VAT survey coverage was derived on an older Microcosm cut and puts VAT about 9% above receipts on the current certified cuts #2633

gov.simulation.vat.survey_consumption_coverage (0.67, product 0.469 with the household share of 0.7) was derived in #2187 on the Microcosm UK 2024-25 national release microcosm-uk-2024-25-national-20261002T230158Z-5c6b3f68 (revision f9d1922c). Its metadata and docs/engineering/vat-coverage.md say to re-derive it when the dataset it should match changes. Two certified Microcosm cuts have replaced that one since:

• microcosm-uk-2024-25-national-20261019T083323Z-6cba5d71, the cut policyengine.py 6.10.2 pins;
• microcosm-uk-2024-25-national-20261023T075039Z-a133f37a, the current default and the cut that policyengine.py#715 certifies.

On both, the standard-rated base is about 10% larger than on f9d1922c (2024-25: £435.6bn on a133f37a against £396.15bn). With the shipped product, total vat sits above OBR receipts, and one point on the standard rate costs more than HMRC's ready reckoner.

Measured on a133f37a with the doc's own calculation, using policyengine-uk main at c025241 (2.152.7) and core 3.33.2. This is a baseline-only Microsimulation with weighted sums; the dataset's SHA-256 is 64f83900ca3a1242….

• One point on the standard rate: a reform run on 6cba5d71 (policyengine-uk 2.135.7 and 2.152.6) costs £9.94bn in 2026-27, against HMRC's £8.8bn (+12.9%). The ready reckoner implies a product of 0.529 on a133f37a. A reform run on a133f37a is pending.
• 6cba5d71: the same calculation agrees to 3 dp (implied products 0.518, 0.512 and 0.504).

Re-deriving by the doc's rule (the latest outturn year, 2024-25) gives a coverage of 0.739, rounded to 0.74 (product 0.518). That sits inside the bound from the receipts route (0.720 to 0.739) and the ready reckoner (0.756).

The fix lowers VAT reform costings and baseline vat (and so gov_tax); the sizes come from runs at the fix branch, in #2634. household_tax counts vat_change, not vat, so baseline household incomes don't move.