Official bin-performance data and constructed value-for-money rating

Status and boundary

This is an **independent, constructed peer-relative indicator** for English waste collection authorities. It is not an official government rating, an inspection judgement, or a complete measure of service quality. In particular, there is no consistent national council-level series for missed collections, complaint handling, assisted collections or resident satisfaction, so those service-quality dimensions are not scored.

Stage 1 published the address-free council data, deterministic importer and builder, runtime lookup and audit material. Stage 2, a separate editorial and product decision, now shows the rating on a resident's result page beside a link to the plain-English explanation at /bin-collection-value-for-money. This document is served publicly from the same site, at /bin-collection-value-for-money/technical-method. The files and scripts it names live in the service's source repository.

The runtime contract is:

officialBinPerformanceFor(gss) // validated record, or null

It returns null for a non-English or unknown GSS code, a stale artifact, an artifact built for another period or methodology, or any dataset containing a partially invalid row. Validation happens for the whole artifact before any row is exposed.

Official sources

Every source in data/official-bin-performance-input.json records its publisher, dataset and table, landing page, direct download URL, publication period, retrieval date, source version, Open Government Licence v3.0 metadata and SHA-256 checksum. Availability and licensing were checked on 30 August

1. Raw downloads are not committed.

official-bin-performance-sources.mjs is the reviewed source manifest. It also records, for every observation class, the exact table or sheet, source columns, row identity, source and output units, and transformation. The importer requires all six local filenames explicitly, checks every pinned SHA-256 before parsing, and then rejects changed CSV columns, ODS sheets or headers, units, duplicate row/GSS identities, and ambiguous crosswalks. The ODS reader is a format-constrained OpenDocument ZIP/XML reader built only from Node standard library modules; no office runtime or package dependency is required.

Spend

The Ministry of Housing, Communities and Local Government (MHCLG) Revenue Outturn multi-year CSV supplies net current expenditure, in £000:

• RO5_envcll_net_cur_exp: RO5 line 281, household waste collection;
• RO5_envrcy_net_cur_exp: RO5 line 284, recycling.

The RO5 specific guidance says line 281 includes household waste collection but excludes trade waste and separately collected recyclable, reusable or compostable waste. Those separate collections belong on line 284. Line 284 can also contain processing, sorting, reprocessing and composting costs, credits and recyclate income. The sum is therefore described as a **bin and recycling service spend proxy**, not as pure vehicle-and-crew collection cost.

Net expenditure is used because it reflects income and credits actually reported against the service. Negative component values are retained as source observations; an annual combined value must remain positive. Comparing each council with same-year peers before taking the multi-year median avoids mixing cash values from different years and removes the need for a separate inflation assumption.

Waste outcomes and denominator

The Department for Environment, Food & Rural Affairs (Defra) Local authority collected waste management annual results are accredited official statistics based on returns made by all English local authorities to WasteDataFlow. The normalized evidence uses:

• Table 1: household waste not sent for recycling, tonnes;
• Table 3: residual household waste per household (former NI191);
• Table 3: percentage of household waste sent for reuse, recycling or composting (former NI192);
• Table 3: Defra waste authority type, Collection or Unitary.

The household count is an explicitly derived denominator:

estimated households = residual household waste tonnes × 1,000
                       ÷ residual kg per household

It is stored under derived_measures, not presented as a source observation. The builder refuses a missing, zero or non-reconciling denominator.

Geographic and demographic context

ONS Explore Local Statistics supplies numeric council observations for:

• population density, residents per square kilometre;
• total population;
• gross median weekly pay, a local labour-cost proxy;
• children in relative poverty before housing costs, a deprivation proxy.

The use of contextual grouping is consistent with the principle in Defra's Local Authority Packaging Cost and Performance model methodology, which groups authorities using factors including rurality, flats, deprivation, households, collected tonnes and collection design. Defra reports that its own residual and recycling groupings explained 89% of sampled cost-per-tonne variance. This rating does **not** reproduce or claim validation from that model: its cost sample and several inputs are not public at the required level.

