SimuXL User Guide

1Start here

What SimuXL is for, the three steps every model goes through, and where to go next.

SimuXL turns a spreadsheet that computes one answer into one that computes the range of answers your assumptions actually allow. You keep your model exactly as it is. You replace the numbers you were never really sure about with the ranges they always were, tell SimuXL which cells you want measured, and it recalculates your workbook a few thousand times and reports what came out.

Typical uses

• Cost risk - a bottom-up estimate whose quantities and unit rates are ranges, answering "how much do I need to be 70% confident?"
• Schedule risk - durations as ranges, answering "what is the chance we finish by the date we promised?"
• Cost and schedule together - a cost and a finish date from one model, answering "how often do we meet the budget AND the date?" - see Joint Outcome
• Risk registers - a probability and an impact per risk, answering "what is the whole register worth, and which entries drive it?"
• Engineering and design - tolerances, loads and material properties as distributions, answering "how often does this design fall outside spec?"
• Any model with an assumption in it - SimuXL never looks inside your formulas, so anything Excel can compute, it can put a range around.

The three steps

1. Turn assumptions into distributions. A cell that held 10 becomes =SXL.Triangular(8, 10, 14). Type it, or use the Input dialog - which can also take it as the percentiles you know, or fit it to your data.
2. Define your outputs. Select the cells whose answer you care about - a total, a finish date, a clearance - and press Output once.
3. Run, and read. Press Run Simulation, then look at the distribution of each output and at which inputs drove it.

What SimuXL will not do to your workbook

• It does not change your model. No SimuXL command writes a value into a cell you did not ask it to.
• It does not put its results among your own cells. An analysis is written to a sheet of SimuXL's own - by default a new sheet in your workbook, replaced on the next run - and none of your cells is edited. If you would rather each analysis arrived in a workbook of its own, that is one setting away.
• It works offline, and sends nothing about your work.

2What SimuXL is

A Monte Carlo add-in that lives inside your own spreadsheet - and the two words this guide uses precisely.

Most of a risk model is already in your workbook. What is missing is that some of the numbers you typed are not really single numbers - a duration is "eight to fourteen weeks, most likely ten", a unit rate is "about $420, give or take", a risk either happens or it does not. SimuXL lets you write those cells as what they actually are, then runs your model thousands of times and shows you the range of answers it produces.

You replace a number with a formula. =SXL.Triangular(8, 10, 14) is a duration that could be as low as 8 or as high as 14 and is most likely 10. Everything downstream of that cell - every SUM, every IF, every lookup you have already built - keeps working exactly as it did.

Two words this guide uses precisely

• An Uncertain Input is ONE distribution: the thing that gets sampled, that appears in the tornado, and that you can freeze or correlate.
• A SimuXL formula cell is a cell containing one or more Uncertain Inputs. One cell can hold two: =SXL.YesNo(0.05) * SXL.Triangular(10, 20, 50) is a 5% risk with an uncertain impact, and SimuXL treats it as two Uncertain Inputs in one cell. The analyses rank the two separately unless you make the cell one input - see Making one input from several.

3Quick start: your first simulation

Five minutes: two uncertain inputs, one output, a fixed seed, and a result you can reproduce.

Build the smallest model that shows what SimuXL does. It takes three short stages, and the recap at the end names what each one was for.

Stage 1 - the uncertain inputs

1. In a blank sheet, put this in cell D1 and press Enter: =SXL.Triangular(10, 20, 50). The cell shows 26.667. That is the distribution's average - see "What a SimuXL cell shows" below.
2. Put this in D2: =SXL.Normal(100, 15). It shows 100.

Stage 2 - the model, and the output

1. Put an ordinary formula in F1 that uses them both: =D1 * D2. This is your model. It can be as complicated as you like; SimuXL never looks inside it.
2. Select F1, then press Output in the Define group and press Save. This tells SimuXL which cell you want measured - once, for this workbook.

Stage 3 - the run

1. On the SimuXL tab, press Run Simulation. The dialog states what it will measure - 1 defined output across 1 sheet - rather than asking you for an address.
2. Leave Trials at 1000. Random seed already holds 12345 - SimuXL fills it so the run is reproducible before it starts rather than only afterwards from its run record - so leave that alone too and press Run Simulation. A small window shows the trial counter while it runs, with a Stop button.
3. A window tells you the run is finished and lists what it built. It does not take you anywhere - your selection is exactly where you left it - and you press Go to sheet to look at whatever you want to see. The SXL Trial Run sheet holds the mean, the standard deviation and the percentiles of F1, plus a row per Uncertain Input showing what it actually drew against what it should have drawn.

What you just did

• You defined two Uncertain Inputs - a Triangular and a Normal - by writing them as formulas, so both are legible in the formula bar to anyone who opens the file.
• You defined one output, F1, which SimuXL wrapped in a marker that travels with the cell and leaves its displayed value alone.
• You ran 1000 trials at a fixed seed. Each trial drew one value per input, recalculated your model, and recorded F1.
• You got a distribution rather than a number, plus a check that each input was sampled the way its formula says.

Then try the analyses

With that run still in memory, press Tornado and then Impact. On a two-input model they will agree. Add =SXL.YesNo(0.05) * 100 to the model and they will stop agreeing - which is the single most useful thing SimuXL will ever show you, and is explained under When something looks wrong. With two outputs - a cost and a finish date, say - press Joint Outcome to count how often both meet the thresholds you set.

4The eight distributions

Every one is an ordinary worksheet function. Start typing =SXL. and Excel offers the list, or use the Input dialog.

The dialogs call them by name rather than by function. Input, the Input as gallery and every analysis sheet list them as Triangular, Beta-PERT, Uniform, Discrete Uniform, Normal, Lognormal, Custom Discrete and Yes-No - the same eight, in that order.

What a SimuXL cell shows when nothing is running

Its average. =SXL.Triangular(10, 20, 50) shows 26.667 because that is the mean of that triangle. The cell is not a random number that changes when you press F9, and it is not the last value that happened to be sampled.

This is deliberate, and it has a useful consequence: your workbook, sitting there untouched, is a deterministic point - every uncertain quantity at its average. That is exactly what the Impact analysis measures deviation from.

Parameters can be formulas

Anything Excel can evaluate is a legal argument, so a distribution can be driven by the rest of your model:

=SXL.Triangular(B4 * 0.8, B4, B4 * 1.3)

The arguments are re-evaluated on every trial, so if B4 itself depends on a SimuXL cell, the relationship holds trial by trial.

In the Input dialog, type the formula into the parameter's box - =B4*0.9 for a Minimum that is 90% of B4 - or press the … button beside the box and click a cell or type the formula there. The box is shaded and shows what the formula works out to now; click in it to see the formula.

Entering an input by percentiles: the Parameters list

If what you know about an input is a judgment rather than its parameters - a value you would be surprised to come in under, a likely one and one you would be surprised to exceed - choose how you enter it in the Input dialog's Parameters list, under Type. A Triangular or a Beta-PERT offers Minimum, Most Likely, Maximum; 10%, 50%, 90%; 5%, 50%, 95%; 10%, Most Likely, 90%; 5%, Most Likely, 95% and Custom…; Uniform, Normal and Lognormal offer their own two parameters, 10%, 90%; 5%, 95% and Custom…. The label above each box says what it holds, and the value you type there means exactly that: with 10%, 50%, 90%, the first box is the value at P10. The distribution does not change - a Triangular entered by its P10, P50 and P90 is still a Triangular. The preview draws it, and the line between the chart and the boxes says the distribution's own parameters your values give - for a Triangular, its Minimum, Most Likely and Maximum.

Choosing another set keeps the same distribution: when the boxes already describe one, each box whose label changes is filled with that distribution's value at its new label, and a box whose label does not change keeps what it holds. Custom… lets you choose, box by box, between the distribution's own parameter and a percentile you type. Beta-PERT's Lambda is always its own box. Set Default makes the set shown the one a new input of that distribution opens on in this workbook.

The set is kept in the formula, after the name, so it travels with the cell and the dialog shows it again when you reopen it:

=SXL.Triangular(80, 100, 130,, "P10,P50,P90")

The three numbers are the values at P10, P50 and P90, not a Minimum, a Most Likely and a Maximum. SimuXL works the Triangular's own parameters out from them every time the cell calculates, so a box can hold a cell such as =D6 and the run follows it. An input entered by its own parameters has no Parameters argument. Each item of the set, in order, is that place's own parameter as the formula spells it (Minimum, MostLikely, Maximum, Mean, StdDev) or P and a percentage.

What you typed is shown beside what SimuXL solved. Find Inputs gives it in its last column, Entered as - P10 80; P50 100; P90 130 for the formula above - and leaves it empty for an input entered by its own parameters; the Input Summary says it on a line of its own in the block it writes for that input, and the HTML and Word reports after the parameters in their Input Summary section. A set held in a cell rather than typed into the formula is not shown.

Not every set of values has a distribution of that kind: for P10, P50 and P90, a Triangular's gap from P50 to P90 can only be between 0.62 and 1.62 times its gap from P10 to P50, and a Lognormal takes only values above zero. The line above the boxes then says why no distribution gives them, and OK says so too. Where more than one does - a Triangular whose Most Likely lies below every percentile you gave, for one - SimuXL takes the Triangular or Beta-PERT whose Most Likely is nearest its Minimum or its Maximum, or the Lognormal with the smaller spread, and the line says so. A set held in a cell rather than typed into the formula (,,B1) is read when the cell calculates, but the dialog cannot show it.

Fitting a distribution to data, or to percentiles you know: Fit…

When you have observations, or judgments, and want to see which distribution they point to, press Fit… beside Type in the Input dialog. It opens SimuXL - Fit a distribution, which fits to one of two sources: A range of data or Percentiles you know. What you choose there comes back into the Input dialog; nothing is written until you press OK.

A range of data: type a range, or press … and point to it - one block of cells, on any sheet of the workbook. SimuXL reads the cells in use, fits to the numbers among them and leaves every other cell out - blanks, text (a number stored as text included), TRUE or FALSE, and errors - counting each kind and naming the cells. It never changes your data: a value beyond Tukey's fences (1.5 times the interquartile range beyond the quartiles) is flagged and kept. Fewer than 3 numbers, or a constant, are refused; fewer than 20 are warned of, and so are values that look discrete - fewer than 10 distinct values among 20 or more, or whole numbers with fewer distinct values than a third of their count. When none of these applies, the space under the range says what was checked, so an empty space never stands for an answer.

• Normal. Mean: the mean of the values. Standard Deviation: their sample standard deviation (divided by n - 1).
• Lognormal. Fitted on the natural logarithms of the values - their mean m and sample standard deviation s (divided by n - 1) - and written as the Lognormal's own Mean (arithmetic), exp(m + s²/2), and Std Dev (arithmetic), Mean × √(exp(s²) - 1), which is how SimuXL's Lognormal is set.
• Uniform. Minimum and Maximum: the smallest and largest values, each moved out by one average gap - the range divided by n - 1 - because a sample's own ends are inside the distribution's (the minimum-variance unbiased estimate).
• Triangular. Maximum likelihood: up to 1,000 values, each value in turn is tried as the Most Likely and the Minimum and Maximum searched for it, and the best is kept; above that, the Minimum and Maximum are searched over a grid and the best points refined - rather than one search from one start. A bound the Most Likely is not on is then finished where the likelihood's slope is zero (Newton's method). The Most Likely can land on the smallest or largest value - most often with few or skewed values - and the Minimum or Maximum is then that value (written to ten figures at or outside it).
• Beta-PERT. Maximum likelihood over the Minimum, Most Likely and Maximum, with Lambda held at 4 (the standard PERT): searched from the best 3 of 27 starting points, and with the Most Likely at the smallest or the largest value tried on its own - the Minimum or Maximum is then that value (written to ten figures at or outside it). The best is then finished where the likelihood's slope is zero (Newton's method). Neither a Triangular nor a Beta-PERT is fitted to more than 20,000 values: each step of its search reads every value.

A-D (Anderson-Darling) and K-S (Kolmogorov-Smirnov) measure how far your values sit from each fitted distribution - smaller is closer. They are descriptions, not tests: the parameters were estimated from these same values, so the usual tables of significance do not apply. Look at the plots as well. Nothing is ranked or called best, and on this side nothing is chosen for you unless the Input dialog's own type was fitted. The histogram draws your values with every fitted curve, the chosen one strongest; the Q-Q plot draws the chosen family - each value against the family's value at the same share - and a close fit lies along its line.

Press Use this distribution, then OK in the Input dialog: the formula holds the fitted numbers, and nothing else, and the fit's record is added to the SXL Fit sheet - the input, the formula, when, the data range, how many cells and numbers, the cells left out, the method, every family's parameters, A-D and K-S, the notes, and both charts. In your workbook, each fit adds its record below the earlier ones and removes nothing; when your sheets go to a new workbook (Preferences > Analyses), each record opens one of its own. Change the distribution or a box after the fit and OK writes the cell without a record, because the record would describe a different input.

