What it does. Each death model gives every simulated patient a survival curve without a transplant at listing. "Calibrated to UK outcomes" multiplies that patient's hazard (their risk of dying each day) by one fixed number per death model: M1 0.63, UKELD 0.87, MELD 3.0 1.30, GEMA-Na 1.34. A factor below 1 means the published model predicts more deaths than this simulated list produces; above 1, fewer.
How the numbers were chosen. Once, offline, on the default UK-sized run (starting list of 613, 83.5 retrieved livers a month, DBD livers ranked by the Transplant Benefit Score, super-urgent, ACLF and variant syndrome patients, suspension on, weight matching ±20 kg, removal for improvement on), seeds 1–10. For each death model the factor was searched for until 8% of elective patients had died or been removed as too unwell within a year of listing, NHSBT's figure for 2023/24 and 2024/25 registrations (annual report 2025/26, Figure 3.9; NLOS 96-month review, Table 6). Liver supply was fitted at the same time so that the list stays near 613.
It does not change with your settings. The factors are stored with the model and are the same for every run, whatever rule, supply or patient groups you choose. That is deliberate: if they were re-fitted for each run, every strategy would show 8% deaths and no comparison would be possible. Changing a setting changes deaths, and that change is the result.
Is it a fudge? In part, yes, and it is a declared one. It is the simplest recalibration used when a prediction model is applied to a different population or period: one number sets the overall level of risk. It does not change who is at higher or lower risk relative to anyone else, so the ordering the death model produces is unchanged. But the single number also absorbs everything else the simulation gets wrong: simulated patients, the assumed changes in bloods, suspension and decline rates. It is fitted to one outcome (deaths at one year) under one set of settings, and runs far from those settings (very different supply, say) are less tied to UK data. The factors were refitted each time the model's structure changed, not when settings change.
As published uses each equation with no factor, which shows how far each model's own predictions are from what this list produces: 11% deaths at 1 year with M1, 9% with UKELD, 7% with MELD 3.0 and GEMA-Na, against 8% observed.