# Winter supply, lake access and nuclear fleet comparisons

WattInfo, 26 September 2026.

## What is compared

The case specification records the demand year, historical weather year, number of new 1.1 GW reactors, continued operation of Gösgen and Leibstadt, import ceiling, lake conditions and any prolonged outage. Within the same demand year, weather sequence and heating assumption, solar, demand, existing hydro inflows, thermal/wind output and pumped-storage equipment remain matched when comparing nuclear fleets, import ceilings or lake constraints. The CSV gives both annual and winter unmet electricity, peak shortage, and gross and net winter imports. Unmet electricity is the additional supply needed under the case assumptions.

The 2050 zero-new-reactor rows retain the same 53 GWp solar and other resources. They measure the effect of omitting nuclear additions. They do not optimise a different renewable, storage or demand package or establish its cost. The three-unit rows assume construction is achieved; no third-unit delivery slot, site or price is established.

Each new reactor has 1.1 GW net capacity and 98% availability outside a chosen 30-day maintenance interval. Those intervals start on days 191, 226 and 261 of the calendar year. The third unit's maintenance therefore extends into October and reduces its winter contribution relative to the first two. This is an assumed calendar, not an intrinsic technology difference or an optimised maintenance plan.

For the three-unit case in 2016/17 weather with existing-plant extensions, recorded lake stocks and up to 2 GW imports, a separate calculation establishes a lower bound on unmet energy after removing every modelled maintenance hour for the third unit. This grants more generation than moving the same maintenance interval to another season, while preserving the unit's 98% availability allowance and every other resource. It restores both September and October hours. A feasible mathematical certificate from the original linear programme is evaluated against that added generation, with fixed lake stocks and all original constraints retained. It proves that at least 1.06 TWh remains uncovered over the year. The modified dispatch problem has not been solved: the bound is not its exact shortage, a winter allocation, a peak requirement or a feasible maintenance agreement. The code bundle includes the certificate, input case hashes and numerical arrays. The October-only diagnostic is separate and does not establish the full value of rescheduling.

## Solar and existing reactors

One multiplier fits a radiation-based hourly solar proxy to BFE's five published annual yields for 2020–2024. Its 2024 output is 835.3 kWh/kWp; annual fit root mean square error is 23.7 kWh/kWp. Annual agreement does not validate winter timing, snow, orientation or a future site mix. Earlier weather years lie outside that fitted period.

An unchanged-model check against the BFE 2015–2019 reports gives a root mean square error of 48.5 kWh/kWp; the model understates all five annual yields. This comparison also contains differences in the installed fleet and statistical methods. It does not identify an equivalent winter-yield correction. The source catalogue includes the five earlier reports.

Gösgen's planned-maintenance factor is 90.45%, based on 2016–2024. Leibstadt's is 85.53%, based on 2018–2020 and 2022–2025. These factors divide total generation by year-specific net capacity times calendar hours. The excluded years contain the separately documented prolonged outages. The Leibstadt value retains 2018 derating. The factors are conditional assumptions, not forecasts or statistical outage probabilities. Scheduled maintenance remains explicit. The extended-outage cases remove one large unit for a chosen 90-day winter period.

## Lakes and power

The free-lake case uses finite pooled natural and pumped storage, with cyclic annual stocks; the solver can start winter nearly full and draw natural stocks to zero. The observed-start case instead fixes the combined starting and ending stock to the BFE October level, converted by the inherited 98% delivery factor. Weekly values are interpolated to 1 October. This comparison tests an observed stock quantity, not the actual owner's dispatch rights or the physical connectivity of every lake.

The floor case also keeps natural stocks above 1 TWh throughout the year. That floor is a scenario choice, not a legal reserve requirement. The power case retains the observed starting stock and reduces natural-hydro turbine power to 80%; it does not impose the 1 TWh floor. The power restriction can increase the largest hourly shortage without materially changing the seasonal energy shortage. Seasonal electricity and power readiness are therefore separately specified services.

Reservoir inflows are a monthly accounting residual, not measured runoff. The model knows the whole year in advance, pools reservoirs and omits the internal grid. Named future hydro projects receive no additional output credit. Owner access and remuneration require contracts.

## Dispatch and schedule ranges

The dispatch first minimises annual unmet energy, then the largest hourly shortage, then uses a preference for lower imports and less cycling. The last preference favours exports over unused surplus. Exported surplus in one schedule does not establish sales, prices or an annual trade forecast.

The public gross/net import ranges hold annual unmet electricity, peak shortage and annual gross imports at the saved optimum, while allowing alternative feasible schedules. They are conditional schedule ranges, not confidence intervals. They do not minimise winter imports independently of annual imports. Report gross imports to show reliance on supply from abroad; net imports subtract surplus exported in other hours.

The separate direct winter-import minimum holds annual unmet energy and the peak gap fixed but lets annual imports change. It therefore finds the smallest gross winter import volume compatible with those adequacy targets. It establishes no rights to foreign electricity or cross-border capacity. The new winter-gap ranges hold annual unmet energy, peak shortage and annual gross imports fixed, and minimise/maximise the winter share of unmet energy. Repaired feasible witnesses preserve the endpoint and held quantities after removing simultaneous pumping and generation.

The CSV also gives the winter unmet energy and annual gross imports in the selected minimum-winter-import schedule. That annual import volume describes one feasible schedule; it is neither a minimum annual requirement nor a trade forecast.

Where a case also has summer shortages, its displayed winter gap can be one of several allocations of the annual total. The annual figure is included alongside such winter values. Hourly shortage timing and lake trajectories are not unique in general. No arbitrary dispatch week is used as a forecast.

## Reproduction and boundaries

The evidence JSON records case IDs, complete specifications, results and SHA-256 input manifests. The CSV exposes every matched case used in the comparison. The code bundle contains model and calibration code, package requirements and input manifests. It also needs archived third-party inputs and project-generated inputs acquired separately: original hourly ledgers, case and variant indexes, the hourly-artifact manifest and cached solar profiles. The ZIP alone is not a complete runnable dataset. The [earlier programme method](/downloads/programme-2050/method.md) and [earlier calculation bundle](/downloads/programme-2050/calculation-code.zip) describe its inherited inputs. The source catalogue records original URLs, dates, file hashes, locators and extraction coverage. Full third-party archives are not distributed here.

These are physical adequacy comparisons. They do not price grid reinforcement, establish bankable owner revenue, assign outage probabilities or determine a reserve purchase. Previous vehicle-service examples and owner financing calculations retain their separately stated assumptions. Their results cannot be subtracted from these supply requirements without a matched run.