Percentiles you know: type two or three percentiles and the values you judge for them, and every family is fitted to them side by side - exactly where it can, otherwise as closely as it can, and each says which. The Input dialog's own type is chosen whenever it fits, otherwise the first family that does. A family that gives every percentile you typed, and is the only one of its kind that does, comes back as a Parameters set holding your values, kept in the formula as described above - when you typed as many percentiles as its set has boxes: three for a Triangular or a Beta-PERT, two for the others. Anything else comes back as the family's own numbers. Press Use this distribution, then OK, and this fit's record is added to the SXL Fit sheet too - the input, the formula, when, the percentiles you typed and the workbook's percentile convention, the method, every family's parameters and whether it gives your percentiles, the chosen family's value at each of them, and one chart: its cumulative curve with your percentiles marked.

Which inputs were fitted. Whenever SimuXL scans your workbook - Find Inputs, a run, Impact, a comparison - it reads the workbook's SXL Fit sheet and finds, for each input, the newest record for its cell. Find Inputs names it in its Fitted column - for example To data: Data!$A$2:$A$63, 60 values, 2026-10-05 16:30 - SXL Fit row 1; the Input Summary names it in the block it writes for that input, in shorter words - what it was fitted to, when, and the record's row; and the HTML and Word reports list the fitted inputs in a table of their own when The listing of every Uncertain Input is ticked under Preferences > Report. When the cell no longer holds the formula that fit wrote, the words begin Not as fitted. A record names its cell by address and by a hidden name, SXL_Fit_0001 and on, which Excel moves with the cell: an input moved since its fit - rows inserted above it, cut and pasted, or on a sheet you have renamed - is still named, and the next of those scans to find it moved rewrites the record's heading and its Input line to the cell's address now (the worksheet function SXL.HarvestInputsArray writes nothing). Without its name - deleted by a macro, say, or one Excel would not add - a record has only its address, so an input moved since is not named and the input now in its old cell is said to be not as fitted. A record written to a new workbook is not read.

5When the at-rest value is not a base case

The workbook sitting there untouched is a deterministic point. It is not the answer's average, and on three common shapes of model it is nowhere near it.

Take the workbook one cell at a time: every SimuXL formula shows its distribution's mean when no simulation is running, so the number already in the cell is that distribution's average. The sheet as it stands is therefore a deterministic point - reproducible, stable when you press F9, and the point the Impact analysis measures every swing from.

A distribution that CHOOSES rather than SIZES

A distribution used to CHOOSE something rather than to SIZE something settles its branch at rest, and settles it the same way every time. A Draw cell reads exactly 0.5 with nothing running, so IF(Draw < P, Impact, 0) reads ZERO for every chance at 50% or below - the comparison is strict, so a risk at exactly 50% is off - and the full impact for every chance above it. The total at rest is neither the register's mean nor any trial it produced.

An average the cell can never draw

On a distribution with gaps in it, the average is a value the cell can never actually draw. =SXL.YesNo(0.05) shows 0.05 and =SXL.DiscreteUniform(1, 6) shows 3.5 - a gate that is 5% on, and a die that rolls 3.5. That is the correct average and it is deliberate; it is also why Discrete Uniform, Custom Discrete and Yes-No compute nothing meaningful at rest when they are choosing rather than sizing.

An output that bends

If your model bends, the base case is not the answer's average and cannot be, whichever way it bends. A formula that is flat on one side of a threshold and sloped on the other, a MIN or a MAX, a product of uncertain factors, any risk gate - each pulls its own average away from its value at the centre. That is a property of the model rather than a defect, and it means Impact measures swings from a point your output may almost never occupy.

6Defining inputs and outputs

The dialogs, what they write into your cells, and why the formula is the definition.

Input as - straight to one distribution

The Input as menu beside it lists all 8 distributions with a line each on what they are for, and opens the same dialog already on the one you picked. Nothing else about it differs - the parameters, the live curve and the formula SimuXL writes into the cell are identical either way.

Erase - taking SimuXL back off a cell

Select the cell and press Erase. It answers all three kinds of SimuXL cell, and it tells you what each one will become before it writes anything. Select several cells to erase several at once.

An output or a control variable goes back to what its cell held before it was defined. Both WRAP something, so the original is still inside the formula: SXL.Output(A1*3, "Cost") puts =A1*3 back, and a control variable leaves the plain value behind. The tint comes off with it, and the cell goes on showing the same number - the wrapper never changed what it computed.

An Uncertain Input is different, because there is nothing underneath it. The distribution call IS the cell, so there is no earlier formula to restore. SimuXL leaves the distribution's mean, written as an expression: =SXL.Triangular(10, 20, 50) becomes =(10 + 20 + 50) / 3. It stays live, so a parameter that lives in another cell goes on working, and it shows its own arithmetic three months later where a bare 26.666666666666668 would not.

What each distribution's cell is left holding, each row produced by the same rule that does the erasing:

Input - the dialog, if you would rather not type

Select a cell - empty, or holding a number or text - and press Input (the Define group, first on the tab). A cell holding one of your model's formulas is refused: the formula is part of your model, and a distribution written over it would break the calculation. A cell holding one SimuXL distribution on its own opens the dialog to edit it. Pick one of the eight distributions, fill in its parameters, and the curve redraws as you type - so you can see the shape you are describing before you commit to it. Press OK and SimuXL writes the formula into the cell.

• It writes a formula, not a hidden definition. The dialog is a convenience; the formula is the input. Anything you can do in the dialog you can do by typing, and the other way round - but a Parameters set held in a cell (see Entering an input by percentiles: the Parameters list), which only typing can make.
• Each parameter box takes a number, a cell (=D6) or a formula (=D6*0.3) - and the preview follows a reference, so you see what that cell's current value does to the shape. A box holding a reference is shaded amber and shows you the VALUE it resolves to, not the text you typed; click into it and it shows the reference again, so you can always see both.
• A name is optional. Leave it blank and the input is reported by its cell address; type one and every chart and table uses it instead. The Tornado, Impact and Tail Driver tables print the cell as well, in a Cell column beside the name, so a named input can still be traced back to where it lives. The name box takes a cell too - =A6 names the input from that cell's text, and a relative reference follows when you copy the cell. The Name button writes into this same argument, so the dialog and Name are two ways to set one thing.
• On a cell that already holds a SimuXL input, it opens in EDIT mode - titled Edit SimuXL Input, with the existing parameters and name loaded, so reopening a cell changes it rather than starting again.
• It refuses on a cell holding an ordinary formula, because replacing =D1*D2 with a distribution would destroy the model you built. It also refuses a cell holding MORE than one distribution, or one inside a larger formula, which you can still edit by typing.

Output - naming the cells you want measured

An output is a cell whose distribution you want to see. Select the cells and press Output (the Define group), once, and press Save. Every run after that measures exactly those cells.

Saving writes the definition INTO THE CELL, as SXL.Output wrapped around whatever the cell already held. =D1*3+D2 becomes =SXL.Output(D1*3+D2, "Total cost"), and the cell goes on showing exactly the number it showed before - the function returns its first argument and does nothing else.

• That is what makes an output survive editing. Copy the cell, fill it down, move it, or insert a row above it, and the definition goes with it, because Excel maintains the formula the way it maintains every other formula.
• Removing an output here puts the cell's original formula back. The marker contains what it replaced, so nothing is lost either way. ⚠️ Excel's own Undo does not reach an add-in's writes, so Ctrl+Z is not the way back - this is.
• A cell holding a plain number can be an output too. 1700 becomes =SXL.Output(1700, "Base"), still worth 1700.

• Outputs on DIFFERENT SHEETS work. A typed range cannot express that at all - an Excel range covers one worksheet, however many areas it has - so if your model reports a total on one sheet and a schedule date on another, this is the only way to measure both in one run.
• Every chart, table and exported column then says the NAME rather than Sheet1!$F$7. Names have to be unique, because an exported CSV uses them as column headings and two columns under one heading is a join that silently addresses the wrong one.
• SimuXL suggests a name from the cell to the LEFT when that cell holds text, which in most cost and schedule models is already the label you would have typed. Type over anything you do not like - pressing Save commits whatever is in the name box, so you do not have to press Rename first.
• Or point at the cell that holds the label, with the … button beside the name - the same affordance the input dialog has. The box then shows that cell's TEXT on an amber ground, and the reference again as soon as you click into it. It is a LIVE link: edit that cell and the output's name follows it. The reference is RELATIVE when the label is on the output's own sheet, so filling a wrapped cell down a column names each copy from the label on its own row.
• Nothing is written until you press Save, including removals - so Cancel really does leave the workbook alone.

7Rules SimuXL enforces

The things that stop a run, why each one stops it, and what to do instead.

SimuXL refuses to run a model it cannot sample reliably, and it refuses rather than guessing, because every failure below produces a run that completes with plausible numbers and no sign anywhere that they are wrong. Each rule is stated the same way - the rule, why it exists, what to do - followed by worked examples and, at the end, the instrument that answers all of them on your own workbook.

Rule: the same SimuXL calls on every trial

Rule

A cell must make the same number of SimuXL calls on every trial.

Why

An Uncertain Input is a cell plus a POSITION, and each position owns a fixed place in the seeded random stream before trial 1. If a conditional skips a call, the positions after it shift up - so one place in the stream stops meaning one distribution.

What to do

Put each distribution in its own cell and let the formula choose between those cells.

It is the rule behind the refusal a risk modeller is most likely to meet, and it is narrower than it first looks. You can absolutely make a cell's answer depend on a chance - the product's headline shape does exactly that. What you cannot do is make a formula choose whether a SimuXL call happens at all.

Where only one call is written, SimuXL is instead being asked to sample a position that sometimes does not exist. Either way the run would complete and the numbers would look entirely reasonable.

Which Excel functions skip an argument - measured, not assumed

Only a function that can decline to evaluate an argument can break this rule, and the unit that matters is the ARGUMENT rather than the function - a distribution is safe in an argument Excel always evaluates and unsafe in one it may skip. These were measured in Excel - the conditionals on 11 August 2026, LET on 23 August - a cell per construct with a distribution in the arm that is NOT returned, each against a control cell proving it fires when it IS returned:

Which Excel functions call MORE than once - measured, not assumed

Excel's six LAMBDA helpers run a LAMBDA many times over, and they are neither of the things above. They skip nothing - each one calls once for every thing it iterates, and what it iterates differs between them. So one call site written in your formula draws many times, and the number changes when your data does. Measured in Excel on 17 August 2026, on the same instrument as the table above:

What to do instead

Put each distribution in its own cell and let the formula choose between those cells. =IF(A1>0, SXL.Uniform(1, 2), SXL.Normal(3, 4)) becomes three cells: =SXL.Uniform(1, 2) in one, =SXL.Normal(3, 4) in another, and =IF(A1>0, D5, D6) where the original was. The model means the same thing, both distributions are visible in the formula bar, and both are Uncertain Inputs the tornado can rank and the Correlation Editor can reach.

Rule: a cell must not draw more than its formula shows

Rule

A cell must not make MORE SimuXL calls than its own formula text contains.

Why

That is almost always a defined name that expands to a SimuXL call. The assumption is then invisible in the formula bar, so nobody reviewing the model can see it - and SimuXL refuses for the same reason it refuses the case above.

What to do

Write the distribution into a cell of its own, where the formula bar will show it.

The row on the run record this rule has a twin in

The rule above is checked when SimuXL scans your workbook, which is one moment. A model whose call count depends on its own numbers can pass that scan and then draw MORE times on some later trial - so the run record carries a counter for exactly that: Evaluations past a cell's known calls.

Each cell gets as many places in the random stream as SimuXL saw that cell use when it scanned. A call past the last of them gets no draw at all: it comes back as the distribution's own mean. The row counts those calls over the whole run, and zero is the number you want.

Rule: a required argument must not be left empty between two commas

Rule

Every required argument must have something in it. =SXL.Triangular(,20,50) - one stray comma - is refused.

Why

Excel hands an empty argument slot to SimuXL as zero before SimuXL is entered, so that cell samples Triangular(0, 20, 50) and shows a perfectly reasonable number. Nothing downstream can reveal it: the call fires, it is counted, every total agrees, and the model runs on a minimum you never typed.

What to do

Put a number in the slot, or delete the extra comma. Find Inputs names the cell and the parameter, and Run Simulation refuses until it is fixed.

When a trial errors, and when an input misbehaves

Rule

An output that is not a number is yours to decide about. A cell that makes an unexpected number of SimuXL calls always stops the run.

Why

An output error is your model's business and you may legitimately want the trial dropped. A call-count disagreement is an INPUT that misbehaved: the result would be drawn from the wrong distribution, and nothing on any sheet could tell.

What to do

Choose what a bad output does under Preferences > Errors. A call-count disagreement is not a choice - fix the cell SimuXL names.

