Skip to content

populace-uk candidate (for UK data controller review)#28

Open
MaxGhenis wants to merge 1 commit into
mainfrom
populace-uk-candidate
Open

populace-uk candidate (for UK data controller review)#28
MaxGhenis wants to merge 1 commit into
mainfrom
populace-uk-candidate

Conversation

@MaxGhenis

Copy link
Copy Markdown
Contributor

Stages populace-uk as a candidate UK data release — deliberately left for @nikhilwoodruff's sign-off (UK data controller; swapping the UK certified default is not a solo call).

The release (populace-uk-2023-72aeefc-20260611, private repo under UK Data Service licence, same split as policyengine-uk-data-private):

  • FRS 2023-24 base + the primary-donor imputations (WAS, LCFS, SPI incl. the synthetic high-income stratum, ETB/VAT, capital gains, salary sacrifice, student loans), 20× OA clone pool (535,080 households), populace-calibrated (bounded ratio 50; 93.24% of the 148-target registry within 10%; max weight 13,683).
  • Gates: parity 0 gaps vs the frozen enhanced FRS at simulation level; no all-zero stored columns.
  • Sound comparison (matched 53,508 households, symmetric refit, train AND holdout required): 6/6 rotations won — holdout 0.31–0.60 vs the enhanced FRS's 0.40–0.65; per-target 123/25/0 on the primary rotation.

Caveats stated plainly: the income-tax (£353B) and UC (£21.8B) aggregates differ from my HMRC/DWP outturn anchors — the registry fits within 10%, so the target definitions likely differ from those anchors, but that reconciliation should be part of review. The pinned manifest uses the populace build schema; conversion to the usdata release-manifest schema accompanies sign-off.

🤖 Generated with Claude Code

populace-uk-2023-72aeefc-20260611 (PRIVATE repo, UKDS licence): beats
the enhanced FRS 6/6 rotations on the sound comparison (holdout
0.31-0.60 vs 0.40-0.65 at matched 53,508), parity 0, exported-nonzero
pass. NOT self-merged: swapping the UK certified default is the data
controller's decision. Note: the manifest is the populace build schema,
not yet the usdata release-manifest schema the generator consumes —
conversion accompanies the sign-off.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant