Council tax integration: Nuffield Workstream 2 deliverable tracker #1854

Activity

1. vahid-ahmadi commented on Sep 15, 2026

   

   vahid-ahmadi

   on Sep 15, 2026

   CollaboratorAuthor

   More actions

   Statutory findings from the Axiom side — these change the Wales items above

   Investigated in parallel via TheAxiomFoundation/rulespec-uk (see TheAxiomFoundation/rulespec-uk#344). Three corrections to the scope as originally written in this issue:

   1. Band I for Wales comes from LGFA 1992 s.5(1A), not separate Welsh legislation. s.5(1) prints the England proportions as 6 : 7 : 8 : 9 : 11 : 13 : 15 : 18 for bands A–H. s.5(1A) substitutes, for Wales and for financial years beginning on or after 1 April 2005, 6 : 7 : 8 : 9 : 11 : 13 : 15 : 18 : 21 — so Band I = 21/9. Both sets are printed in the Act as ratios relative to each other; the "relative to Band D" form (A = 6/9 … H = 18/9) is a derivation, not Act text.

   2. The operative provision for gross liability is s.10 (Basic amounts payable), which the original scope above omitted. s.5 gives the proportions; s.10 gives the amount payable.

   3. s.12 was omitted with effect from 1 April 2026 by the Local Government Finance (Wales) Act 2024 (asc 6), ss.18(6) and 23(7). Anything modelling Welsh discounts needs to account for this — and 2026-27 is the first year affected, which is the year the data refresh targets.

   Further detail relevant to the discount/premium work item:

   • s.11(2)'s 50% no-resident discount is derived, not printed — the Act says "twice the appropriate percentage", and s.11(3) sets the appropriate percentage at 25 per cent unless varied by order.
   • ss.11–11D and Schedule 1 are now England-scoped.
   • s.11A's discount floor is "at least 10" per cent.
   • s.11B's "relevant maximum" is year- and duration-dependent (ss.11B(1A)–(1C)).
   • s.11C's cap is 100 per cent; ss.12A/12B Welsh caps are 300 per cent.

   Schedule 1 paragraph numbers for the student and severe-mental-impairment disregards are not stated here deliberately — the headings were confirmed to exist but the numbering was not verified. To be checked against the Act before encoding.

   Axiom will not be the venue for liability

   For the record, since it affects where this work lands: Axiom's omission of council tax liability is deliberate, not a gap. 82 of its 100 encoded CTR modules carry an explicit module.deferred_outputs block, several naming liability deferral directly (Sandwell: "council-tax charging, statutory discounts, joint-liability apportionment, and calendar-day calculation are future outputs"). 68 of 100 use a locally-prefixed liability variable and 30 declare the slot in a module-level inputs: block. Axiom encodes the reduction and expects liability injected by the caller.

   So the liability modelling committed to Nuffield belongs here, in policyengine-uk, not upstream in Axiom. Issue #344 asks them whether they would want an LGFA 1992 Part I charging encoding contributed; that is a separate question from this deliverable.
2. vahid-ahmadi commented on Sep 18, 2026

   

   vahid-ahmadi

   on Sep 18, 2026

   CollaboratorAuthor

   More actions

   Microcosm compatibility — constraints on everything in this tracker

   PolicyEngine/microcosm consumes both policyengine-uk and policyengine-uk-data, and maintains a formal parity register against uk-data's target inventory. It already has a council tax surface, and in places a richer one than uk-data. Anything built for this deliverable has to hold to that contract, so recording the constraints here.

   1. Band amounts are rates. They must never be wired as additive loss-matrix targets.

   packages/microcosm-build/src/microcosm/build/uk/uk_data_target_parity.json carries local_council_tax_band_d_rate at status reviewed_exclusion, classification non_linear: "Band D currency amounts are per-rate, not additive household controls", with verdict_basis "Keep excluded until a rate-aware non-linear objective exists". Microcosm reached this independently of PolicyEngine/policyengine-uk-data#483.

   This binds the eight new band_A_amount..band_H_amount columns from PolicyEngine/policyengine-uk-data#482 as well — they are amounts, therefore rates, therefore not additive targets. #482 emits no targets from them, which is correct and should stay that way.

   2. Band counts are the surface that legitimately calibrates.

   local_council_tax_band_counts is ported_local_declared and actively bound — Microcosm's council_tax/band_a..i contract rows "bind 2,511 authority cells on the taxbase basis with the 296 English band-H cells and Shetland's band-H cell signed deferred (microcosm#929)". So voa/council_tax/{code}/{band} stays. Counts are additive; amounts are not.

   3. Net council tax at LA grain is blocked upstream, not just here.

   local_council_tax_net sits at blocked_source: "The pinned feed has no comparable 2025 local-authority council-tax net series across the 361-area roster", with an explicit fence against combining unlike country totals, 2026 Welsh budgets, or gross dwelling stock. So retargeting onto total_council_tax_net is not an available escape hatch for #483.