• An output cell that is not a number - a #DIV/0!, some text, or the empty string an =IF() gives when its condition is not met. None of those can go into an average. What SimuXL does about it is yours to choose: see Stop the run on a calculation error under Preferences > Simulation.
• An unexpected number of SimuXL calls in a trial. SimuXL knows how many your model should make. If a trial makes a different number, some Uncertain Input either did not fire - and is then silently stuck at its average - or fired twice and consumed its neighbour's random number. This always stops the run, whatever the setting above says.

The message box tells you the outcome and, if something went wrong, one sentence saying what. The detail - which trial, which cell, which error, and what SimuXL did about it - is on the run record in the workbook that opens, because that is the artefact you keep and screenshot. A run that excluded some trials and kept others still reads Completed as its status, because it did complete - so the status word carries the excluded count with it everywhere it appears, and every sheet carries a line saying which trials its figures describe. A run that kept none at all is reported as a failure, because there is nothing to compute a result from and a sheet of dashes headed Completed reads as a SimuXL defect rather than as a model that errors on every draw.

Find Inputs - what SimuXL thinks your model contains

Find Inputs is in the Other drop-down of the Define group.

Press Find Inputs at any time. SimuXL recalculates the workbook once, notes every SimuXL function that fires, and writes a listing of each Uncertain Input - its sheet and cell, its distribution and the arguments it was actually called with. The listing goes to a new workbook or onto a SXL Uncertain Inputs sheet of this one, whichever you have chosen under Preferences > Analyses, and it is replaced each time you press it. Nothing else in your workbook is touched - no formula, no value, no formatting.

It reads every sheet of your workbook. The scoped recalculation box on the Run dialog narrows what Excel recomputes during a run - it never narrows what SimuXL scans, and there is no way to leave a sheet out.

Reading the arguments this way rather than parsing the formula text is what lets a parameter be a formula, a name or a cell reference - Excel does the hard part, and SimuXL sees the numbers your model really produced.

It lists everything the scan finds, not just the inputs

The same scan that finds your Uncertain Inputs also finds your defined outputs and your control variables, out of the same formulas, and Find Inputs reads your made inputs - SXL.MakeInput cells - the way a run reads them before it starts. The listing carries a block for each: INPUTS, ISSUES, OUTPUTS, CONTROLS, MADE INPUTS. A marked cell SimuXL will not use appears in its block with the reason in the Status column, rather than only in the message that refuses. An output or a made input SimuXL cannot use stops a run and Impact; a control variable it cannot use stops a comparison and leaves an ordinary run alone.

It reports what it cannot account for, rather than guessing

• A call inside a branch that was not taken - =IF(A1>0, SXL.Normal(1,2), 0) with A1 at zero - is reported, and it stops a run. It is narrower than it looks - gating an uncertain input in one cell is supported - and it has a remedy: see Rule: the same SimuXL calls on every trial above.
• A SimuXL function name this build does not have is reported, not ignored.
• An argument that cannot be a valid parameter is reported.
• A call reported from a different open workbook is excluded and said so. A recalculation is application-wide, so SimuXL sees other workbooks' cells and must not merge them into yours.

8Running a simulation

Trials, a seed, stopping and pausing, and how to watch a run go past.

A run can be paused or stopped. A small window shows the trial counter and the elapsed time for as long as the run lasts, with Step, Pause and Stop buttons; Excel's status bar carries the same count.

• Pause holds the run between two trials so you can look at the sheet, and Resume carries on from exactly where it stopped. Nothing is lost and nothing is re-drawn: the seeded stream picks up at the next trial, so a paused run that finishes is an ordinary complete run. The elapsed time on the report does not count it: SimuXL stops its clock while a run is held, deliberately, so Elapsed and Per trial go on measuring your model rather than how long you spent looking at the sheet - and the run record's Held for you row says how much was not counted.
• Stop ends the run for good. Everything completed so far is kept, so the report and the analyses still work - over however many trials there are, and every artefact says so.
• Neither is instant. SimuXL sees a press the next time it looks at the window, which is always between two trials, and acts there. It looks after every trial on a watched or paced run, and about every 66 ms of calculation on a run at full speed. On an ordinary model that is a fraction of a second, which on a fast one can be many trials past the counter you were reading; on a model whose single recalculation takes seconds, it is seconds.

Watching a run - pacing, and Step

A normal run turns Excel's screen updating off, because redrawing the sheet on every trial would cost more than the trial does. That is right for getting an answer and useless for checking that your model does what you think it does - so there are two ways to slow a run down and watch it.

• Tick "Watch the run" in the Run dialog and set a hold in milliseconds. SimuXL leaves the screen painting and pauses that long after each trial, so you can see the drawn inputs land and the model recalculate. The dialog tells you what it will cost before you commit: at 250 ms a trial, 10,000 trials is over forty minutes.
• Press Step on the ribbon (the Simulation group) to start a run that holds after every trial. Press it again - on the ribbon or on the progress window - to draw exactly one more. This is the way to check a formula that only misbehaves on some draws.
• Step works during a run too. Pressing it while a run is going holds it, and from then on each press draws exactly one more trial. Resume carries on without holding; the sheet stays visible from then on, and a paced run goes back to its pace.

The other two ticks on the Run dialog

• Show the distribution as it builds draws one output's histogram in the progress window and keeps it up to date while the run goes past, so a distribution that is obviously wrong shows itself in the first few hundred trials instead of at the end. On a run with more than one output the window carries a picker for which one to follow, and you can change it mid-run. The picture refreshes on a clock - every 250 ms - rather than on every trial, so what it costs does not grow with the trial count; leave it unticked and the run allocates nothing for it at all. It changes no number: the run, its record and every analysis are identical either way.
• Scoped recalculation - faster on large models recalculates only the sheets between your Uncertain Inputs and your outputs on each trial, instead of the whole workbook. On a large model that is worth a few milliseconds a trial, which is minutes over a long run. ⭐ SimuXL works those sheets out from your formulas, refuses to scope at all if it cannot prove the answer, and checks itself against a full recalculation on a sample of trials while the run goes - if a single check disagrees it stops scoping and finishes the run the ordinary way. The run record says what it did either way, so a run you are unsure about can be settled by reading it rather than by running it twice. A comparison run recalculates the whole workbook whatever this box says.

If a run does not finish

Reset clears a stuck run state - use it if something went wrong and SimuXL still believes a simulation is in progress. It changes nothing in your model.

9Latin Hypercube

Spreading the draws evenly - what it buys, what it costs, and why a seed alone does not identify a run.

SimuXL normally draws every Uncertain Input independently every trial, which is plain Monte Carlo and is what a run does unless you choose otherwise. Tick Latin Hypercube on the Run dialog and it does something different: it splits each distribution into as many equal slices as there are trials in a design, and takes exactly one draw from every slice. Nothing about your model or your assumptions changes - only which random numbers come out.

• It helps most where SimuXL currently has to print a warning instead of an answer. A Yes-No at 5% fires a lucky or unlucky number of times under plain Monte Carlo, and the whole upper tail of anything it drives is built out of those firings. Under Latin Hypercube it fires in exactly 5% of the trials whenever that 5% works out to a whole number of trials per design, and within one trial per design otherwise. A run is laid out as 20 designs, so a 4,000-trial run gives designs of 200 and a 5% gate fires exactly 10 times in each of them, at every seed. A gate at 0.15% in the same run works out at 0.3 firings per design, so most designs do not fire it at all and a few fire it once - close to nominal over the run, and not on it. The remaining wobble is entirely in the sizes the gate multiplies, not in how often it fires.
• Where the gain lands depends on the model, and the honest summary is that it varies enormously. Measured over 300 repeated runs at 4,000 trials: on a 5% risk gate the P90's run-to-run spread is about five times narrower, and on a product of uncertain factors the mean's is about five to six times narrower. A sum of independent triangulars - the textbook case, and the one no real model looks like - improves by more than a hundredfold on its mean and hardly at all on its tail. On any real model it is worth having and on none of them is it magic.
• Correlations still work, and SimuXL swaps the method quietly. Plain Monte Carlo correlates by transforming the drawn numbers; that transformation would undo the even spread, so under Latin Hypercube SimuXL correlates by reordering them instead. Both land on the same achieved correlation, so every number in the correlation report means what it always meant. You do not choose between them and the run record says which was used.

10Naming, freezing and correlating

The three things you can attach to an Uncertain Input - a name goes into the cell's own formula, and the other two are what SimuXL stores in your workbook.

Name

Gives the selected cell a name that appears in every report in place of Sheet1!$D$6. The three analysis tables print both - the name, and a Cell column beside it - because a name tells you what an input is and only the address tells you where to go and change it. It works on an Uncertain Input or on a defined output - press it on either and the box opens on the name already in that cell's formula. Leave the box empty to remove it. On a cell that is both - an output whose own formula is a distribution - it names the input; use Define Output to name that output.

Freeze / Release

Holds the selected input at its average for the whole run, so you can ask "what does my answer look like if this one is not uncertain?". Press again to release.

Your formula is never touched. Freezing does not write a value into the cell; SimuXL simply stops pushing a sample into it, and the function falls back to the average it shows when nothing is running. The cell is tinted grey so you can see at a glance that it is frozen, and your own fill colour is put back when you release it.

Correlate

Correlated inputs move together: a high draw on one makes a high draw on the other more likely. Correlate opens the Correlation Editor, which lists every Uncertain Input in the workbook - no selection needed.

1. Pick Variable A, the input every pair in this batch will be correlated with.
2. Tick as many Variable B as you like. The search box narrows the list, and ticks survive a change of filter - tick three under one search, three under another, and you have six.
3. Type one coefficient, strictly between -1 and 1. The word beside it and the ceiling note below it both follow what you type.
4. Press Add Pair. One pair is written per ticked B, all sharing that coefficient.
5. Press OK. Nothing reaches the workbook before that, including removals, so Cancel really does leave it alone.

• Only the inputs you correlate are affected. Declaring a correlation between A and B does not change C's draws at all, so a before-and-after at the same seed is a fair comparison.
• Every sheet that names the sampler also says what you correlated. On a run with nothing correlated the Sampler row says every draw was independent; on a run with something correlated it says how many pairs there were and what induced them, and a Correlated pairs row states the same fact beside it. That row is on the Convergence and comparison sheets too, so a reader holding one of them never has to open another to find out what the Sampler row's claim leaves out.
• You declare a Gaussian r, and inducing it costs a little. The rank correlation that actually arrives is (6/pi)*arcsin(r/2), so 0.8 becomes 0.786 - and that is the Expected Spearman column. The table carries Requested r, r used, Expected Spearman, Achieved Spearman and Ceiling (measured) side by side on every row. That much is how the method works, not an error.
• A rare Yes-No cannot reach a high correlation with anything, however you declare it. At a 5% chance the ceiling is about 0.38 - an input taking only a few distinct values caps the rank correlation it can reach with anything at all. The editor says so beside the coefficient box before you commit, and the run report fills in its Ceiling (measured) column afterwards when that limit is what you are seeing.

Input Settings

Input Settings is in the Other drop-down of the Define group.

Lists everything this workbook has saved against its Uncertain Inputs - freezes and correlations - and whether any of it no longer matches the formulas. A model with nothing frozen or correlated stores nothing against its inputs, and this report says so. It shares its drop-down with Find Inputs: the two of them report on your inputs rather than on a run.

11Making one input from several

Rank a risk's Yes-No and its impact as one input in the Tornado, Tail Drivers and Impact, by marking the cell that combines them.

A risk on a register is usually two Uncertain Inputs - a Yes-No that says whether it happens and a distribution for how much it costs - combined in one cell: =D6*E6, or both in one formula. Every analysis ranks Uncertain Inputs, so on its own the risk arrives as two half-answers. A made input is that combining cell, ranked as one input in their place.

1. Select the cell that combines the risk's inputs - H6 holding =D6*E6, say.
2. Press SimuXL, Define, Make Input, or right-click the cell and choose SimuXL ▸ Make Input...
3. Type a name, or leave it blank to name it after its cell. SimuXL rewrites the cell as =SXL.MakeInput(D6*E6, "Ground risk") and tints it light blue. The cell shows the same number it did.

For a whole register, select every combining cell at once and type the letter of the column that holds the risks' names - A, say. Each cell takes the name on its own row, as =SXL.MakeInput(D6*E6, A6), the same relative reference a fill-down would give it.

Find Inputs lists your made inputs before any run, in its MADE INPUTS block: each one's name and the inputs it is made of, and every SXL.MakeInput cell a run would refuse, with the reason.

From the next run the Tornado and Tail Drivers - and Impact, the next time you measure it - rank the made input as one row in place of the Uncertain Inputs it is built from, followed through other cells, other sheets, defined names and tables. Each sheet counts the inputs it carried under Excluded, and the made input's Distribution column names them - the first three, then how many more. The Input Summary folds nothing: it adds the made input's own distribution after the inputs, which keep their rows.

Impact is where the number changes. It swings a made input's inputs together - all at their low ends (the risk does not happen), all at their high ends (it happens at its worst), and a middle point where it happens at its most likely size, with every Yes-No among them at 1 and every other input at its most likely value. Its bar is the output's move from the risk not happening to the risk at its worst, so on a total that adds a 5% risk of 200 to 600, one bar spans the whole 600 where its two inputs drew two shorter ones. Its Swung From and Swung To read the made input's own cell.