The ONS statistical-neighbours methodology supports winsorising, standardising and using Euclidean distance for small comparator groups. This implementation uses its own disclosed waste-focused feature set rather than silently adopting all 29 ONS indicators.

All these publications state Open Government Licence v3.0 terms. The licence reference stored with each source is the Open Government Licence v3.0.

Exact authority identity and responsibility

GSS codes are exact identifiers; names are descriptive only. The input contains reviewed one-to-one crosswalks for source codes retained after an authority change:

Any duplicate source mapping or duplicate successor is invalid. The importer also names every 2022-23 predecessor row for the Cumbria and North Yorkshire reorganisations, plus the Somerset Waste Partnership row. Those observations are excluded rather than aggregated into Cumberland, Westmorland and Furness, North Yorkshire or Somerset. Abolished authorities are not scored as if they still existed, and predecessor figures are not imputed into successors.

Four current councils are excluded because Defra publishes their outcome observations only as joint services, so a council-specific score would be ambiguous:

• Cambridge City and South Cambridgeshire;
• Babergh and Mid Suffolk.

Collection authorities are compared only with Defra Collection authorities; Defra Unitary authorities are compared only with Unitary authorities. This is a hard gate, not a distance penalty. It prevents a collection-only district from silently borrowing the cost structure of an authority holding materially different waste responsibilities.

The release roster contains exactly 296 sorted current English GSS codes across observed and explicitly excluded councils. Its SHA-256 is e521db89c273f9bf229c6396ba4495e6b609cabde3f99ed4d286052ac0822830. Both input and output carry the complete roster. Runtime rejects a missing, unexpected, duplicate, jointly rated-and-excluded, or hash-mismatched code. The expected count and SHA-256 are also pinned in the immutable OFFICIAL_BIN_PERFORMANCE_RELEASE_CONTRACT; an artifact cannot redefine its own expected roster by substituting a syntactically valid code and recomputing its self-declared hash. This data-release identity is deliberately separate from the scoring methodology identity, so a future official-data roster refresh need not falsely imply that the formula changed.

Method version 1.1.0

The immutable runtime method ID is:

official-bin-vfm:1.1.0:daf2fdddf62761160fe5f3c4c05d441a6cb605916da68759e695a4c53711361e

The suffix is the SHA-256 of a canonical JSON representation of the complete immutable contract in official-bin-performance-contract.mjs. It binds source fields and units, the authority partition, transformations, winsorisation, standardisation population, distance formula, weight interpretation, tie breaking, annual gates and aggregation, percentile tie convention, banding, confidence threshold, rounding, and review/expiry policy. Changing any choice changes the hash and method identity. Runtime compares the full canonical contract as well as its hash and version, so relabelling an old file cannot silently reinterpret it.

1. Select peers before looking at performance

For each authority type separately, transform the four context variables:

Each feature is winsorised at the English 1st and 99th percentiles and standardised across all eligible authorities before the authority-type hard partition. The standard deviation is the population form (divisor n), and quantiles use linear interpolation at (n-1) × p. Weighted Euclidean distance selects the nearest 20 authorities, always excluding the council being scored. Ties break by GSS code. The peer selection uses no spend or outcome value.

At least 12 selected peers must have observations in each compared financial year. The peer median is used rather than the mean. A council needs two of the three years and must include 2024-25. No missing value is imputed.

2. Derive three peer-relative components

For every usable year:

cost ratio = council (line 281 + line 284) £ per estimated household
             ÷ peer median £ per estimated household

residual ratio = council residual kg per household
                 ÷ peer median residual kg per household

recycling gap = council recycling rate − peer median recycling rate

The median of each council's two or three annual comparisons becomes its cost index, residual-waste index and recycling-rate gap. A cost or residual index of 100 means the council equals its peer median; below 100 is better for those two measures. A positive recycling gap is better.

