Method & sources

Pandora is a deterministic engine — no LLM, no runtime randomness beyond seeded Monte Carlo. All figures are computed in EUR internally, then displayed. FX values use the disclosed snapshot below. Identical inputs always yield identical outputs.

30- and 35-year anchors: Morningstar, State of Retirement Income, Dec 2025; other horizons are Pandora's disclosed extrapolation. Secure anchor = 3.00%; Flexible premium = +1.20%.

Monte Carlo: up to 5,000 runs (Quick renders 1,000 first, then refines to 5,000 in the background), lognormal monthly returns, σ = 0.12. Seed = FNV-1a 32-bit hash of canonical inputs (excluding run count) — reproducible.

1. User statement override — always beats our estimate.
2. Years-based dated slices per system (UK flat, EU/US earnings, other flat/earnings).
3. Fallback offset — only if there are no dated slices.

A state pension is a function of the system you contribute to, never of what you plan to spend. Expectation estimates are anchored to the average full-career benefit of your country of residence; systems where expatriate workers accrue nothing (UAE, Thailand, Vietnam, and provident-fund systems) credit zero regardless of what you select. A figure you enter from your own statement always wins and is never trimmed. Dated slices start at each system's pension age; expectation estimates apply across the whole horizon (disclosed simplification).

Bridge & events use the same depletion solver as the Stress test, calibrated to the rate(H) anchors.

Stress test = -2.00% real for the first 10 retirement years, then 4.00%.

As of 2026-09: EUR/USD 1.15, EUR/GBP 0.859, EUR/CHF 0.94, EUR/AUD 1.65, EUR/CAD 1.48, EUR/SGD 1.45. (verify annually)

Display currencies convert via a quarterly-updated snapshot. Pandora deliberately does not use live FX: live rates would break the reproducibility promise (identical inputs, identical results) and add false day-to-day precision to a multi-decade plan.

The withdrawal anchors assume a globally diversified portfolio measured in real terms; they are deliberately identical for every passport. Passport effects are modeled where they truly occur — taxes, market access, estate exposure, and pension rights — through the flag engine and the state-pension ladder, not through the market math.

We replay your plan across 1,000 possible market histories. Worst 10% (P10) is the poor outcome — bad-sequence markets early in retirement. Median (P50) is the middle case. Best 10% (P90) is the lucky path. The Secure number is sized so a P10 shortfall is survivable, not fatal.

28 countries. Each carries a tier, currency zone, state-pension model, EU coordination flag, US totalization flag, exit-tax flag, and special-regime marker. Full country attribute types visible via the engine's registry.

20 rules across Tax, Pension-rights, Currency, and Structure; deductions 4–22 by severity; gold rules mark advantages.

Pandora is built and operated by Pensora. Scoring methodology © Pensora. The method is public; the exact weightings and corridor matrix are proprietary.

• Morningstar, State of Retirement Income, Dec 2025
• OECD Pensions at a Glance, Nov 2025
• Mercer CFA Global Pension Index, Oct 2025
• World Bank / Holzmann, Oxford WAR, Feb 2025
• IRC §§1291–1298 · IRC §2102
• PRIIPs 1286/2014
• EU Reg. 883/2004
• HMRC PTM

v3.2 — bridge-aware two-phase solver, one-off events, other-income streams, plain-language layer, consent + formatting fixes.

v3.2.1 — spending waterfall + self-funded line; editable spend + metro toggle; US-listed question made conditional with honest default; rounding polish; empty-profile nudge.

v3.2.2 — global input-currency selector; per-pot optional balances; one-currency results panel; state-pension anchor line; funding vs Monte Carlo reconciliation; 5,000-run Monte Carlo with quick-refine; Pro-mode per-tier ledger; Quick-mode declutter.