What a made input cannot follow, and what to change

SimuXL decides which inputs a made input is built from before the run, by reading the formulas - it does not watch Excel calculate. Anything whose cells are only known while Excel calculates is therefore out of its reach, and so is anything Excel will not show it. What happens then depends on where it sits: inside a made input - in its formula or anywhere in what that formula reads - SimuXL refuses rather than guess; on the path from an output back to its inputs, it runs but folds nothing. The table lists the common cases; every message and note names the cell and the reason.

SimuXL, Define, Erase takes SXL.MakeInput off the cell again and leaves the formula it wrapped.

12Control variables

The cells that hold a DECISION rather than a chance, so a later comparison can set them by name.

What a control variable is

An Uncertain Input is a chance SimuXL takes for you and an output is a number SimuXL measures. A control variable is neither: it is a lever you set - a markup, a contract type, an escalation rate - held in a cell so that a comparison can set it by NAME, one value per alternative, without touching anything else in the model.

=SXL.Control("Bid Markup", 0.08)

The marker returns the value unchanged, so the cell goes on showing 0.08 and every formula that reads it reads the same number it always did. There is no stored list anywhere - the cells themselves are the definition, so a control survives a row insert, a copy and a sheet rename for the same reason an output does.

1. Type the value your base case uses into a cell of its own.
2. Select that cell and press Define ▸ Control Variable.
3. Press Add, type the name the comparison will use, and press Save.

What a control variable may never do

A control variable sets a VALUE. It can never change which SimuXL calls fire: a cell whose distributions appeared and disappeared with the alternative would be sampled differently in each one, and the comparison could not say what had changed. To take an uncertainty out of an alternative, freeze that input instead.

This is the one rule worth reading twice, because the most natural-looking lever in a risk register breaks it. A cell like =IF(ActiveVariability="Yes", SXL.Triangular(10, 20, 50), 30) looks like exactly the switch a scenario wants - and it changes whether a SimuXL call happens at all, which is the shape the call-count rule refuses. Put each distribution in its own cell and let the formula choose between those cells.

What uses them

The Alternatives button sets up your alternatives in a dialog and lays them out as a table, one row per control, with a column per alternative - see the Alternative Table section. Set Alternative then puts your model on one of those columns, and Restore Base Case puts it back. Run Comparison runs the whole table in one press - every column, against the same random numbers, and writes the results onto one sheet. What that sheet holds is numbers AND pictures: three charts per output, drawn on the sheet below the table - see the Alternative Table section. Comparison Report then writes the whole thing as an HTML and a Word document you can send to somebody who does not have your workbook.

A control stays out of every count and every analysis that is about uncertainty - a control cell is not an Uncertain Input, so there is nothing about it for a tornado to rank, and a run behaves exactly as it did before it was defined.

13The Alternative Table

Set up in a dialog and kept on a sheet you can still type into: one row per control variable, one column per alternative, and a Base case column SimuXL keeps up to date.

What it is

Press Comparison ▸ Alternatives and SimuXL opens a dialog to set up the alternatives you want to compare - see Setting up the alternatives, below. OK writes them to a sheet called SXL Alternatives, holding one row for each of this workbook's control variables and one column for each alternative, with its name in the header row and its value for each control underneath.

It is an ordinary worksheet and it is meant to be. You can widen the columns, colour it, print it, put notes beside it and refer to it from your own formulas. It is saved in your workbook like anything else, so it is still there when you reopen the file, and a colleague who opens the file can read the comparison without SimuXL installed.

Setting up the alternatives

Alternatives opens one dialog for the whole table. The list on the left holds your alternatives - a new table starts with Alternative 1 and Alternative 2 - and Add and Remove change how many there are. Choose one, and on the right give it a Name and a Description, then change the control variables it changes: pick a row in the list of control variables and type the value in the Value box under it.

OK writes only what the dialog changes. A cell whose value is already what the dialog holds is not written. A cell holding a formula is never written over: OK names it instead, so you can change it on the sheet. Removing an alternative the table had removes its column from the table - the alternatives after it move one column left, each with its name, description and values, and nothing outside the table's rows moves - and OK asks before it does.

Typing values without leaving a dialog

Press Comparison ▸ Edit Values and SimuXL shows one alternative at a time: every control row, what the Base case column holds, what this alternative holds, and what each row does in a comparison. Pick a row, type its value in the box underneath, and pick the next one. Change which alternative you are looking at with the list at the top.

It types into cells that already exist and does nothing else. It never adds a row or a column, never removes one, and never renames an alternative. Adding, removing and renaming alternatives is done in Alternatives, above, or on the sheet - where you can undo your own typing with Ctrl+Z, which you cannot do to anything SimuXL writes.

The three columns SimuXL reads

The table is found by its header row - Status, Control Variable, Base case, in three cells side by side - so you can move the whole thing down or across and SimuXL will still find it. One row says Description, and that row is worth its own heading below.

The Status column

Status answers, for every row at a glance, whether that control is taking part in a comparison. SimuXL rewrites it every time you press OK in Alternatives, and it says one of two things: Defined when the workbook has a control variable of that name, and Not defined when it does not.

Holding a control at its base value

Sometimes you want a control variable to sit still while you compare the others. Press Comparison ▸ Hold at Base and SimuXL lists every control variable in this workbook with what it does in a comparison beside it. Tick one to hold it at its base value; untick it to let your alternatives change it again.

The same thing can be typed straight onto the sheet: put Held at base in that row's Status cell and clear it to release. The button and the cell are the same setting - the button writes that cell, and nothing about a hold is stored anywhere else. So a colleague can read your sheet and see what was held without having SimuXL installed.

A control variable with no row at all in the table is held too - nothing can change what has no cell - and the list says so. Untick one of those and SimuXL adds its row, holding its base value in every alternative column, and asks first: it is the only thing this dialog does that writes into your alternatives.

SimuXL says which controls are held in three places: the message after OK in Alternatives, the dialog before a comparison runs, and the SXL Comparison sheet itself. The last one matters most - it is what tells a reader six weeks later that a control showing the same value everywhere was held deliberately rather than simply not mattering.

Naming and describing an alternative

An alternative's name is whatever you type in the header row above its column. That is what SimuXL calls it everywhere - in the picker, on the comparison sheet, and in any message about it - so a name you would recognise in a list six weeks from now is worth the ten seconds.

An alternative's description goes in the row whose first cell reads Description, directly under the header row. Alternatives creates that row for you; give an alternative its description in that dialog, or type into the cell under its own column. Nothing else has to be filled in, and SimuXL writes into it only from the dialog.

A description is text and only text. It is never read as a value, never set into your model, and never compared against anything - so write whatever helps you choose.

A mitigation cost in your model

An alternative that reduces a risk usually costs something, and that cost belongs in your model, like every other cost. Give it a control variable - in an empty cell, type =SXL.Control("Mitigation cost", 0) - and add that cell to the formula of the output it costs, such as your total. Its base value, 0, is your model as it stands, with no mitigation; each alternative sets its own cost in Alternatives, the way it sets any other control variable, and an alternative that costs nothing leaves it at 0.

The Base case column is SimuXL's on the control rows

On every control row, the Base case column is redrawn from your control cells every time you press OK in its dialog. SimuXL writes an alternative column or the Description row only where the dialog holds something different from the sheet, and never a Base case cell. So the base case cannot drift away from your model - if you change a markup in your model next month, press the button, press OK, and the table catches up.

Adding a control variable later

Define a new control variable and press Alternatives again: the dialog lists it at its base value under every alternative, and OK adds a row for it, under the last control, holding what the dialog shows. Change it in the dialog wherever an alternative should use something different. You never have to delete the sheet to pick up a new control, and your alternative columns survive.

Putting your model on one alternative

Press Comparison ▸ Set Alternative and choose a column from the list. Every control variable in the workbook is set to that column's value, and any control the column has no row for is put back to its base value - so what your model holds afterwards is that alternative and nothing left over from the last one. Then run a simulation as you normally would.

Restore Base Case puts every control variable back. Press it before you save, before you send the file to anybody, and before you trust a number the model shows you - a model left on an alternative still calculates, and the figures still look ordinary.

Running the whole table

Press Comparison ▸ Run Comparison and SimuXL runs one simulation for the Base case column and one for each of your alternative columns, then writes a sheet called SXL Comparison holding all of them side by side: one row per alternative, the mean and standard deviation, a percentile ladder from P0 to P100 in tens, and how far each alternative's mean and standard deviation moved from the base case.

A dialog opens first, listing every column with its description beside it - see Naming and describing an alternative below. Every column is ticked, so pressing the button straight through runs them all; untick the ones you do not want this time - they stay on the sheet, so you never have to delete an alternative to avoid running it. The base case always runs, because every difference on the comparison sheet is measured from it. The trial count is yours to change and starts at whatever Report Preferences says; the seed comes from there too, and the dialog shows you which one it will use. There is no seed box here - a comparison has one seed for the whole batch, so it follows your seed setting rather than asking: a fixed seed or the last run's seed is recorded as (from Preferences), and under a new seed each run SimuXL takes one fresh seed for the whole comparison and records it as (chosen by SimuXL). Nothing is written to your workbook until you press Run Comparison.

A comparison also clears what SimuXL remembers about your last single run, so the Analysis buttons have nothing to work from until you run one again - and neither do any =SXL.Result(...) cells you have written, which report that no simulation is being held until you run one. The Run Comparison dialog says so before it starts, and the comparison sheet says so afterwards.

Every output gets three charts, on the comparison sheet itself. The first is a cumulative distribution - one S-curve per alternative, on one pair of axes; when your workbook reads percentiles as Above, it is an exceedance curve and is named so, each curve falling from left to right because a point is the share of trials at or above its value. Where two curves CROSS is the finding a table of percentiles cannot show you: it is the value above which the alternatives swap places. The second is a histogram of each alternative, turned on its side so its values run up the chart: each bar is the share of that alternative's trials in its band of values, and every alternative is counted in the same bands and drawn to the same scale, so a longer bar is a band more of its trials fell in. The third is a distribution shape panel - one violin per alternative, side by side, with a rule across each at its median. Each violin is its histogram set on its own centre line - as wide at each band as that band's bar - and drawn as one outline, so reading the second chart beside the third is the way into the violins if you have not met them before. The widths share one scale, so a wider shape is a more concentrated result and a narrower one is a more uncertain result; a violin scaled to its own peak would make every alternative look equally certain. The base case is drawn heavier than the rest on all three charts.

Two targets for every alternative

Tick Count two targets for every alternative in the Run Comparison dialog and, when you press Run Comparison, the Joint Outcome dialog opens: choose two outputs, a threshold for each and how each is met, as you would for a single run. Nothing runs until you press Use these targets, and Cancel there writes nothing. The tick is offered when SimuXL can read at least two output cells before the run.

The comparison sheet then carries a TARGETS block above the outputs: the two targets, then one row per alternative with the share of trials in which Both met, each of the two outcomes in which one target was met and the other missed, Both missed, each target met whatever the other did, and the change from the Base case for both met and for each target. A change is the alternative's share minus the Base case's, in percentage points, printed with pts - a fall from 62.0% to 50.1% is -11.9 pts - and the cell holds the points, so a formula that reads it gets -11.9.

SimuXL, Report, Other Reports, Comparison Report writes the whole comparison as a document you can send to somebody - an HTML file and a Word file, both from one press, beside your model and named after it. It is the "SXL Comparison" sheet laid out for a page: the same record, the same alternatives, the same percentile tables and all three charts for every output. SimuXL does not re-run or re-measure anything to write it, so a number in the document and the same number on the sheet cannot disagree. Each output prints as one table with your alternatives across the top and the statistics down the side - so reading down a column is one alternative's distribution and reading across a row is what one percentile does as the alternative changes, which is the comparison. A comparison of more than seven alternatives is wider than a page, so it is printed in parts with the statistic names repeated on each and a line saying which columns you are looking at.

14The analyses

The ways to see what your model does, two ways to write the whole run up, plus the raw data. None of them edits your model.

Convergence - was this run long enough?

Convergence is in the Other drop-down of the Results group.

Re-measures the last simulation after fewer trials and reports how much each output's answer was still moving. Nothing is re-simulated - it reads the values the run already kept. What a row is depends on how the run was drawn, and the sheet says which it did: on a plain Monte Carlo run each row is half the one below it, and on a Latin Hypercube run each row is one more of that run's own designs.

A chart beside each output's block draws the same six lines the report's convergence figure draws - the mean, its 95% band, and the three tracked percentiles - on the same span, so the sheet and the report are one picture of one run.

• On a plain Monte Carlo run the mean gets a 95% band, and that band is arithmetic: it shrinks as the square root of the trial count, so four times the trials halves it and a hundred times narrows it tenfold.
• On a plain Monte Carlo run a percentile gets no such band, because there is no formula for one that survives a distribution with a gate in it. What is reported instead is how far each tracked percentile moved when the trials were doubled, measured in standard deviations of the output itself.
• The mean settles first and the tail settles last, and the tail is usually the number that gets quoted. A run whose mean is settled to a tenth of a per cent can still have a P90 that moves visibly every time you lengthen it.

