Simple enough to explain. Strict enough to test.

Official restrictions and forecasts are kept separate.

Otter keeps official readings, company estimates, forecasts and legal restrictions separate. Each public number includes its source and the date it was measured or checked.

A hosepipe restriction is a legal decision made by the water company. Its notice is always the authority. A forecast can never create, change or remove an official restriction.

Your postcode finds the water system that serves you.

We turn a full postcode into a map point and check which official water-company supply area contains it. The map contains 129 areas across England and Wales, using Environment Agency and Dŵr Cymru data.

A warning uses the same area the water company uses: the whole company, a named group of areas or one local supply area. We use a smaller area only when a drought plan or official notice makes that boundary clear.

Scotland has one national supplier but many separate water systems. Detailed public customer boundaries are not yet available, so Scottish postcode results remain limited.

Start with today’s level, then look at similar times in past years.

A reservoir forecast starts with the latest official level and applies the average change seen over the same dates in earlier years. This captures the combined effect of weather, customer use, transfers and day-to-day operations without pretending we can separate every cause.

Rainfall, demand or a more complex method is added only if it performs better on past data that was kept out of the model’s training. The ranges show typical past errors, not guarantees.

We show a forecast only when it beats the simple assumption that today’s level will stay unchanged and its likely range works well enough in later tests. Groundwater is compared with the same time of year, not described as “percent full”. Partial forecasts are labelled when only one well or reservoir is available.

Warn about pressure early, without pretending to know whether a ban will happen.

A warning is not a forecast. It answers a narrower question: do current conditions give people a reason to pay closer attention to possible hosepipe restrictions? Historical tests determine the confidence label and limitations. They do not become a percentage chance that a ban will happen.

Every public warning uses the same four levels: 1 means no unusual pressure; 2 means dry weather is raising supply pressure; 3 requires measured reservoir, river or groundwater stress; and 4 comes from an official restriction or supplier operational warning. Official notices always override the experimental levels. If the underlying weekly check is more than 14 days old, the level disappears and the page says the data is unavailable.

We still show how often each rule caught past restriction starts and how often it warned during no-ban periods. Missing events, false alarms and unavailable demand or operating data are shown plainly instead of being hidden behind a single accuracy claim.

In Yorkshire the fixed rule spotted both later restriction starts and stayed quiet in 12 of 14 past no-ban years. Adding a dry-weather condition reduced false alarms but missed the 2025 restriction, so we rejected it. The surviving rule can support an early watch, but never a percentage chance of a ban.

In London the simple rule spotted all four recorded announcements but also raised alarms in 11 of 25 no-ban seasons. Only 2026 was kept for a later test, and the test used past data rather than a warning issued at the time. It is therefore a cautious public experimental warning, not a validated probability model.

We also tried one shared Yorkshire–London rule. The first version spotted six of seven restrictions and missed London 2026. A more sensitive version spotted all seven but stayed quiet in only 18 of 41 no-ban seasons. We rejected it: the same testing approach can transfer, but the same warning number cannot.

We do not use the withdrawn UKCEH monthly reservoir history. Replacement histories come from named percentage-full readings in Environment Agency reports under the Open Government Licence. Unclear months are left blank. Each history is joined to a fresh company reading and retested; only forecast lengths that pass the later accuracy check are shown.

Affinity Central is the groundwater-led fifth test. The frozen Lilley Bottom rule caught the 2012 calibration restriction with 32 days of lead, then missed the untouched 2026 event. That failed holdout is retained. Affinity Central can still show a public level-2 dry-weather alert, but the stronger level requires groundwater confirmation and the page states that demand, Grafham support and abstraction headroom are not fully observed. Independently, the 14- and 60-day groundwater level horizons beat their naïve baselines and are published for Central zones 1–6. A failed ban-warning replay does not suppress a passing level forecast.

South East Water, Anglian Water, South West Water and United Utilities use the same layered public warning. Climate can raise only the lighter level 2. A level-3 warning requires matching measured resources in the relevant supply area. South West's reservoir warning is not transferred to the Isles of Scilly, Anglian excludes Hartlepool, and United Utilities is labelled low confidence because its usable reservoir history begins after the 2010 ban.

Southern Water Hampshire and Isle of Wight is the sixth test and the second warning to pass publication. Four Environment Agency gauges reconstruct the supplier's Total Test Flow. A 28-day exponential recession is projected 35 days towards the statutory 355 Ml/day hands-off flow. The rule was frozen on the 2022 event and five post-2019 no-ban seasons, then caught both untouched later starts: 15 days before the 2025 announcement and 5 days before the 2026 announcement. Four of five no-ban seasons stayed quiet. The retained 2019 false TUB signal coincided with a real River Test drought-permit application. It is therefore an ordinal watch, not a ban probability. Separately, a numerical flow model trained through 2014, selected on 2015–2018 and evaluated on 2019–2025 publishes only its independently passing 14-, 60- and 90-day Total Test Flow horizons in Ml/day. Its 30-day horizon is withheld after failing the all-days interval sensitivity. The flow trajectories are not relabelled as storage percentages or complete seven-zone resources.

SES Water is the seventh research stress test. A rule frozen on the pre-2012 record combines two consecutive low-seasonal-tail months at Chipstead with low rolling six-month rainfall at Redhill. It replays the 2005 fitting episode and detects the 2012 restriction 34 days before announcement. The rainfall condition improves later quiet seasons from 10 of 14 to 13 of 14, but the result remains research-only: the Chipstead architecture was informed by the 2011/12 event, no untouched later restriction is available, and the reproducible Bough Beech archive starts in 2016, after both known restriction events. Bough Beech is now published as a separately gated resource component, not as a restriction probability.

The remaining national resource paths use independently sourced histories only. Yorkshire and London retain their licensed reservoir histories; Severn Trent retains its direct regional total; Southern Water retains river and groundwater indicators; Affinity, SES and Wessex retain their groundwater indicators; and Essex retains its Environment Agency reservoir history. Each output is labelled to its actual scope rather than being promoted to a complete supply-system measure.

Severn Trent's Strategic Grid result now leads with the supplier's broader direct-supply regional reservoir total. Its frozen 28-day trend passes held-out 14-day and four-week gates, while 60 and 90 days remain withheld. It is not presented as the full conjunctive supply-demand system.

Backtests are scope-aware. The current source-audited event register contains 31 rows across 12 model suppliers. South East Water’s 2026 Kent restriction and its later western expansion are separate events, so the model cannot earn credit for predicting the wrong area.

South Staffs supplies a second Level-2 counterexample and a data quality correction. Two withdrawn 2019 Environment Agency maps were storing watermark digits as Blithfield values: 30% is corrected to 80%, and 18% to 98%, with the rejected OCR values retained in lineage. The corrected level model still passes no publication horizon. The statutory Level-2 curve signalled in 2025, but the regulator records that South Staffs decided not to impose a TUB. Its 1976 restriction predates the public archive, so the separate warning audit remains research-only.