3. Construct the final band

The three peer-adjusted measures are converted to percentile ranks across all eligible English authorities. Percentile ranks, rather than raw magnitudes, stop an extreme cash value from dragging other councils' scores. The composite is:

60% peer-adjusted cost percentile
25% peer-adjusted residual-waste percentile
15% peer-adjusted recycling-rate percentile

Composite percentiles are published as deciles: 1 is the lowest decile and 10 the highest. A decile distribution is therefore a property of the definition, not evidence that the formula was tuned to look attractive.

Cost receives the largest weight because the label is value for money. The two outcomes stop an apparently cheap service receiving full credit merely because it leaves more residual waste or recycles less. The weights are judgement calls, disclosed and sensitivity-tested; they are not official weights.

Confidence and exclusions

standard confidence requires all three years, all 20 peers in every annual comparison, and a twentieth-peer distance no worse than the 95th percentile of the estate. Other publishable rows are limited and carry exact reasons. Councils below the minimum gates are listed in excluded_councils and return null.

The required cohort periods are exact. The 2018 City of London and 2019 Isles of Scilly pay observations remain visible as source observations but are stale for this method, so those councils are excluded before standardisation and peer selection. This audit correction reduced rated coverage from 291 to 289; it was not tuned away.

The artifact expires after 30 April 2027, shortly after the next expected annual Defra update. Expiry removes the rating rather than presenting stale comparisons.

Rebuild and read-only evaluation

The committed normalized input is council-level and address-free. It preserves source observations separately from the household denominator derived during normalization. Raw CSV, ODS and XLSX downloads stay outside Git; their hashes in the input pin the exact evidence used.

Deterministic import from the six exact local downloads:

node scripts/import-official-bin-performance.mjs \
  --revenue-outturn /path/Revenue_Outturn_time_series_data_v3.2.csv \
  --defra-lacw /path/LA_and_Regional_Spreadsheet_2024-25.ods \
  --population-density /path/ons-population-density.csv \
  --population /path/ons-population-count.csv \
  --pay /path/ons-gross-median-weekly-pay.csv \
  --child-poverty /path/ons-child-poverty.csv

Running it with the six hash-verified release files reproduces data/official-bin-performance-input.json byte-for-byte. It performs no download and has no default source directory.

Deterministic build:

node scripts/build-official-bin-performance.mjs

Read-only, address-free estate evaluation:

node scripts/evaluate-official-bin-performance.mjs

The evaluation reports coverage and exclusions, authority-type cohort sizes, annual peer counts, confidence, band distribution and three pre-declared sensitivity scenarios: equal cohort weights, density-heavy cohort weights and a cost-only final score. It does not rewrite the artifact.

scripts/audit-official-bin-performance.mjs takes the same six explicit source arguments. It uses the production importer and builder only to verify byte-identical reproduction. A separate small reference implementation then recalculates feature transformations, winsorised population standardisation, same-type/self-excluding peer selection and distances, annual peer medians, component percentiles, the weighted composite percentile and final band for the fixed stratified sample. That implementation imports no production builder or builder helper. The audit is recorded in docs/official-bin-performance-spot-audit-2026-08-30.md.

Known limitations

• RO accounting allocation and income treatment can differ between councils.
• Line 284 is broader than collection and can include treatment and credits.
• Density does not fully capture route length, road topology, islands, communal bins, flats, tourism or seasonal population.
• Pay and child-poverty measures are proxies, not observed contract wage or collection difficulty.
• The household denominator is algebraically recovered from two rounded Defra measures and is labelled estimated.
• Recycling rate rewards garden waste and scheme design as well as operational efficiency; it is not a complete environmental outcome.
• Missed bins, complaints, accessibility, resident satisfaction and contract quality are absent because no consistent official national series was found.
• A peer-relative band says where a council sits under this disclosed method; it does not identify why, prove mismanagement, or establish how much any council could save.