A Yes-No at 1% fires about 10 times in 1,000 trials, and the whole upper tail of anything it drives was built from those 10 while the mean looks beautifully settled - so the sheet names it. Scale the trial count against the number of times the rare outcome actually happened, not against the trial count. If the run did not keep its input draws this check could not run at all, and the sheet says so rather than reporting no gates.

Histogram - the distribution of each output

The first thing to look at, and usually the only picture a reader wants. One chart per output cell: columns showing the SHARE OF TRIALS that landed in each bin, and the cumulative curve on the right-hand axis. The columns sum to 100% - they are not a density, whose height would change with the bin count.

View Results - the same picture, in a window, writing nothing

Every other analysis opens a workbook the moment you press it. This one opens a window instead, and writes NO SHEET until you press Write to sheet - which then writes the output on screen, with its chart, where the Histogram button writes, from the same code and the same preferences. So you can try six parameter sets without collecting six workbooks to close. To write several outputs at once, press Histogram and tick them. When SimuXL writes its sheets into this workbook, either one replaces the SXL Histogram sheet each time.

The window is modeless - leave it open, change a parameter in your model, run again, and open a second one to compare. Each window keeps the run it was opened on and can say which: the run time and the seed are the first two rows of the Statistics panel, which you reveal with the Percentiles and Statistics boxes. That matters because two histograms of one output from two seeds look identical at a glance and mean different things.

It has four tabs - Distribution, Tornado, Impact, Tail Drivers. The three analysis tabs show the same TABLES their ribbon buttons write, as read-only grids, and each carries its own Write to sheet handing off to that button's own writer. One difference from pressing the button: each of these three tabs always covers every output of the run, because a window with no picker in front of it can only mean all of them, where the ribbon button asks you which to include first. Impact measures HOW FAR one input can move an output; Tornado measures how CONSISTENTLY the output follows it. They can rank the same model differently, and reading one without the other is the most common reason a model looks broken.

Each analysis tab opens on its CHART, with its table behind a Show table box. Tick Show table and the same grid appears underneath it, on the same tab, listing the rows the analysis kept; untick it and the chart has the tab to itself. The chart is re-laid either way, so revealing the table shrinks the picture rather than covering it. Every tab starts with the box clear, including the next window you open, and it changes only what you are looking at: the analysis, the sheet Write to sheet produces, and every number in either, are the same whether the table is showing or not.

Configure this chart - and where a base value lives

Configure this chart... in the viewer sets one output's OWN chart settings: its base value, and optionally its own markers, shaded split and colour pair overriding Preferences > Charts. Pressing Save there stores them in your workbook - that is the one thing this window writes, and it writes it so the setting survives closing Excel. Cancel writes nothing.

On a DATE output there is a date picker beside the box, and the two are live together. Pick a date and the box fills in; type a date - or the Excel serial number - into the box and the picker moves to it. They are the same value said two ways, and whichever you use is what gets stored. ⚠️ Excel's calendar starts on 1 January 1900, and so does SimuXL's. A date before that has no Excel date number at all, so typing one is refused and the box says so. A NUMBER below 1 is not a date either: it stays a number in the box, which is where a plain figure belongs on any output. Excel also carries a date number 60 for a 29 February 1900 that never happened - a quirk it has kept since the 1980s - and no picker can show a day that does not exist, and typing that date is refused with a sentence of its own - not because Excel lacks the day, but because Windows does, so there is nothing for the box to read it back as or for the picker to show. Its Excel number is 60, and typed that way it is taken and kept as the number. ⭐ Charts and reports then DRAW it the way Excel does, as 29 February 1900, because the cell on your sheet does. ⚠️ The picker cannot be scrolled back past 1 January 1900, or past whatever later date your own regional calendar begins at. A cost output and a clock-time output get no picker, because neither is a date.

A base with a time of day is rounded to midnight. If a stored base carried a fraction of a day - noon on 25 January reads 44951.5 - the box shows 1/25/2023 and pressing Save stores the date. Nothing on a chart shows a time of day, and a base value that is half a day off the date beside it is a difference nobody can see and everybody has to explain.

The line is called "Base" by default, and there is a User defined label box if you want it called something else. A base read from your workbook as it stands is called "Base" too - the chart label carries the VALUE, and the sentence under the chart says where the number came from, because a derived figure should not read like a stated one and a bracket on the label is not where that belongs.

• The rule is dashed, so it cannot be mistaken for a percentile marker's solid rules.
• A base outside the distribution is drawn anyway, and the chart stretches to reach it. The columns fill the span between the outer edges of the grid - which is not always your smallest and largest trial: with no shaded split it is exactly those two, and with a shaded split the whole grid is nudged so a column edge lands on the split value, which puts the first edge a little below your minimum and the last a little above your maximum. A base outside that span is still drawn: SimuXL adds empty bins until the grid reaches it, so the distribution is squeezed into part of the plot and the base line sits out in the empty part, which is what shows their relationship. It will not squeeze the distribution below a quarter of the width. Past that the value is stored and not drawn, and Configure this chart... says so, names how far out you went, and prints the furthest figure that would draw.
• The value is a row on the histogram sheet as well as a rule on the chart, because a rule can be clipped and the label carries a claim a reader has to be able to check.
• A nominated cell has to name its sheet - Cost!$B$4, not $B$4. An unqualified address means a different cell depending on which sheet happens to be in front when the chart is drawn, and a base value that moves is not a base value. The … button beside the box fills it in correctly for you.

Tornado - how CONSISTENTLY each input drives each output

Ranks the Uncertain Inputs by how reliably the output follows them, using signed rank correlation. It ignores size: an input that always nudges the answer a little in the same direction scores highly.

Each row names the input twice: the Input column is its NAME if it has one, and the Cell column beside it is always the address. An input you never named reads the same in both, which is the point - the name column falls back to the address, so without the second column a NAMED input had nowhere to say where it lives. The same pair appears on the Impact and Tail Driver sheets, in the HTML report and in the Word report.

One column is headed Contribution to Rank Variance, and the word Rank is load-bearing. It is not a decomposition of your output's actual variance, and the sheet says so in its own words. Each share is never negative - the sign of the input's ρ is the direction - and where inputs share an effect, because they are correlated with each other or one keeps its own row while also sitting inside a made input, the shares overlap and a Shares overlap row under that output says so. It names each correlated pair at least one of whose shares is 0.05% or more, and counts the other correlated pairs among the rows shown, both of whose shares are under 0.05% and round to zero; an input inside a made input is held to the same test, and an output where nothing passes it has no Shares overlap row. The Highest |ρ| possible column beside it is filled in when an input's rank correlation is capped by how rarely it varies rather than by how little it matters - the figure is the largest |ρ| that input could have reached with this output at all.

Impact - how FAR each input can move each output

Swings each input across its whole range, one at a time, and reports what each output did in its own units - dollars, weeks, whatever your model produces. It ignores likelihood.

It needs no simulation. It re-evaluates your live model, so you can press it on a workbook that has never been run. The base case is your workbook exactly as it sits, with every uncertain quantity at its average.

Tail Drivers - what actually ran atypically at each END of the range

Takes the trials where an output landed in its lower or upper tail, and reports which inputs ran unusually high or low in those trials. Unlike the other two this looks at what happened rather than at what could. Which end is the good news is a property of your model and SimuXL does not assume it - on a cost or a duration the lower tail is the good news, on a revenue or a schedule float it is the upper one.

A driver is one Uncertain Input or one made input, not a risk, and it is ranked by how far its average in the tail trials sits from its average over every trial, counted in its own standard deviations. That rewards being OVER-REPRESENTED in the tail, not occurring there: a 50% risk that fires in 60% of the upper-tail trials barely registers, while a 5% risk that fires in 40% of them leads. To see a risk's gate and impact as one driver, make them one input - see Making one input from several.

• Two tables per output, one per tail, and they routinely have different winners. They are different sets of trials, not two views of one ranking.
• Rows marked ~ are inside the noise floor. With many inputs and few trials, an input entirely unrelated to the output still shows a shift by chance, and a top-five table shows the largest shifts - which is exactly where noise wins. Each row carries in the Noise floor (SD) column the floor it had to clear. That is the same figure on every row unless an input is extremely lopsided, because only a very skewed input reaches further by chance in the direction it leans - so a column reading the same on every row is the ordinary case and not a fault. It is a rough guide, not a significance test: roughly one unrelated input exceeds it anyway.
• A negative bar means the input ran LOW in that tail, not that it pushes the output down. The chart title names the tail.

Joint Outcome - how often two outputs BOTH meet their thresholds

Pick two outputs of the last simulation - any two: a cost and a finish date, a peak flow and a duration, a margin and a yield - and for each, choose under Met when whether it is met at or below or at or above its threshold, and type the threshold. SimuXL takes each trial's two values together, from the same trial, and counts the trials in each of four outcomes, named after your outputs: Both met; First missed, second met; First met, second missed; and Both missed - for outputs named Total cost and Finish, met at or below and on or before their thresholds, "Total cost met, Finish missed" is a trial that came in at or under its cost threshold and finished after its date. The share in which both are met is counted from those pairs - never multiplied from the two separate shares, which is right only when the two outputs are independent - and two outputs driven by the same risks rarely are.

• A value exactly on its threshold meets it, whichever way it is met. A cost equal to a threshold it must stay at or below meets it, and so does a yield equal to one it must reach at or above.
• For an output shown as a date - first or second - the threshold is a day, and the list reads on or before or on or after. Type it as a date the way your computer writes a short date - the line under the box shows one - or as the number behind one; any time on the threshold day counts as that day, so it meets the threshold either way. An output in any other format - a duration, or a date in a cell not formatted as one - is compared exactly, and its threshold is a number in the output's own units.
• The four counts add up to every trial the run kept. A trial the run excluded - one in which any output of the run could not be read as a number - is in no count and not in the denominator, and the sheet says how many there were. The shares are shown to one decimal place, so the four of them can add to anything from 99.9% to 100.2%.
• The sheet is SXL Joint Outcome: each output with its threshold and what met means for it, the four outcomes with their trials and shares, each output's separate share beside the joint one, the joint-confidence frontier's table when it is drawn, the READING block, and under them a scatter chart - the first output along the x-axis and the second up the y-axis, as the dialog's First output (x-axis) and Second output (y-axis) say - one point for each trial it draws, a line at each threshold, and each corner's outcome, share and count; which corner is which follows the directions you chose. A date is drawn at its day, the day it is counted by. Each axis's ticks fall on round values - for a date, whole days, and Mondays when they are a week or more apart - no more than the labels leave room for, at most 8 - of those, the spacing that frames the trials closest, with four labels or more where it can - in the output's own format, a number's label the exact value of its tick. When the output's format cannot label its axis that way - its labels too long for the axis, or too coarse to show any round tick without stretching the axis to over twice what the trials need - the axis is labelled in plain numbers instead (a date as m/d/yyyy). When a run keeps more than 3,000 trials the chart draws 3,000 of them, spread evenly through the run, each the pair of values one trial gave, so the cloud stays readable; its legend says how many and what share - Trials - 3,000 of 10,000 shown (30.0%) - and the sheet and the report say the same. The counts, the four outcomes' labels and the frontier always use every trial. A workbook whose structure is protected cannot take the sheet, so SimuXL writes it to a new workbook and its message says so.
• SimuXL does not read your choices back from the workbook. In one Excel session the dialog opens on the thresholds you typed and the directions you chose last - in any workbook - and on each output you chose last that the run still has. Where it has not, the first output is the first that is not a date and the second the first other that is - a cost and a finish date, in a cost-and-schedule model - or simply the next output when there is none of that kind; and the first time, both are met at or below - on or before, for a date. The sheet records what it counted. Press Joint Outcome again with other thresholds to count the same trials against them.

The joint-confidence frontier. Under the second output the dialog has a tick, Draw the joint-confidence frontier at, and a level in percent. It starts ticked, and the level starts at the workbook's Shade either side of under Preferences > Charts when that is above 0 and below 100, otherwise at 70; in one Excel session the dialog opens with the tick as you last left it, and keeps a level you typed. For each threshold of the first output, SimuXL takes the trials that meet it and finds the lowest threshold of the second output at which both are still met in at least the level's share of the trials - at 70% of 10,000 trials, the 7,000th lowest second value among them. Pressing Joint Outcome with a row's two thresholds counts both met in at least that many trials, and with the row's second threshold made stricter by one in its last digit shown - by a day, for a date - in fewer. For a date the threshold is the earliest; for an output met at or above its threshold it is the highest (the latest for a date). The line is the same whichever output is swept; the table reads it by the first.

