Most planners show you a result and leave you to find the change that fixes it by trial and error. Solvers turn the question around. You name a target, such as a 90% chance of success, and the solver searches for the answer: a spending figure in money, a retirement month, a savings amount.

The solvers

  • Maximum spending: the most you can spend, optionally limited to chosen expenses, while keeping the chance of success at your target.
  • Earliest retirement: the earliest month a chosen milestone can happen, reported as a month and an age.
  • Required savings: how much a chosen savings flow must be. If you name a flow that cannot be solved for, the tool says which flows can.
  • Required portfolio: how large your invested balances must be, as a scale on the accounts you choose.
  • Safe withdrawal rate: the highest constant withdrawal rate that works over a horizon of years (30 by default).
  • Goal seek: any variable against any objective, within bounds you set, for questions the named solvers do not cover.
  • Tax-strategy ranking: compares candidate tax strategies (see withdrawal strategies).

Every solver returns the same things: an answer written as a sentence, the supporting details, the answer at neighbouring targets so you can see the trade-off, and the plan edits that would apply it. Nothing is applied until you choose to.

How one works

A solver runs the plan many times at different values of the thing it is solving for, each run a full Monte Carlo simulation, and narrows in on the value that meets the target.

target: 90%answer: about 39,800 a year30,00040,00050,000100%75%50%
Illustrative. Each dot is a full Monte Carlo run at one spending level; the solver narrows in on the target.
A spending solver at a 90% target. Each dot is a full simulation at one spending level. Illustrative figures.

Because each solver evaluation is a simulation, runs are jobs: they start in worker threads and report progress, and you can wait for the result or poll for it. Settings trade accuracy for speed: fewer trials per evaluation make a faster, noisier answer, and a cap on evaluations bounds the search. Runs use the seeded default draws, so asking the same question twice gives the same answer.

A worked illustration

On the synthetic UK sample household, asked at 90%, the spending solver answers in pounds a year what the household could spend instead of its planned £31,600 (about £36,100 in the project’s sample run), and offers the edit as a change set. Asked about retirement at an 85% target, the earliest-retirement solver gives a month and an age (about 56 and a half). These are illustrative results for a made-up household and one set of assumptions; yours will differ.

The same questions work over MCP. A prompt named how_much_can_i_spend runs solve_max_spending and quotes the figure per year; can_i_retire_earlier runs solve_retirement_date, explains the trade-off and offers a branch. Over the REST API they are POST /api/v1/plans/{id}/solve/max-spending, /earliest-retirement, /required-savings, /required-portfolio, /safe-withdrawal-rate and /goal-seek.

Reading an answer

A solver result carries two statuses. The job status says whether the work finished. The solver’s own outcome says whether it converged on a value or ran out of evaluations first. A solver that stops at its evaluation cap gives its best value so far, and says so.

An answer at 90% is not “safe”. It is the point where, under your assumptions and the chosen market history, nine of ten replays fund every year. Check the neighbouring targets, and read the failure analysis before acting.

What it does not do

  • It answers inside your assumptions. Garbage in, garbage out: if growth, inflation or tax settings are off, the answer is precisely wrong. The assumption audit catches common mistakes.
  • It is not an optimiser of everything at once. The tax-strategy ranking compares generated or custom candidates; it has no beam search or search-depth setting yet.
  • Randomised lifespan and care costs are not inside the simulations it runs.
  • It does not give advice. It tells you what your model says is possible.
  • No Flex versus No-Flex comparison table. Risk-based guardrails are compared instead.

Go further

See where your plan lands.

Free to self-host today, with every feature. The hosted service opens later, at $9 a month or $90 a year.