The projection runs your plan under three models at once and shows all three, because a single success number hides how much the answer depends on what you assume markets do.
Forward · Baseline (normal shocks)
Draws each year's return from your assumptions (return, volatility, correlations) with normally-distributed shocks, applied as growth factors, so a single year can never lose more than 100%. This is the reference the other two are read against, which is why it is named for its job rather than for its distribution.
Forward · Fat tails (Student-t)
Same assumptions, but the shocks come from a fat-tailed Student-t distribution: ordinary years cluster closer to average while the rare bad years get worse, with total volatility held fixed. The Tail df knob sets how violent those rare years are (lower = fatter tails; ~6–8 is typical). Watch the failure depth, not just the success rate. A slightly higher success number with deeper shortfalls is not an improvement.
Historical replay
Ignores your return assumptions entirely and replays the 1928–2025 record using your allocation and historical returns, sampled per the historical sampling preset. It needs per-asset weights, so it runs under the per-asset allocation models, not Blended.
Read them as a range: Forward answers "what if markets behave like my assumptions", Historical answers "what if they behave like the past". When the three disagree sharply, the disagreement is the finding. Your plan is sensitive to the market model, so favor decisions that hold up under all three.