DFS Simulation Sensitivity: Change One Thing at a Time

By Splash Play Staff · A guide to reading the tools

To see how much a DFS result depends on one assumption, change that one input, run it again and compare the same metric before and after. If you change the projection, the player pool and the exposure limits together, the result moves and you cannot say why. One change, one comparison.

Write the test down first

Before you touch anything, note four things:

The site and slate.
The settings you are leaving alone.
The one thing you are changing.
The number you will compare.

Writing it first keeps you from choosing the metric afterward, once you have seen which one flatters the change.

A worked example

Made-up numbers. A player is projected for 24 points on 28 minutes. You think he could play 32 with a teammate out. On the same per-minute production, 32 minutes is about 27.4 points (24 × 32 ÷ 28).

Example note: One change: 28 to 32 minutes. Projection 24 to about 27.4 (+3.4, or about 14.3%). Everything else unchanged. Not confirmed: whether he actually gets the minutes.

That last line matters. The test tells you what the projection would be if you are right about the minutes. It does not tell you that you are right.

Run it again, then compare like with like

If you edit a projection, run the simulation again so the results include the edit. Then compare the same thing on both sides: cash rate with cash rate, or top-10 rate with top-10 rate. Comparing the old projection with the new ceiling and calling the difference a gain is an easy way to fool yourself.

Small differences may be simulation noise. A tiny gap alone does not prove the assumption has no effect; check the run size and whether the difference persists before drawing that conclusion.

Keep both versions

Save the baseline and the alternative. Don't overwrite the first with the one you like better. If you want to test a second change, start again from the baseline and label it. A chain of five tweaks, each on top of the last, ends in a result nobody can trace.

What sensitivity is for

It finds the conclusions that are fragile. If four more minutes turns a player from a fade into your most-used play, your whole build rests on that one assumption, so go and check it: the injury report, the rotation, the last game he played without that teammate. If the change barely moves anything, stop worrying about it.

It will not tell you which input is true. That comes from the news, not from the model.

Where to see it

Simulation results are part of Squirt Squad VIP and Apex. Open the NFL tool for the current slate, read the field guide for the step-by-step workflow, or compare memberships. A sensitivity test makes a comparison clear. It does not make an outcome more likely.