Read your results
The sensitivity sweeps, the driver ranking, and how to tell a sturdy valuation from a fragile one.
The Sensitivity page answers one question: which assumptions is this valuation actually hanging on? During the run, the agent designs a handful of sweeps: experiments over the one, two or three drivers it judges most important and most uncertain, each with a set of economically chosen test values. The engine then runs every combination deterministically and records the observed fair values. Nothing on this page is a language-model estimate; every number is a real engine rerun.
Each bar spans the fair values observed across one driver's tested range, with any other swept drivers held at their base values, anchored on the base fair value (the dashed line). The bars are sorted with the widest spans on top.
Read it in one pass: the top bar names the assumption this valuation is most exposed to. If the terminal growth bar spans three times as much as anything else, this valuation is mostly a bet on the long run. Hover any bar for the observed low and high and the swing as a share of the base value.
Each card below the ranking is one experiment. A two-driver sweep renders as the classic two-way matrix: rows and columns are the tested values, each cell the fair value the engine produced for that combination. Cell shading scales with the move against the base case, and cells with an amber outline are break-even: combinations at which the company is worth roughly today's price (within 5%). Those are the cells to argue about; the question is whether the world they describe is plausible. Cells marked n/a are infeasible combinations, usually a growth rate too close to the discount rate for a terminal value to exist.
One-driver and three-driver sweeps render as a table with one row per tested combination.
Select any cell or row for its detail: the swept values against their base, the observed fair value, and a per-point Excel model. The workbook is the same as the base model with those assumptions changed, so you can trace exactly how the change flows through the projection to the final number.
No single driver's tested range should carry the value past the current price on its own unless the write-up treats that assumption as the crux. When one assumption flips the verdict from undervalued to overvalued, that assumption deserves your attention before the headline number does.
Reports produced before the sweep rework show the earlier format on this page: a tornado chart plus a list of single-assumption tests.
Each bar is one recorded test, starting from the base fair value and extending to the value the model produced with that single assumption changed. Green pushed the value up, red pulled it down, largest swings on top.
Every test gets a row: the assumption tested and the direction of the swing. Expand a row for the before and after values, the agent's reasoning, and (where supported) a per-scenario Excel model.
For how volatility turns the fair-value gap into odds and timelines, see Reading the Volatility page.