   4. New policyengine-uk variables must be engine-computed, not supplied inputs.

   packages/microcosm-frame/src/microcosm/frame/adapters/policyengine_uk.py classifies variables via _is_engine_computed, which tests for adds/subtracts/formula/formulas. A new variable with a formula is adapter-safe. One that has to be supplied as an input column is not, and the adapter has precedent for requiring explicit input handling (see validate_uc_claimant_input).

   This is a live consideration for #1855: council_tax_band_d_amount reads an LA-keyed parameter rather than being a dataset input, which keeps it computed and therefore adapter-safe. It is also another argument against resolving #1855 by making liability a supplied input.

   5. There is an existing Axiom adapter too.

   adapters/axiom.py sits alongside the policyengine ones, so Microcosm is the point where the rulespec-uk charging encoding (TheAxiomFoundation/rulespec-uk#346) and the policyengine implementation would meet. Worth keeping the two consistent in naming and entity level rather than letting them diverge.

   Action: @juaristi22 added as reviewer on #482, #1855 and TheAxiomFoundation/rulespec-uk#346, and on the two PRs opened against #483 and the discounts work.

   🤖 Generated with Claude Code
3. vahid-ahmadi commented on Sep 21, 2026

   

   vahid-ahmadi

   on Sep 21, 2026

   Last edited by vahid-ahmadi

   CollaboratorAuthor

   More actions

   Council tax workstream — status for review

   @MaxGhenis @juaristi22 — everything open on the Nuffield WS2 council tax line, in one place. Four PRs are reviewable; two items need a decision, one needs an action only Max can take.

   Reviewable now

   PR	What
   PolicyEngine/policyengine-uk-data#482	Eight band amounts from MHCLG Table 9 (2026-27), keyed on ONS code
   PolicyEngine/policyengine-uk-data#484	Drops the Band D calibration target
   #1864	LGFA 1992 s.11 statutory discounts
   TheAxiomFoundation/rulespec-uk#346	Five CT-CHG worklist entries (ss.5, 10, 11, 11A, 11B)

   #482 is purely additive — 361 rows unchanged, 16→24 columns, and all 294 directly-matching England rows reconcile to the MHCLG source at £0.00. All 296 authorities obey the statutory proportions (worst deviation £0.0044, per-band rounding). Note band_d_amount was already on 2026-27 figures, so this adds the other seven bands rather than refreshing anything.

   #484 resolves #483. María — your parity register already answered this: local_council_tax_band_d_rate is a reviewed_exclusion, classification non_linear, "Band D currency amounts are per-rate, not additive household controls". So the fix is to drop the target, not to add a mean-returning compute route — a mean would stop the number being wrong but would still bind a rate into an additive objective, which is the thing the fence exists to prevent. Band counts (voa/council_tax/{code}/{band}, 2,563) are untouched and still bound. Retargeting onto total_council_tax_net was the other option and is closed off too, since local_council_tax_net is at blocked_source on your side. Tell me if dropping rather than fencing-in-place is the wrong shape for your register.

   #1864 encodes s.11(1)/(2)/(3) with the 25% as the parameter and the 50% derived as twice it, since "twice the appropriate percentage" is what the Act prints. Microcosm-safe: both new variables define a formula, so _is_engine_computed classifies them as engine-computed and neither can be requested as a supplied input column.

   Decision needed — #1855

   Computed liability gives England £61.21bn against the ~£40bn we cited to Nuffield. Most of that is an artefact, and I think it's a design question rather than a bug: the variable looks up Band D via household("local_authority"), but no dataset builder populates local_authority, so every household decodes to the enum default (Maidstone). LA analysis in uk-data runs through per-LA reweighting (local_authority_weights.h5), not per-household tagging — so a household-level enum lookup may be the wrong mechanism entirely.

   This gates the validation report and policy examples, so I'd rather not build further on it until it's ruled on. council_tax itself is untouched — still the input, so CTR results don't move.

   Action only Max can take — TheAxiomFoundation/rulespec-uk#346

   Blocked on corpus ingest for ss.5, 10, 11, 11A, 11B. Write access to axiom-corpus doesn't carry the ed25519 ingest key: its Actions secrets hold AXIOM_CORPUS_RELEASE_SIGNING_KEY and no ingest private key, the only ingest key in variables is the public verification one, and no workflow signs a manifest. Detail and the three options are on #344.

   Two open questions on #344

   1. Ingest key — provision it, run the five extract-uk-legislation calls yourself, or is there a contributor path I've missed?
   2. Oracle — policyengine-uk-rust does have the band multipliers and the 25% discount, but on unmerged branches (vahid/council-tax and ten others), not main, where council_tax is still an FRS input. Build the axiom-oracles mapping against that branch, wait for it to land, or has this moved to 10ds-microsim?

   Also filed

   • data/councils/registry.csv status and scheme_url have drifted: 98 of 99 encoded councils still say pending TheAxiomFoundation/rulespec-uk#347 — data/councils/registry.csv drift: 98 of 99 encoded councils still say pending, and only 5 of the 99 have a scheme_url. The roster is correct and the layout test only uses it as an allowlist, so nothing catches this. Asked whether those columns are still meant to be live before rewriting 98 rows.
   • Encode LGFA 1992 s.5 and s.10 council tax charging (proposal for #344, blocked on corpus ingest) TheAxiomFoundation/rulespec-uk#345 — closed. The hand-written LGFA encoding, superseded by Uprate UK poverty thresholds #346 once the worklist route was clear. (Compute gross council tax from LA Band D amounts and statutory band ratios #1855 is open and is the liability PR above.)

