Planning definitions:
where does the trend change sign?
A planning headline can depend on which records count. This tool traces the space between two published definitions and identifies the exact point where a rise becomes a fall.
One crossover in 16 tests. The East of England's annual comparison changes direction at a 47.96% retained weight on the mentions excluded by the reserved-matters filter. The other 15 comparisons keep their direction throughout this particular test. That does not establish robustness to other definitions or later revisions.
Move the definition, inspect the result
At 0%, all mentions caught by the source's reserved-matters text filter are excluded. At 100%, all are restored. Intermediate values are an illustrative common weight applied to both periods, not observed housing counts, a probability or measured classification accuracy.
-26.6%
Fall under this weighting
Zero-change point: 47.96% retained weight.
19 matched authorities. Baseline: 2025-Q3.
Horizontal axis: retained weight on removed mentions. Vertical axis: percentage change. The same weight is used in both periods.
Change bridge: +2,219 in the retained group plus -4,627 in the removed group gives -2,408 when the weight is 100%.
The arithmetic behind the reversal
The Q3 planning definition-sensitivity dataset separates all unit mentions in approved records into a retained group and a group removed by a case-insensitive "reserved matters" phrase match. This application takes those published aggregates and adds a continuous sensitivity test.
In the East of England, the retained group rises from 2,376 to 4,595 mentions, a change of +2,219. The removed group falls from 6,679 to 2,052, a change of -4,627. Restoring enough of the second group overwhelms the first group's rise. The balance is reached at 2,219 ÷ 4,627, or 47.96%.
This is an accounting identity, not a causal explanation. We cannot tell from these aggregates whether individual removed records repeat an earlier consent, represent a distinct stage, or refer to the same dwellings. Nor can we conclude that a particular intermediate weight is the correct one.
One assumption, made explicit
Let F be unit mentions left after the filter, R the removed mentions, and w the common retained weight between 0 and 1. Subscripts 0 and 1 mean baseline and current period.
T(w) = F + w × R
Change(w) = 100 × [(F₁ + wR₁) / (F₀ + wR₀) - 1]
Zero-change weight = -(F₁ - F₀) / (R₁ - R₀)
The baseline totals are positive in all 16 comparisons. Their direction therefore follows the straight-line numerator, (F₁ - F₀) + w(R₁ - R₀). If that expression has no root between 0 and 1, its sign cannot change inside the interval. The percentage curve itself need not be a straight line because its denominator also changes.
Holding w equal in both periods is restrictive. This tool does not test separate weights for each period, correct misclassification, estimate sampling error, or assess the probability of future construction. A stable sign here only means stable under this common-weight interpolation.
All results, including the stable directions
These tables compare Q3 2026 with Q3 2025 and Q2 2026. "None" means no zero-change point strictly inside the 0% to 100% interval. The panel total overlaps all regional rows and must not be added to them.
| Geography | Authorities | 0% retained | 100% retained | Zero-change point |
|---|---|---|---|---|
| Panel | 253 | -27.1% | -25.4% | None |
| East of England | 19 | +93.4% | -26.6% | 47.96% |
| London & South East | 102 | +3.9% | +1.3% | None |
| Midlands | 55 | -29.7% | -21.7% | None |
| North East & Yorkshire | 20 | -24.5% | -12.4% | None |
| North West | 27 | -38.8% | -15.3% | None |
| South West | 23 | -68.5% | -64.4% | None |
| Wales | 7 | -55.4% | -74.6% | None |
| Geography | Authorities | 0% retained | 100% retained | Zero-change point |
|---|---|---|---|---|
| Panel | 253 | -28.4% | -20.8% | None |
| East of England | 19 | +57.1% | +26.5% | None |
| London & South East | 102 | -46.5% | -38.3% | None |
| Midlands | 55 | -16.9% | -10.3% | None |
| North East & Yorkshire | 20 | +6.7% | +25.5% | None |
| North West | 27 | -57.9% | -28.0% | None |
| South West | 23 | -7.2% | -25.3% | None |
| Wales | 7 | -62.4% | -64.9% | None |
What this changes in a regional note
The South West's annual decline survives this test: -68.5% after the filter and -64.4% with all mentions. That makes its direction less sensitive to this particular choice than the East of England's. It still says nothing about the probability of a specific scheme completing.
The Construction Capital South West Q3 report provides the wider regional context. If citing a pipeline number, name the definition, source edition, geography and coverage beside it.
Keep funding and planning evidence distinct
A rise in unit mentions is not a rise in financed projects. Construction Capital's Development Finance guide and service overview explains the funding process and the role of scheme-specific planning, costs, valuation and exit evidence. This sensitivity result does not establish lending appetite, loan size or terms.
The ICPSR Q3 planning approval study asks a separate question about decisions and regional mix. The companion unresolved-outcomes explorer tests decision denominators. Neither is a check on how many unique dwellings these unit mentions represent.
Download and reproduce
No account, email address or finance enquiry is required to use the tool or download its calculations. Run python3 reproduce.py reproduced beside the source CSV to rebuild the 16-row table. Python 3's standard library is sufficient.
For a spoken explanation of the binary filter reversal, see the Construction Capital planning-data video. This tool adds the change bridge and breakpoint test.
Coverage, rights and attribution
The underlying matched panel covers 253 authorities in seven custom regional groups, including only seven authorities in Wales. It is a partial panel, not a national planning census. Q3 2026 is provisional. These are mentions in approved records, not deduplicated net new dwellings, starts or completions. Missing unit counts contribute zero upstream; the aggregate export does not expose unit-observation coverage. The tool cannot rebuild or independently validate that upstream extraction.
Cite: Lenzie, Matt (2026). Planning definitions: where does the trend change sign? Construction Capital, CC-DF-2026Q3-C04, v1.0, 12 October 2026. Adapted from CC-DF-2026Q3-04, frozen edition 5 October 2026. Identify any changes when reusing the analysis.
This tool and the linked Construction Capital publications share authorship and commercial affiliation. This is a methodological adaptation, not an additional independent study, peer review or institutional endorsement. The selected source aggregates come from Construction Capital's Landstack-derived export. Underlying source rights remain reserved; no blanket licence to the Landstack database is granted. Original commentary: CC BY 4.0. Application and reproduction code: MIT. See component licences and the retained source rights notice.
Construction Capital is a trading name of Lenzie Consulting Ltd, registered in England and Wales, company number 08174104. We are a commercial finance broker and introducer, not a lender, and we are not authorised by the FCA. Where a deal is a regulated activity we arrange it through lenders who hold the relevant FCA permissions. Every figure is indicative, varies by lender and deal, and is never an offer of finance.