• On the chart it is a purple staircase through every corner of the frontier, from the frame's edge to its other edge, and the legend names it. When it has more than 9,998 corners it joins every n-th one, keeping the first and the last; between two it runs at the first one's height, never stricter than the frontier there, and the sheet says how many it joined.
• On the sheet, the JOINT-CONFIDENCE FRONTIER table gives a row at the frontier's first point, at each tick of the x-axis between that and its last corner, and at its last corner: the first output's threshold, the second output's best threshold there, and Trials meeting - how many trials meet the first threshold. A row with few trials past the number needed rests on few trials, so one trial can move it; the first row is the least steady. Under the table the sheet says where the frontier starts - no stricter first threshold gives the level - and where it stops moving - a looser one buys nothing.
• The frontier is not drawn for a comparison's targets, and untick it to count without it.

The reading. Above the chart the sheet has a READING block: findings written by rules from every counted trial - never from the trials the chart draws - and then a bottom line. The words say what the trials show and never why, and they give no advice about your project.

• Each alone, and both. The share of trials meeting each threshold and both together, and which output is the harder to meet - the one that misses its threshold alone more often. When the two counts of trials in which only one is missed are within what chance alone could make (McNemar's test, at about 95%), it says neither is clearly the harder.
• Together or apart. Both met, against what multiplying the two shares would give; how often the second is met where the first is met, and where it is missed; and the two outputs' rank correlation over every counted trial, printed to three decimals, with SimuXL's word for that figure. What it then says of meeting the two together leans on the four counts as well: a correlation is over every value, and meeting is at your two thresholds. Where the counts lean the same way as the correlation it says: below 0.1, that both met is within what chance alone could make of multiplying the shares (a test of independence at about 95%); where, in meeting terms, the correlation runs against, that the two are a trade-off; at 0.9 or more with the two running together in meeting terms, that they move almost as one, with both met beside the harder one on its own; otherwise, that each is met more often where the other is met than where the other is missed. Where the counts lean another way, it says so and gives what they say.
• Where your thresholds sit, when the frontier is drawn: whether both are already met at its level; if not, holding each of your thresholds in turn, the other threshold with which both reach the level and how far it lies from yours (for a date, in days and weeks), resting on the trials that meet the one held past the number needed - so a threshold of yours that sits at the frontier's start gives its figure with the hedge that one trial can move it - or that none does, because the one held is met on its own in fewer trials than the level needs, with that count.
• What a step along the frontier buys. The frontier is read where the first output's threshold is met on its own in a set share of trials - at 75%, 80%, 90% and 95% for a 70% frontier - each with the second output's threshold there, the share of trials it meets on its own, and the trials past the number needed. A point resting on fewer than 100 trials past that number says one trial can move it, and a step that gives the second output the same threshold as the step before - the first step, as at the frontier's start - is not read. Where exactly the number needed meet the first output's own point at the level - the frontier's start - one trial sets the second threshold there, so the steps are read above it and the block says so; the other end is named as the second output's own point at the level, its figure rounded toward the stricter side so the bound it states holds as printed.
• One threshold for two numbers. When both outputs are numbers met the same way in the same format, one line says each is counted against its own threshold, and that one threshold for the two together is a question about their total, which SimuXL counts once the total is defined as an output. No pooled figure is printed: SimuXL cannot tell two separate costs from one that contains the other.
• What the harder one moves most with: its top 3 inputs by the Tornado's own rank correlation over the same trials, or each output's when neither is clearly the harder. They are what it moves with, not what causes it, and are left out when the Tornado has nothing to rank.

Input Summary - what each input actually drew

One row per Uncertain Input with its realised average, standard deviation and percentiles against the theoretical values - then one per made input - plus a histogram for the ones you tick. This is where you check that a distribution was sampled the way its formula says - a parameter error is far easier to see than to calculate.

HTML Report - the whole run in a browser, for checking

Writes the run record, the statistics, a distribution chart for every output (unless you switch the charts off) and whichever analyses you selected into one self-contained HTML file - the charts are inside it, so there is nothing to go missing. It lands beside your model, named after it, and opens in any browser. If the workbook has never been saved there is nowhere beside it to go, so the file joins the log under %LOCALAPPDATA%\SimuXL\ - and SimuXL tells you the exact path either way.

This is the one to LOOK at. You cannot edit it and it will not drop into a document you have to hand on, so it is not the file to write when somebody needs a report they can work with - use it to check the run reads the way you expect, then write the Word Report. It opens in any browser without SimuXL or Excel, and the charts are vector, so it prints and zooms without going fuzzy. Nothing in your model changes, and every figure comes from the run already in memory - SimuXL does not recalculate your model to produce it.

Word Report - the same run, for pasting into your own document

SimuXL, Report, Other Reports, Word Report writes it - it sits in that menu because the one-click slot went to the look-check, and it is still the report to hand on. The same report as the HTML one, from the same run, written as a .docx. It is not a picture of a report: the headings are real Word headings and the tables are real Word tables, so you can open it, copy the run record or the percentile table straight into a chapter of your own document, and restyle it to your house style in one action. Named after your model and written beside it, exactly as the HTML report is.

Every chart is in it too, drawn into the document as a picture at print resolution - the same chart the HTML report shows, from the same run, because the picture is rendered from that report's own figure rather than built a second time. Click one and Word will let you resize it or drop it into a different document; it is an ordinary inline picture.

Export Trials - the raw numbers

Every trial's input draws and output values: one row per trial, one column per input and output. This is the data behind every SimuXL chart and statistic, and it is the only SimuXL output meant to be read by a program rather than a person.

• To a new workbook, as two sheets - the record, and a data sheet whose first row is the header and second row is data, so "Save As CSV" works.
• Or straight to a CSV, with a plain-text record written beside it. Both land in your model's own folder and are named after your model.
• Numbers are written at full precision with no formatting, because thousands separators and four decimal places are correct for a person and a lie to a machine.
• A worksheet holds about a million rows and 16,384 columns. If your run has more trials than the one, or more inputs plus outputs than the other, the worksheet option is disabled and says which limit it hit - before you choose it, not after.

Release Data - give the memory back

Release Data is in the Other drop-down of the Results group.

SimuXL keeps a run's per-trial data in memory afterwards: every input's draws, so the analyses can rank them, and every output's values, which are kept whatever else happens. On a large model that is a lot of memory - above a few hundred megabytes SimuXL asks before the run, and above its ceiling it runs WITHOUT retaining the draws and says so, in which case the per-input analyses are unavailable for that run - though the output values are still held. Release Data frees it and reports how much.

15Interpreting results

Which number to quote, which analysis to believe, and what a reader needs in order to check you.

The summary table

Every run opens with one row per output and these columns: Output, Trials, Mean, Std Dev, Minimum, P5, P10, P25, P50, P75, P90, P95, Maximum. They are not equally useful, and which one matters depends on the decision you are making.

Which analysis to believe

Four analyses rank or bound the same model, and they will not always agree. Disagreement between them is information; it is not a sign that one of them is broken.

Making a result defensible

A SimuXL run is reproducible exactly - not approximately - by anyone who has your workbook and three facts: the seed, the sampler and the trial count. All three are on the run record of every artefact SimuXL produces, which is why the run record is the part of a report worth keeping rather than the part worth trimming.

• Type a seed for anything you will compare against or quote. A clock seed is recorded and reproducible after the fact, but a typed one means the run you meant to make is the run you made.
• Say which sampler. Plain Monte Carlo and Latin Hypercube at one seed are two different runs - see Latin Hypercube.
• Keep the HTML report, not a screenshot. It is one self-contained file, it carries the run record, the statistics and the charts, and it opens on a machine that has never had SimuXL installed.
• Note what was frozen. A run with an input frozen is not a full simulation, and every report marks it - but only the report you kept can tell anyone that.

16Preferences and percentiles

Which analyses run, where they go, what the report carries, and what P90 means - stored per workbook, apart from five that follow you instead.

Preferences > Report on the ribbon, and Preferences beside Run Simulation, open the SAME dialog on different tabs. Six tabs. It decides which analyses SimuXL builds for you after a simulation, where their sheets go, which of them the HTML Report carries, and how percentiles are worded.

Simulation - what the Run dialog opens with

The Simulation tab sets the trial count the Run dialog suggests, and which seed it offers - a new one each time, the last run's, or a fixed number you choose. Neither decides anything: the Run dialog still asks, and a run uses whatever you leave in its boxes.

To set them: tick Also save the trial count, seed, sheet destination, summary window and update check to my MACHINE at the foot of the dialog - below the tabs, beside Save - and press Save. They are written to SimuXL_Defaults.txt, beside the log, so a brand new workbook does not start from scratch. The settings line on every analysis sheet names which of the sources it actually read.

The same tab carries Show a summary window when a run finishes and Check for updates once a week when Excel starts, in a box captioned These settings follow ME, not this file because both live only at machine scope - no workbook stores either. The summary window itself offers to stop appearing after every run; its tick is the way back on. A run that was stopped, that failed, or that excluded any trial still shows the window whatever this says, because those are the runs whose outcome you cannot infer from the sheet. The update tick is described under the section named Checking for updates.

Errors - what a run does with something it cannot use

The Errors tab holds the two settings that decide whether a run happens at all: Stop the run on a calculation error, and Refuse to run when a worksheet could not be read. Both are stored in the workbook, because each is a judgement about the model rather than about you - see Rules SimuXL enforces for what each one does.

Analyses - what happens after a run

Tick the analyses you want built automatically, immediately after each simulation finishes: Histogram - the distribution of each output, Tornado, Impact, Tail Drivers, Input Summary. They are the same analyses as the ribbon buttons, over every output, with no picker to answer - so ticking them here saves pressing the buttons after every run. In a workbook whose settings have never been saved, all five are ticked to begin with, so every run also builds all five until you untick the ones you do not want; a workbook whose settings were saved keeps the ticks it was saved with. The one thing this tab is needed for is Impact in the HTML Report, which cannot get there any other way except by pressing Impact on the ribbon afterwards.

The same tab chooses where SimuXL writes its sheets: a sheet in this workbook - the default - replaced on each run (the SXL Fit sheet is the exception: each fit adds its record below the last), so a hundred runs leave one sheet rather than a hundred workbooks; or a new workbook each time, if you would rather each result arrived somewhere it can be thrown away without touching the model it measured. This covers the Trial Run report you get after every simulation, not just the analyses - which matters if you run a model dozens of times in a sitting and would rather not collect dozens of workbooks. SimuXL's own sheets are named so it can recognise and skip them, so putting them here does not confuse a later harvest. A workbook whose structure is protected cannot take a sheet, so SimuXL writes that sheet to a new workbook instead, and says so - in the message after a command, and in the summary after a run when the summary is shown.

SimuXL's own sheets are named SXL - the Histogram is SXL Histogram - so SimuXL can recognise them and skip them when it harvests your model. A sheet you own is never one of the ones replaced.

Report - what the HTML Report contains

The Report tab is a checklist of the sections the HTML Report writes: a distribution chart for each output, the Tornado, Impact, the Tail Drivers, the Input Summary and the listing of every Uncertain Input. Untick the charts for a numbers-only report. The report ends with a short paragraph naming whatever you left out, and why - so a reader can never mistake an omitted section for an analysis that found nothing.

Percentiles - the tails, and Below or Above

The Percentiles tab holds the presentation choices the analysis sheets make: which way round a quoted percentile reads, which percentiles define the lower and upper tails, and how many drivers each tail table shows.

• Every option here is stored in the workbook, not in the add-in, so it travels with the file and two models can disagree. Five exceptions can also follow YOU - the trial count, the seed choice, where analysis sheets go, whether a summary window appears after a run, and whether SimuXL checks for updates once a week - used only by workbooks that have not chosen for themselves. Percentile direction deliberately cannot: it decides what P90 MEANS, and one workbook must not read two ways on two machines. That is the test the whole split applies - a setting may follow you only if it changes how you WORK without changing what any number or any sentence means.
• An option is written only when it differs from what this workbook would otherwise inherit - from your own machine defaults, then from SimuXL's built-ins. So a model that agrees with all of those stores nothing at all, and the dialog says so; and choosing SimuXL's own default AGAINST a machine default of your own is a real choice, so it is written down.
• The Input Summary and the Tail Drivers print which source they read the percentile direction from - the workbook, your machine, or the built-in - on the sheet, beside the numbers. They are the two sheets a percentile setting changes.
• Nothing here changes a sheet that already exists. Run the simulation again, or rebuild an analysis, to see a change applied.

Below or Above: the one setting worth understanding

Below (the default) is the ordinary reading: P90 is the value 90% of trials fall at or below - Excel's PERCENTILE.INC. Above reads the label from the other end: P90 is the value exceeded 90% of the time, which is the ordinary 10th percentile. Same data, opposite wording, and each sheet prints the sentence in full so there is never any doubt.

It never changes your model, and it never changes which trials an analysis looks at. What it changes is which percentile a label like P90 refers to - so a sheet that quotes percentiles will show different numbers under it, because it is quoting different percentiles of the same data. The table below says which sheets those are.

17Where things are

The Help menu, the log, the reports, and what SimuXL keeps inside your workbook.