   Two things I'd like a steer on

   Double-discount risk. FRS-reported council tax is generally the billed amount, already net of any single-person discount, while uk-data documents the input as "gross liability before CTR" — silent on statutory discounts. If that reading is right, applying #1864's discount to the FRS figure would discount twice. This is my inference, not something the repo asserts, and one of you will know outright. It matters for how #1855 and #1864 compose.

   Naming alignment. Microcosm has an adapters/axiom.py alongside the PolicyEngine ones, so Microcosm is where the rulespec charging encoding and the PolicyEngine implementation actually meet. Worth agreeing variable naming and entity levels now, while both are unbuilt, rather than reconciling later.

   Still open against the Nuffield commitments

   Schedule 1 disregards, s.11A determinations, s.11B/11C premiums, exemptions, Wales and Scotland liability, the validation report, and policy examples. All itemised in #1854.

   Three statutory findings that moved scope along the way: Wales Band I comes from LGFA 1992 s.5(1A) itself, not separate Welsh legislation; s.10 is the operative provision for liability and I'd originally missed it; and s.12 was omitted with effect 1 April 2026 by the Local Government Finance (Wales) Act 2024, which matters because 2026-27 is the year our Band D figures come from. Scotland is also not on flat ninths — SSI 2016/123 raised E–H to 473/360, 585/360, 705/360 and 882/360.

   🤖 Generated with Claude Code
4. vahid-ahmadi commented on Sep 26, 2026

   

   vahid-ahmadi

   on Sep 26, 2026

   CollaboratorAuthor

   More actions

   Correction: I cited the wrong instrument for the Scottish band proportions, in this thread and in several others.

   I wrote that SSI 2016/123 raised the Scottish E–H proportions. That is wrong. SSI 2016/123 is the Local Authority (Capital Finance and Accounting) (Scotland) Regulations 2016 and has nothing to do with council tax.

   The correct instrument is SSI 2016/368 — the Council Tax (Substitution of Proportion) (Scotland) Order 2016. Its article 2 substituted the proportions

   240 : 280 : 320 : 360 : 473 : 585 : 705 : 882
   

   into LGFA 1992 s.74(1) for financial years beginning on or after 1 April 2017.

   Two things that follow, beyond the reference itself:

   • The operative provision for Scottish proportions is s.74(1), in the Scotland Part of the Act — not s.5, which is England and Wales. I had been eliding that.
   • Expressed on the 360 = Band D basis, A–D at 240/280/320/360 are identical to the original s.74(1) proportions, so only E–H changed in 2017. Anything encoding Scottish ratios needs A–D defined from 1993, not from 2017.

   Caught while addressing review feedback on #1855, where the same wrong citation had been carried into the parameter file. Corrected there, and in the bulk/worklist.yaml note on #346 (commit ed3830c).

   Apologies for the noise — a wrong statutory citation is worth more than a footnote in this repo.

   🤖 Generated with Claude Code
5. vahid-ahmadi commented on Oct 24, 2026

   

   vahid-ahmadi

   on Oct 24, 2026

   CollaboratorAuthor

   More actions

   Status, 24 Oct 2026 (main at 4ed874f):

   • Gross liability: Compute gross council tax from LA Band D amounts and statutory band ratios #1855 merged 22 Oct. It adds council_tax_band_d_amount (Band D amounts for England, Wales and Scotland, 2025-26 and 2026-27), band ratios for all three nations including Welsh Band I, and simulated_council_tax.
   • Statutory discounts: Add statutory council tax discounts across Great Britain (LGFA 1992 ss.11 and 79; Welsh Regulations 2026) #1864 merged 22 Oct. It adds council_tax_discount_rate and simulated_council_tax_less_discounts (LGFA 1992 ss.11 and 79, and the Welsh 2026 Regulations).
   • Disregarded persons, exemptions and empty-home and second-home premiums: not modelled. The variable documentation says the result is an upper bound.
   • Relationship to the council_tax input: documented but not decided. council_tax is still the input.
   • Eight-band CSV and the 12-authority alias map: Extend LA council tax to all eight bands (MHCLG Table 9, 2026-27) policyengine-uk-data#482, still open. In policyengine-uk the amounts are keyed on the LocalAuthority enum.
   • Band D target: Drop ons/council_tax_band_d/{code} target emission (#483) policyengine-uk-data#484 (for #483), still open.
   • Validation report and policy examples: not started. council_tax_band_d_amount needs a dataset that supplies local_authority. The Enhanced FRS does not, so every household defaults to Maidstone. Validation therefore needs Microcosm UK.
   • Add LGFA 1992 council tax charging sections (ss.5, 10, 11, 11A, 11B) to the bulk worklist TheAxiomFoundation/rulespec-uk#346: still open.

   🤖 Generated with Claude Code