Model updates

Changelog

Every meaningful change to the simulation engine, methodology, or calibration is documented here. The goal is a model that gets less wrong over time, transparently.

Bug fixEngine

Model bias correction: fiscal-cost feedback

Internal analysis found that maxing all tax and spending levers produced a net-positive outcome when reality would produce severe negative consequences. Three structural fixes:

fix

Laffer sign-reversal bug — tax → GDP edges were accidentally producing positive effects at extreme tax rates because the Laffer curve flipped the sign. Fixed: GDP-drag edges now use raw (pre-curve) deviation.

new

10 fiscal-cost feedback edges — fiscal pressure now drags GDP (crowding out), raises political risk (bond market confidence), and pressures housing costs (rate rises). Corporation tax and NICs now have direct political-cost pathways.

cal

Political recalibration — income tax → satisfaction increased 0.4 → 0.6; income tax → GDP increased 0.3 → 0.35. Budget deficit → fiscal pressure now compounds over time (stock: true).

Net effect: max-tax-max-spend at 5 years now produces a clearly negative outcome instead of a net-positive one. The full causal graph and fiscal model are documented on the methodology page.

Engine

Non-linear response curves (Tier 1.5)

Replaced linear response on 21 policy levers with diminishing-returns and saturation curves. This is the single largest accuracy improvement since launch — it means maxing every spending lever no longer produces an implausible utopia, and raising taxes beyond a peak now reduces effective revenue as behavioural responses erode the tax base.

✓

4 tax levers now have Laffer-style inverted-U curves (income tax, NICs, VAT, corporation tax) — sourced from HMRC ETI coefficients, OBR analyses, and IFS Mirrlees Review.

✓

NHS and social care spending now saturate logarithmically — sourced from IFS, Health Foundation, and Nuffield Trust analyses of marginal productivity.

✓

5 housing levers use sigmoid/logarithmic curves with a physical absorption ceiling — sourced from Letwin Review, MHCLG, and CITB capacity data.

✓

10 further levers (training, enforcement, skills, defence) get heuristic concave/sigmoid curves — documented as expert judgment, not literature-cited.

Curves change magnitude of response, not direction. Pushing a lever the right way still moves outputs the right way — it just stops being proportional at the extremes.

Calibration

Detailed view launch: 38 levers, 19 outputs, 99 edges

A Detailed view of the simulator launched at pro.britain2036.app with the full Tier 1 lever set: welfare levers (two-child cap, PIP eligibility), wealth tax (3-lever composite with convex drag), non-dom regime, ECHR status, net zero target, defence spending with crowd-out, and climate-damage overlay. Includes international standing, child poverty index, and inequality index as new outputs.

Infrastructure

Snapshot regression gate

Automated regression testing: 24 scenario baselines on this view, 36 on the Detailed view. Every deploy must pass all snapshots before going live. Prevents accidental behavioural changes to the engine.

← MethodologyRun the simulator →