The Help menu

Everything about SimuXL itself - rather than about your model - is under one button. Help, at the right-hand end of the SimuXL tab, opens a menu holding User Guide (this page), View Log, License, Check for Updates and About. Nothing there changes your workbook.

The log

SimuXL keeps a plain text log of what it did and of anything that went wrong - including the full detail of a failure, which a message box does not show. Help > View Log opens it. Help > About prints its exact path, so you can attach it to a bug report.

It lives under your own user profile: %LOCALAPPDATA%\SimuXL\. Paste that into an Explorer address bar to get there. If SimuXL could not write there it uses your temporary folder instead, so About always prints the path it actually used rather than the one it prefers.

Reports

The Trial Run report and the analysis sheets go where Preferences > Analyses says: each to a sheet of its own in your model, replaced on the next run - the choice for a workbook SimuXL has never saved settings in - or each to a new, unsaved workbook. Either way there is never a "a sheet of that name already exists, overwrite it?" question, and none of your own cells is touched.

Two commands write a real file instead, because a file is the whole point of them. Export Trials asks where to put it; HTML Report puts it beside your model without asking, and tells you the path. Both default to your model's own folder and name the file after your model.

What SimuXL stores in your workbook

One very hidden worksheet, __SimuXL_Handles, holding the freezes and correlations you have created and the settings you have saved in it - your Preferences for this workbook, a Set Default Parameters set, a chart's own settings from Configure this chart - and, while your model sits on an alternative, or a comparison stopped before putting it back, each control variable's base value, which is how it is put back. Names are not there - they are in the cells' own formulas. It is made the first time there is something to keep - a freeze, a correlation, a saved setting, or a comparison or an alternative setting your control variables - and it stays once made. A model where none of those has happened has no such sheet.

SimuXL also keeps hidden workbook names, each pointing at one cell: one for each input you freeze or correlate, and one for each fit you use - SXL_Fit_0001, SXL_Fit_0002 and on, written on its record's Cell name line. Excel moves a name with its cell, which is how a fit's record on SXL Fit still finds its input after you insert rows above it, cut and paste it, or rename its sheet. A model with none of those has no hidden names.

Two sheets you can see hold what SimuXL reads back, rather than a result. SXL Fit holds a record of each fit you used, and every scan reads it to name the inputs fitted - when your sheets go to a new workbook instead (see Reports, above), each record opens one of its own, and those are not read. SXL Alternatives holds the Alternative Table: SimuXL creates it and keeps its Base case column, you set up the alternatives in Alternatives or type them, a comparison reads them, and SimuXL never deletes or replaces the sheet. Every other sheet SimuXL names with SXL is a result - an analysis, a listing or an export - and goes where Reports, above, says, except Input Settings' listing and Export Trials' two sheets, which always open in a new workbook.

18Licensing

The license agreement, a trial on each machine, then a key issued for that machine.

SimuXL runs as a trial for 30 days on each machine it is loaded on, counted from the first time Excel loads it there. After that a simulation needs a license key issued for that machine. Everything else keeps working: you can still open a model, define inputs and outputs, and read the sheets a run already wrote. What a key buys is the trial loop itself - Run Simulation and Run Comparison.

The license agreement

SimuXL is licensed by Simuraytics Software LLC under the SimuXL Software License Agreement, Version 1.2. The first SimuXL button you press on a machine shows it once, and Help > License shows it at any time, with the box I agree to the SimuXL Software License Agreement, Version 1.2. Tick it to agree: Activate works only once it is ticked. The trial runs either way.

SimuXL keeps the version you agreed to and the day, in SimuXL.agreement beside its log, and Help > About carries an Agreement line saying whether its box was ticked on this machine. The agreement is also License.txt beside the add-in and at simuxl.com/License.txt. Third-party notices... on the same dialog shows the licenses of the components SimuXL includes, which are also THIRD-PARTY-NOTICES.txt beside the add-in.

Getting a key

• Press Help > License. The dialog shows where the license stands and this machine's id - six groups of four characters. Press Copy machine id and paste it into your order.
• The key you are sent back was issued for that id and works only on that machine. It starts SXL- and is a few lines long: paste it into the box on the same dialog and press Activate, with the agreement ticked. SimuXL checks it and keeps it beside its log; you do not need to enter it again.
• You may instead be sent a short activation code, starting SXLA- - three groups of four characters. Paste it into the same box and press Activate, with the agreement ticked: SimuXL sends the code, this machine's id and SimuXL's version number to the activation service once, over the internet, and keeps the key that comes back. The code activates one machine - the first one that uses it - and after that the key works offline exactly as a pasted key does. A code that has been cancelled (after a refund, say) activates no machine, this one included, and the dialog says so.
• If you move to a new computer, send the new machine's id and a key is issued for it.

Perpetual or subscription

A perpetual key runs forever, on every SimuXL release dated on or before the date it names - Help > About shows that date beside the licensee. A release after it needs a new key. A subscription key runs every release until its expiry date; in the last 14 days before it, Help > About and the Run Simulation dialog count the days left; for 14 days after that it still runs and says so, and then simulations stop until a renewed key is entered.

If a key is refused

The dialog says which check refused it: not a key means the paste lost its start or end - paste the whole thing again from the message it came in; does not verify means a character changed in transit; a different machine means the key was issued for another machine's id; expired or newer than the license covers name the date that decides it. A refused key is not kept, and an active trial keeps running while you sort it out.

19Checking for updates

One signed file on the release page, read when you ask and quietly once a week.

Press Help > Check for Updates. SimuXL reads one small signed file from its release page and compares the version it names with this build's. The request carries SimuXL's version number and no machine id, licensee, key or workbook name - though, like any web request, it reaches the web host with your IP address. It downloads nothing. The dialog then says one of five things: this machine runs the newest release; a newer release is published, and whether the license on this machine covers it; this build is newer than the published one; the file it read is not a release manifest SimuXL accepts, with the reason; or the release page could not be read, with the reason.

What it sends, and what it does not

One request for the file simuxl-latest.txt on SimuXL's release page (the dialog prints the exact address it read), carrying SimuXL's version number in the way every browser carries its own. No machine id, no licensee, no key and no workbook name leave this machine. Like any web request, it reaches the web host with your IP address. The file it reads is signed with the same key that issues licenses, and a file whose signature does not verify is refused rather than believed - so nothing on the internet can make SimuXL announce a release that was not published.

SimuXL never downloads or installs anything. When a newer release exists, Open release page opens the release page in your browser; you download SimuXL-Setup.exe from there and run it with Excel closed, as you did this one.

The quiet check once a week

SimuXL also makes the same check on its own, once a week at most, about 30 seconds after Excel starts. It shows nothing at start-up - no window, no message - and sends exactly what the button sends: the one request above, and nothing more. What it found is on the Updates line of Help > About; pressing the button checks again. A check you make yourself counts, so pressing the button today means no quiet check for 7 days; and if the release page could not be reached, SimuXL waits the same 7 days before trying again on its own.

To turn it off: under Preferences > Simulation, untick Check for updates once a week when Excel starts (it sits in the box captioned These settings follow ME, not this file, because it belongs to your machine and no workbook carries it). The tick is recorded the moment it changes, like the machine tick at the foot of the dialog, so Save is not needed for it. With it off, nothing is read at start-up at all; the button still works whenever you press it.

Updates and your license

A subscription key runs every release until it expires, so a newer release is always yours to install. A perpetual key runs releases dated on or before the date it names (Help > About shows it): if the newer release is dated after that, the dialog says so and names the date, and installing it would stop simulations on this machine until a key covering it is entered. Help > About carries one line saying what the last check concluded and when.

20A signed add-in, and the one-time trust step

Every SimuXL file is signed by Simuraytics Software LLC; one step trusts that signature on your machine - the installer offers to run it for you - and the add-in then loads without being blocked.

Both SimuXL64.xll and SimuXL32.xll carry a digital signature by Simuraytics Software LLC, SimuXL's licensor. The certificate behind it is SimuXL's own rather than one bought from a certificate authority, which is why your machine does not trust it until you tell it to - and until then Windows treats a copy downloaded from the internet as untrusted and Excel refuses to load it.

The one-time step

The installer runs this step for you. It is a ticked option on the installer's tasks page, before anything is installed, and can be unticked there. To run it yourself - you unticked it, or this machine's certificate store was reset - open the folder SimuXL is installed in (%LOCALAPPDATA%\Programs\SimuXL unless you chose another), where SimuXL-Publisher.cer and SimuXL-Trust.bat sit beside the add-in.

1. Double-click SimuXL-Trust.bat. It adds SimuXL-Publisher.cer to your own Trusted Root Certification Authorities and Trusted Publishers stores, and takes out any earlier SimuXL certificate it replaces - no administrator is needed, nothing else on the machine changes, and every SimuXL file signed with that certificate is trusted from then on, including a copy Windows marked as downloaded from the internet. Windows asks whether to install a certificate claiming to represent Simuraytics Software LLC; answer Yes. On a machine that trusted SimuXL's certificate before 4.14.0 - issued in the developer's own name - Windows first asks whether to delete that earlier certificate; answer Yes to that too.
2. The window then reads the signature on each SimuXL file beside it and prints Valid for each - that is the pass. Close it and load the add-in as usual.

After that, every SimuXL file signed with the same certificate is trusted on that machine - this release and the next - and no Unblock step is needed for a downloaded copy. 4.14.0 issued SimuXL's certificate again, in the name of Simuraytics Software LLC and with the same key; the step - the installer offers it - takes the earlier certificate out, so only the current one stays trusted, and it does the same whenever the certificate is issued again. A machine whose administrator locks the certificate stores may need the administrator to run the step instead.

What About says

Help > About carries a Signature line reading what the running file's signature says on this machine: Signature: signed by Simuraytics Software LLC, trusted on this machine. once the step has been run, and before it a line ending run SimuXL-Trust.bat once. naming what is still missing. A file signed by anyone but Simuraytics Software LLC says so on the same line; it is not a file SimuXL released.

21The installer

One file, SimuXL-Setup.exe from simuxl.com, installs the right add-in for your Excel, registers it, and runs the trust step - no administrator needed.

The Download button on simuxl.com gives you SimuXL-Setup.exe. Close Excel, run it, and it does these five things:

1. Reads which Excel you have - 64-bit or 32-bit - from Excel's own program file, and shows the answer on a page you can correct. Only the matching add-in is installed.
2. Installs into your own Programs folder (%LOCALAPPDATA%\Programs\SimuXL), with SimuXL-Publisher.cer, SimuXL-Trust.bat, this guide, the license agreement (License.txt) and the third-party notices (THIRD-PARTY-NOTICES.txt) beside the add-in. No administrator is asked for and nothing outside your own profile changes.
3. Registers the add-in with Excel the way Excel's own Add-ins dialog does, so the SimuXL tab is on the ribbon the next time Excel starts - there is no Browse step.
4. Trusts that one folder in Excel - just the folder it installed into, not its subfolders - so Excel loads the add-in even if your Trust Center is set to block add-in files (File Block). No other Trust Center setting is changed.
5. Runs the trust step (ticked by default). Windows asks whether to install a certificate claiming to represent Simuraytics Software LLC; answer Yes. On a machine that trusted SimuXL's certificate before 4.14.0 - issued in the developer's own name - Windows first asks whether to delete that earlier certificate; answer Yes to that too. The section above says what that step does and how Help > About shows it.

The installer is signed with the same certificate as the add-in. Windows SmartScreen may still say it protected your PC from an app it does not recognise: SmartScreen judges a file by how many machines have already run it, not by its certificate, so a fresh release can meet that screen even on a machine that trusts the publisher. Press More info: the publisher reads Simuraytics Software LLC on a machine that already trusts SimuXL's certificate, and may read Unknown publisher on one that does not yet - a new machine, or one that trusted only the certificate before 4.14.0 - because the certificate is SimuXL's own and not one bought from a certificate authority. Either way, press Run anyway.

Updating and removing

A newer release's installer installs over the older one: close Excel first (the installer asks if Excel still has the add-in open), run it, and start Excel. Uninstall is under Windows' Installed apps as SimuXL. Close Excel first: if Excel still has the add-in open - even with no window on screen - Uninstall says so and changes nothing, so nothing is half-removed.

Uninstall removes SimuXL completely, and that includes copies this installer did not put there. It removes the files it installed, the one trusted location it added, and every SimuXL add-in entry in Excel - not only its own - so Excel is left with no SimuXL at all. If any of those entries points at a SimuXL .xll somewhere else on your computer - a copy you registered by hand, or an older version - Uninstall lists those files and asks whether to delete them too; answer No to keep the files, and the add-in entries are removed either way. Nothing that is not a SimuXL .xll named by one of those entries is ever touched.

What Uninstall never removes: your settings, your licence file, your evaluation record, your log and the trusted certificate. They stay exactly where they are, so reinstalling later finds your licence still in place - and so that removing and reinstalling SimuXL does not restart a free evaluation period.

