Designed Experiment App

The Experiment app runs a designed experiment: it varies two or more model inputs (called factors) over a planned grid of values, runs every combination, and fits a regression model so you can see how each factor affects a response.

You will need: Java 21 and a model bundle. This guide uses the LK (s,S) inventory model from the KSL Book Examples bundle. New here? Read Common UI & concepts and the Single-Model guide first.

What you'll be able to do

• Choose a model and pick the inputs to study as factors, each with low/high levels.
• Build a full-factorial design and preview its design points.
• Run all design points at once.
• Read a regression that quantifies each factor's effect on a response.

---

1. At a glance

You move left-to-right through the tabs — Model → Factors → Design → Simulate → Regression — building the study, then reading the fitted model.

Use this app when…	Use a sibling app when…
You want to measure the effect of inputs over a grid.	You have a few hand-picked configs → Scenario app
You want a regression / response surface.	You want the best inputs found automatically → Simopt app

---

2. Before you begin

Load the model from a bundle (see Common UI → Models and bundles). Results are written under your working directory. Enabling the database persists the run to SQLite for later analysis (e.g. in the Results app); the in-app Comparison Analyzer and regression work on the most recent run's results in memory and don't require it.

---

3. A guided tour of the window

1. Menu bar — File, Bundles, View.
2. Tabs — the workflow, left to right: Model, Factors, Design, Simulate, Regression, Comparison Analyzer, Reports. A summary banner under the toolbar always shows the current model, factor count, design, replications, and total runs.
3. Simulate — runs every design point (on the Simulate tab).
4. Console drawer — the run log.

---

4. Tutorial — a 2×2 factorial on the inventory model

Step 1 — Pick the model (Model tab)

On the Model tab, select the LK (s,S) Inventory model. The tab shows its controls (inputs you can study) and responses (outputs you can measure).

Step 2 — Define the factors (Factors tab)

Click Add Factor for each input to vary. Bind each factor to a model control and give it two levels:

Factor	Bound control	Levels
Reorder Point	Inventory.reorderPoint	20, 40
Order Quantity	Inventory.orderQuantity	40, 80

The factor editor (lower half of the screenshot in §1) lets you pick the control from a drop-down and type comma-separated levels. The banner updates to "2 factors · …".

Step 3 — Choose the design (Design tab)

Pick Full factorial. With two factors at two levels each, that's 2×2 = 4 design points. Set Replications → Uniform, 10 so each point runs 10 times (40 runs total). Click View design points… to preview the exact combinations before running.

Step 4 — Run (Simulate tab)

On the Simulate tab, choose an execution mode and click Simulate. A status table shows each design point advancing to Completed.

Reading the results — the regression

The headline output is on the Regression tab: a model of a response as a function of the factors. Below is the fit of Total Cost for this experiment (coded factors, −1/+1).

The four design points and their average total cost:

Point	Reorder Point	Order Quantity	Avg Total Cost
1	20	40	120.34
2	20	80	127.87
3	40	40	125.49
4	40	80	139.94

The fitted first-order model (R² = 0.82) and its coefficients:

Predictor	Estimate	Std Error	t	p-value	Sig.
Intercept	128.41	0.37	344.9	0.0000	***
Reorder Point	4.31	0.37	11.6	0.0000	***
Order Quantity	5.49	0.37	14.8	0.0000	***

How to read it. Both coefficients are positive and highly significant (p < 0.001), so over the studied range, raising either the reorder point or the order quantity increases total cost — order quantity slightly more (5.49 vs 4.31 per coded unit). The model explains 82% of the variation (R² = 0.82). The residual diagnostic plots check that the fit's assumptions hold:

---

5. Reference — every tab explained

Tab	What it's for
Model	Choose the model; view its controls and responses.
Factors	Define factors, bind each to a control (or RV parameter), set levels.
Design	Choose a design family (Full / Two-level factorial, Central composite, or Custom points) and set replications. View design points… previews the grid.
Simulate	Run all design points; watch the per-point status table.
Regression	Fit a linear model to a chosen response; open/save the report.
Comparison Analyzer	Compare responses across design points (box plots, CIs, MCB). Shared with the Results app.
Reports	Generated batch reports (HTML / Markdown / CSV).

---

6. Common tasks

Task	How
Study a third input	Add Factor and bind it; the design grows automatically
Reduce runs for many factors	Use a Two-level factorial (or fractional) design
See the exact runs before simulating	Design → View design points…
Add the interaction term to the fit	Choose an all-terms model on the Regression tab
Save / reopen the study	File → Save / Open (a .toml document)

---

7. Troubleshooting & gotchas

Symptom	Cause	Fix
Banner shows a validation warning	A factor's control isn't resolvable, or it has < 2 levels.	Re-bind the control; give each factor at least two distinct levels.
Two-level design rejected	A factor has more than two levels.	Two-level designs need exactly two levels per factor.
Regression looks "perfect" (R² = 1)	A saturated model (as many terms as points).	Add replications, or fit a smaller (first-order) model.
Comparison Analyzer is empty	No run yet this session (it reads the last run's in-memory results).	Simulate the design points first.

---

8. See also

• Common UI & concepts · Scenario app · Simopt app · Results app
• KSL Book — designed experiments and response surfaces.