# CANDIDATE A: BOUNDED MHCLG PS2 CROSS-CHECK. DEFINITIONS AND VALIDATION RULE

Fixed Sunday 20 September 2026, about 18:45 UK, BEFORE any London official
figure was computed. Not moved afterwards. Kit context: v475 (Candidate A
HOLD; national thesis DEAD; London/Datahub confound is the only open question).

## 1. WHAT THE OFFICIAL PS2 MEASURE IS

| item | value |
|---|---|
| dataset | MHCLG "District planning application statistics (PS2) - full dataset", open data CSV `PS2_data_-_open_data_table__202603_.csv`, published 19 Jun 2026 with the Jan-Mar 2026 statistical release; sha256 2bfd2293b5354bd39f07039206f604ac6b613ac94b1211c82371e27933a650bf; retrieved 20 Sep 2026 17:34 UTC from assets.publishing.service.gov.uk |
| return form | PS2, the quarterly district-matters return. It "collects more details about the decisions made during the quarter, broken down by development type, whether permission/consent was granted or refused, and time taken" (MHCLG District Matters data dictionary, sha256 61662c18...04d4) |
| reporting period | calendar quarters labelled `YYYY Qn` (Q1 = Jan-Mar ... Q4 = Oct-Dec). Latest published: 2026 Q1. Calendar 2021 and calendar 2025 are each fully covered by four published quarters |
| authority identifier | `LPACD` (ONS/GSS code) and `LPANM`. London boroughs E09000002-E09000033; City of London E09000001 |
| period basis | DECISION DATE: decisions made during the quarter (same basis as PlanningLens decision year) |
| numerator | `Total granted; <type> (all)` |
| denominator | `Total decisions; <type> (all)` = granted + refused (verified exactly for every 2021 and 2025 row). Withdrawn/turned away are not decisions and are not in PS2 |
| major/minor/other | all present. "Other" in 2021 and 2025 = change of use + householder developments + advertisements + listed building consents (alter/extend) + listed building consents (demolish) + relevant demolition in a conservation area (verified: other total equals the sum of these six for every 2021/2025 row) |
| householder | included, as `(other) householder developments` |
| certificates of lawful development, notifications, "all others" | NOT collected on PS2 in 2021/2025 (columns are `..`/blank); counts of certificates appear on PS1 only. So PS2 already excludes certificates |
| prior approvals (permitted development) | NOT in PS2 at all; reported on PS1 (PDR block) as not-required / granted / refused. PlanningLens's frozen substantive definition KEEPS prior approvals (TYPE_KEEP_STRONG token `prior_approval`; 5-25% of London decisions in the mix tables). This is the largest known cohort difference and is NOT reconciled here |
| "all" vs "excluding PAs" | the `(all)` columns are used (all decisions including those with performance agreements, extensions of time or EIAs) |

## 2. THE TWO OFFICIAL COMPARATORS (fixed in advance)

- **PS2-CORE (PRIMARY, used for every assessment)**: decisions = major total + minor total + (other) change of use + (other) householder developments, summed over the four calendar quarters; rate = granted / (granted + refused) x 100. Rationale: these are applications for planning permission, which is what the frozen substantive definition is built to capture (full, outline, reserved matters, householder, change of use, dwelling, s73, telecom, combined, text-decided). The consent regimes (advertisements, listed building, conservation-area demolition) are excluded from CORE because the kit does not let me verify how the frozen text rules treat them.
- **PS2-ALL (SECONDARY, reported only, never used to assess)**: grand total (all) over the four quarters.

## 3. PLANNINGLENS INPUT

The frozen substantive approval rate and N for 2021 and 2025 exactly as printed in the v475 receipt `pl_candA_clean_scan_20260920_1330_out.txt` (section 3 load lines, 1 decimal place; the five extreme risers also at 2 dp from Q6). Rates are NOT recomputed and the definition is NOT altered. Precision of the receipt figures is +/-0.05pp, immaterial at the tolerances below.

## 4. COUNCILS

All 32 London boroughs in the clean panel (City of London excluded under R5, LLDC/OPDC under R3). The receipt's 8a table says 28 are the two Datahub schema families and Hackney is the 24-col mixed store; three boroughs sit on other schemas and cannot be named from the kit, so all 32 are carried and this is disclosed. Control, permitted by TODO_NEXT item 2 "if it costs nothing": the five clean-panel top fallers (Oadby & Wigston, Warrington, Dartford, Doncaster, Slough), same method, reported separately.

## 5. PER-COUNCIL ASSESSMENT RULE (PS2-CORE only)

1. **NOT COMPARABLE** if any of the four quarters is missing or non-numeric in either year, or if PS2-CORE 2025 decisions < 200.
2. **LEVEL test (2025)**: d = PL2025 - CORE2025.
   |d| <= 5.0pp: OK. 5.0 < |d| <= 10.0pp: PLAUSIBLE. |d| > 10.0pp: CONFLICT.
3. **DIRECTION test**, applied ONLY where |PL move 2021->2025| >= 5.0pp (a PL "major mover"): m = CORE2025 - CORE2021.
   same sign as PL move and |m| >= 2.0pp: OK. |m| < 2.0pp (official flat): PLAUSIBLE. opposite sign and |m| >= 2.0pp: CONFLICT.
   Councils with |PL move| < 5pp get the level test only.
4. **Assessment**: AGREES BROADLY if every applied test is OK; MATERIAL CONFLICT if any applied test is CONFLICT; otherwise PLAUSIBLE DIFFERENCE.

Why 5 / 10 / 2. Five points of level tolerance covers the known cohort gaps (prior approvals in PL only; text-classified rows; reference dedup; consents out of CORE). Ten points is the size of the London movements themselves: a level gap that large means PL's 2025 level is as far from the official figure as the change it claims. Two points of same-direction official movement is roughly two standard errors of a difference of two proportions at N about 2,000 (SE about 0.8-1.1pp), so it is distinguishable from noise while allowing the official cohort to move less than PL's.

## 6. LONDON SIGNAL RULE (decision F), fixed in advance

Let R = the 14 London boroughs PL shows rising >= 5pp, and C = {Hillingdon, Brent, Bexley}.

- **LIKELY FEED ARTEFACT / KILL** if ANY of: (a) >= 5 of the 14 in R are MATERIAL CONFLICT; (b) >= 2 of C are MATERIAL CONFLICT; (c) the official pooled London PS2-CORE rate moved <= 0pp from 2021 to 2025 while PL pooled London moved >= +5pp.
- **VALIDATED ENOUGH TO INVESTIGATE** if ALL of: (a) >= 2 of C are AGREES BROADLY and none is MATERIAL CONFLICT; (b) at least 9 of the 14 in R have direction OK; (c) at most 2 of R excluding Richmond and Hackney are MATERIAL CONFLICT; (d) the official pooled London PS2-CORE move is >= +2.0pp.
- **MIXED / HOLD** otherwise.

## 7. WHAT THIS CHECK CANNOT DO

It cannot show that London planning "became easier". At most it can show that official data independently records a similar change in approval rates. It does not reconcile prior approvals, does not rewrite the frozen definition, does not recalibrate Candidate A, and does not touch the PlanningLens repository, databases or website.