdivisions a few points below expectation while rebuild, end-terrace and extension all run above it. The same shape recurs across the outer boroughs: friction trained on the awkward, amenity-poor conversion sitting beside a lighter touch on new build, and wherever the two sit together they manufacture a cancelling middle.

VI Genuine balance: Merton's designed equilibrium

Not every small number is a deception. Merton's +8.7 sits on an internal range of three points: conversions +9 (n=129), rebuild +10 (n=86), mid-terrace +7 (n=43). This is not opposite postures cancelling; it is a borough that has built a level surface. Its adopted plan, November 2024, opens the door to small sites about as wide as any in outer London and then pairs that openness with two calibrated frictions applied across the board: a requirement to re-provide a family-sized home wherever one is lost, and an affordable-housing contribution on schemes as small as two units. Openness plus uniform friction yields a borough that is mildly, evenly generous, and whose headline is therefore a fair description of almost any site within it. Hillingdon reaches flatness from the other direction: a genuinely restrictive garden-and-backland policy drawn so narrowly it barely touches the mainstream cohort. In both, the near-zero number is honest. The difference between them and Lewisham is not the headline. It is everything the headline omits.

VII Movers: the plan as a verb

The third reading is temporal. A small number can be an honest snapshot of a borough caught mid-change, where the relevant fact is not the level but the direction. Wandsworth is the clearest case: its differential is a mild +4.0, but the borough changed political control in 2022 for the first time in over four decades, and its 2026 plan review has begun rewriting the terms on which small sites are consented, including a per-dwelling affordable contribution on schemes of fewer than ten units. None of that has yet worked through into a settled differential, because plans act on the flow of new decisions, not the stock of old ones. The headline describes the borough that was; the plan describes the borough that is arriving. Enfield is in the same position from a different start: its decisions in the data fall under a Core Strategy and Development Management Document from 2010 and 2014, while a new Local Plan, submitted for examination in 2024, is expected to replace them. Greenwich is moving the other way, from an already measured base toward a tighter one.

The outcomes in all three companion notes are drawn from decisions between 2022 and 2026, and several of these middle boroughs adopted new plans inside that window or changed political control at its very end. Harrow adopted a new plan in March 2026. Lewisham and Hackney both passed from long-standing Labour control to Green administrations in May 2026, after the last decision in the dataset. A differential is always a measurement of a borough's recent past. In the stable boroughs of the first two notes that past is a fair guide to the present. Among the movers of the middle it is a guide to a regime that may already be gone, which is simply another reason the borough number, here, is where analysis starts rather than stops.

VIII What this means: where the detail is mandatory

The three findings give a single rule for reading the differential. At the restrictive end, the borough number is broadly honest, and the work is to decide whether the restriction is legitimate. At the permissive end, the borough number is broadly honest, and the work is to identify which of four mechanisms produced it and whether it can be copied. In the middle, the borough number is not reliably honest at all, and the work is more elementary: to recover the typology-level structure the average has erased before drawing any inference whatsoever. The extremes reward interpretation. The middle first demands disaggregation.

For an applicant the operational consequence is sharp. A near-zero borough on the league table is not a signal of safety or of risk; it is a signal that the league table has stopped being informative and the site-type cut must take over. In Lewisham the relevant number is not −1.9 but +14 or −6, depending entirely on what is being built. In Merton the headline can be trusted. In Wandsworth the headline is already out of date. The three look identical from the table and demand three different readings.

IX Conclusion

Ten identical positive numbers can be produced four different ways; two identical negative numbers can mean opposite things. In the crowded middle of the table, a small number is not a description of a borough but a sum that has cancelled its own information away. Behind Lewisham's −1.9 lies a twenty-point internal split; behind Hillingdon's −1.7 lies almost nothing; behind Wandsworth's +4.0 lies a borough that no longer decides the way the number says. The differential found the variation, and at the extremes it reported it faithfully. In the middle it hid it, not by error but by arithmetic, and the only remedy is to read past the borough to the site. That is what the whole instrument was built to make possible: not a league table of boroughs, but a way of seeing the specific decision a specific site is about to meet.

Sources & confidence

Outcome differentials and type-disaggregated cells: LSSPD, build 4 June 2026 (borough_differential.csv, type_differential_cells.csv). Internal range = maximum minus minimum type-level differential among cells with ≥30 decisions; cells below that threshold are suppressed, so the stated ranges are conservative (Harrow's true range is wider; Hackney has too few reliable cells to plot). Plan audit: Lewisham Local Plan (adopted 16 July 2025); Harrow Local Plan 2021–2041 (adopted 24 March 2026); Brent Local Plan 2019–2041 (adopted 16 February 2022); Redbridge Local Plan (adopted 15 March 2018); Wandsworth Local Plan (adopted 19 July 2023, partial review adopted 4 March 2026); Enfield Core Strategy (2010) and Development Management Document (2014), new Local Plan submitted 2024; Hackney LP33 (adopted 22 July 2020); Greenwich Core Strategy (adopted 30 July 2014); Merton Local Plan 2024 (adopted 20 November 2024); Hillingdon DM Policies (adopted 16 January 2020, Policy DMH6).

What this can and cannot tell you. The descriptive claims are firm: these ten boroughs have near-zero headline differentials, and their internal type-level ranges differ enormously, from around two points (Greenwich, Hillingdon) to roughly twenty points (Lewisham, Harrow). The attribution of each internal pattern to specific policies is an interpretation supported by the adopted plan text, not a controlled test. Political control and plan status are given for the 2022–2026 data window; several of these boroughs adopted new plans within it or changed control in May 2026, after the last decision in the data. The figures locate where a borough's variation lives. They do not predict the outcome of any individual application.