22The right-click menu, the keyboard, and SimuXL from your own macros and formulas

The ways in: the SimuXL tab, a SimuXL entry on the cell right-click menu, a typed Application.Run call for the commands with no button, and - for a run - a call your own macro can make and read the answer back from. And in the other direction, three worksheet functions answer questions about SimuXL itself: which build is installed, and what the last run measured. Ctrl+Shift keyboard chords are disabled in this build.

Right-click a cell

Right-click any cell and there is a SimuXL entry on the menu, carrying the commands that are about the cell you are standing on: Define Input, Make Input, Define Output, Erase, Name, Freeze / Release and Run Simulation. The Define Input as entry opens the same list of 8 distributions as the ribbon menu, in the same order.

Keyboard chords are disabled in this build

This build registers no Ctrl+Shift keyboard chords. (The keys do whatever Excel itself assigns them.) They are switched off while a crash tied to Excel's keyboard-macro dispatch is investigated: the same command that runs safely from its ribbon button could bring down Excel when reached through a chord. Every command is on the SimuXL tab or the right-click menu, apart from the two below. The chords will return once the cause is fixed.

Convert and Revert have no button. Convert turns inputs defined by the older VBA SimuXL's dialog into formulas, keeping what each cell held in a very hidden sheet, __SimuXL_Conversion, and Revert puts them back from it: press Alt+F11, then Ctrl+G, type Application.Run "SimuXL_Convert" or Application.Run "SimuXL_RevertConversion" and press Enter.

Running a simulation from your own macro

SXL.RunSimulationArray runs a simulation and hands the run record back to your code, with no Run dialog and no summary window to dismiss. It takes the trial count, and optionally a seed, whether to use Latin Hypercube, and whether to try scoped recalculation:

answer = Application.Run("SXL.RunSimulationArray", 10000, 12345)

• What comes back is the same table the SXL Trial Run sheet shows - the outcome, the trial counts, the seed and where it came from, the elapsed time, every output's statistics and the run's own notes. It arrives as a two-dimensional array you can index, or write onto a sheet of your own.
• Written onto your own sheet it carries the values and not the number formats. SimuXL applies those to the SXL Trial Run sheet as a separate step, so a mean that is a DATE reads as an Excel serial such as 45964.2115 in a cell you have not formatted. Format the cells, or read the figure off the sheet SimuXL wrote.
• If SimuXL would not run, a single piece of text comes back instead, beginning SimuXL:. Test it with If VarType(answer) = vbString Then. The words are the same ones the Run command would have shown you in a box.
• Trials must be a whole number, 1 to 1,000,000. Leave the seed out and SimuXL takes one from the clock and records that it did, exactly as a blank seed box does.
• Which analyses get built, where their sheets go and what to do about a value that will not read as a number are your workbook's own settings. Set them once in Preferences. The one preference a call from code does NOT use is the seed policy: omitting the Seed argument always takes one from the clock, because nothing was offered for you to accept.
• Notes about the ANALYSES are not in what comes back. Which ones were built, which were skipped and why is written to Help > View Log - the run record is written before the analyses run, so it can never carry one.
• It is not a worksheet function. Typed into a cell it refuses, because a formula being recalculated cannot run a trial loop that recalculates the same workbook.

Asking which SimuXL is installed

=SXL.Version() in any cell answers with the build you are running - version, artefact and bitness. It is the first line of Help > About, which asks you to quote every line it prints in a bug report; which Excel you are running is another of them, described below.

• If it resolves at all, SimuXL is loaded - which is the health check a script wants, because the alternative to a build name is #NAME? rather than a wrong answer. ⚠️ A #NAME? does not prove the opposite on its own: a mistyped function name gives one on a perfectly healthy install, so check the spelling before you check the add-in.
• It answers about SimuXL and not about Excel. Which Excel you are running is a separate fact with a separate way of going missing, and it is printed on the SXL Uncertain Inputs sheet, on the run record and in Help > About, where an unanswered question can be told apart from an answer.
• It is recomputed rather than frozen, so a workbook saved under one build and reopened under another names the build you have now instead of the one it was saved with. ⚠️ You cannot swap the add-in file while Excel is open and expect this to follow: Excel holds the file open - see "A newly installed build seems broken" - so a new build means closing Excel and reopening it.

Reading a run's figures into your own cells

=SXL.Result("Total Cost", "P90") puts one figure from the last simulation into a cell of your own - so a summary block, a dashboard or a printed page can be built out of formulas instead of copied numbers, and it is right again after the next run without anybody remembering to update it.

• The first argument is the output's NAME, exactly as the SXL Trial Run sheet lists it. Ask for a name that run did not measure and SimuXL tells you which names it did.
• The second is one of: Mean, SD, Min, Max, Trials, Median, P0, P5, P10, P20, P25, P30, P40, P50, P60, P70, P75, P80, P90, P95, P100. Anything else is refused and the message lists them.
• P90 here always means the value 90% of trials fall at or below - Excel's PERCENTILE.INC - whatever Preferences > Percentiles is set to. A cell holds a number and no caption, so the reading is fixed rather than following a setting a colleague might change; it is the same reading the SXL Trial Run sheet's own ladder uses, and that sheet says so.
• It reports the last COMPLETED run in this Excel session, and only for the workbook the formula is in. The last run is remembered once per Excel, not once per workbook, so a formula in a different workbook is refused rather than answered - it would be correctly formatted and entirely wrong.
• It refreshes itself. Nothing about the formula changes when you run again, so SimuXL recalculates these cells on every recalculation rather than letting them go quietly out of date.
• The figures live in this Excel session and not in the file. Close Excel and reopen the workbook, press Release Data, or run a comparison, and every one of these cells shows a sentence instead of a number until the next simulation. If a page has to survive being emailed, paste its values or send the SXL Trial Run sheet.
• A run that did not finish is refused rather than quoted. A simulation you stopped part-way has real figures over fewer trials than you asked for, and every sheet that shows them shows that caveat beside them - a cell cannot, so it says so instead and sends you to the SXL Trial Run sheet.
• When SimuXL cannot answer you get text beginning SimuXL: rather than an error, and the sentence says what to do; =ISTEXT(...) is the test. ⚠️ An Excel error in either argument is a different case and comes straight back unchanged, so an =IFERROR(...) you have wrapped around it still catches it.
• The figure arrives as a plain number with no number format. A mean that is a DATE reads as an Excel serial such as 45964.2115 until you format the cell - the same thing the run record's array does, and for the same reason.

Reading the run in progress into your own cells

=SXL.Live("Total Cost", "P90") answers from the simulation that is running RIGHT NOW, over the trials finished so far. It takes the same two arguments as SXL.Result and answers from the same list of statistics.

• ⚠️ Whether you SEE it climb depends on one setting. On an ordinary run Excel's screen updating is off, so the cell's figure is up to date the whole time and the screen does not show it - click anywhere on the sheet and the current number appears. To watch it move on its own, tick Watch the run in the Run dialog and give it a hold of at least 1 ms, which leaves the screen painting for the whole run. Measured on Excel 16.0, build 20228: every watched run's cell climbed unaided and every ordinary run's did not.
• It is a different function from SXL.Result on purpose, and the difference is the tense. SXL.Result is the last COMPLETED run; this is the one in progress. A single function that quietly switched between them would leave a cell meaning two things at two moments with nothing on the sheet to say which - and a cell has room for a number and no caption at all.
• When nothing is running it says so and points you at SXL.Result. That is the state a workbook is in almost all of the time, so the sentence is written to be useful rather than apologetic: the figure you want when no run is going is the last completed run's, and that is the same formula with one word changed.
• Ask for Trials beside anything else. Every figure here describes the trials finished so far, and a P90 over the first eleven trials is a real number that is an estimate of nothing. Trials is the denominator that tells you how much to believe the cell next to it.
• The figures refresh on a clock - every 250 ms - rather than on every trial. Keeping a running percentile up to date on every single trial can cost more than the simulation itself: measured at 20.7 seconds of extra work on one 200,000-trial run, against 0.6 seconds for the same figures kept up to date every 250 ms. Nothing is missed - each refresh takes in every trial since the last one - and on a run long enough to be worth watching the difference is invisible.
• Collecting starts when the first live cell asks, and costs nothing until then. A run with no live cell in front of it allocates nothing and does no extra work, so leaving this feature unused is free. The first evaluation in a run may show a short "started collecting" sentence; the next evaluation has figures.
• When the run ends, every live cell goes back to its no-run sentence. It does not freeze on the last figure it showed, because a stale live number and a finished one look identical and only one of them is trustworthy. Read the finished figures with SXL.Result or on the SXL Trial Run sheet.
• A cell that both reads a run in progress and draws from a distribution is refused before the run starts, by name. Such a formula takes one branch while SimuXL is counting its calls and another while they report, which stops the run for a reason that is not the real one. Put the readout in a cell of its own and refer to that cell.

23When something looks wrong

Most of what follows is SimuXL working correctly and being surprising. The first two are not.

Every SimuXL formula shows #NAME?

The add-in is not loaded. Excel does not know the function exists. Check File > Options > Add-ins > Manage: Excel Add-ins and confirm SimuXL is ticked. This is also what a colleague sees when you email them a model and they do not have SimuXL - the numbers you last calculated are still in the file, but the formulas cannot recalculate.

A cell shows "SimuXL: something is wrong with an argument"

That is not an error, it is SimuXL telling you which argument it could not use - for example SimuXL: MostLikely must lie between Minimum and Maximum., or a standard deviation of zero, or a Discrete whose probabilities do not sum to 1. Read the message and fix that argument. A cell full of #VALUE! would tell you none of that, which is why SimuXL does not do it.

Every SimuXL formula shows #VALUE!

The opposite diagnosis to #NAME?, and a much better one: the add-in did load and something inside it threw. Since a bad argument gives you the message above instead, this points at SimuXL rather than at your formula - press Help > View Log, which will have the detail, and quote the version from Help > About.

A newly installed build seems broken

A run is refused before it starts

SimuXL refuses for a small number of reasons and it names the cell every time. All of them, with the remedy for each, are under Rules SimuXL enforces - and the commonest by a distance is a conditional that chooses between SimuXL calls rather than between values.

A rare risk ranks last on the tornado

This is the most important thing in this guide, and it is arithmetic behaving exactly as designed.

Take =SXL.YesNo(0.05) * 100 + SXL.Triangular(10, 20, 50). The 5% gate produces the entire upper tail of that output and owns about 87% of its true variance. The tornado scores it around 15% and puts the triangle first - a near-perfect inversion.

The reason is that rank correlation cannot see a rare event clearly. An input that takes the same value in 95% of trials has 95% of its ranks tied, and that caps the correlation it can reach with anything at about 0.38, however much of the outcome it actually drives. The tornado's own Highest |ρ| possible column tells you when you are looking at that limit - and it is measured against the OUTPUT, so it is not the same figure as the run report's Ceiling (measured), which is measured against the partner INPUT.

A correlation reports far lower than the one I asked for

The same ceiling, in the same costume. If one end of the pair is a rare Yes-No, no sampling method can reach a high rank correlation with it. The limit has a closed form - SQRT(3p(1-p)) for a Yes-No firing with probability p - which gives 50% -> 0.87, 25% -> 0.75, 10% -> 0.52, 5% -> 0.38, 1% -> 0.17. A k-outcome Discrete Uniform caps at SQRT(1 - 1/k^2): 4 outcomes -> 0.97, 6 outcomes -> 0.99, which is why a die is never the reason. The correlation report fills in its Ceiling (measured) column when that is what has happened, and the figure it prints is measured from your run's own draws - the realised firing frequency rather than the probability you typed - so expect it to differ from the formula in the third decimal.

The tornado and Impact rank my model differently

They are supposed to be able to. Tornado asks how consistently an output follows an input and ignores size; Impact asks how far one input can move it and ignores likelihood. A high-probability nuisance leads the tornado; a low-probability catastrophe leads Impact. Each sheet names the other and says what it answers.

Two sheets quote different values for the same percentile

Check the Below / Above setting, and the section on percentiles above. Some sheets follow that convention and some deliberately do not, and each sheet states its own.

A tail driver table lists inputs I know are irrelevant

Look for the ~ marks and the Noise floor (SD) column beside them. With many inputs and few trials, an input with no connection to the output still shows a shift by pure chance - and a top-five table is exactly where that shows up. Each row carries in the Noise floor (SD) column the floor it had to clear. That is the same figure on every row unless an input is extremely lopsided, because only a very skewed input reaches further by chance in the direction it leans - so a column reading the same on every row is the ordinary case and not a fault. At 2,000 trials over 1,000 symmetric inputs the floor lands near 0.22 standard deviations, and it falls as the trial count rises. More trials is the cure.

A run's answer changed when I changed nothing

The two runs used different seeds - or the same seed under different samplers - so they drew different random numbers and differ by sampling noise. Either the seed box was left empty - SimuXL then takes one from the clock, which is different every time - or a different number was typed. The run report prints the seed either way, so open the report for the run you liked, set the sampler it names and type its seed back in to get that run again exactly.

Nothing here explains it

Press Help > View Log. The log records what SimuXL did and the full text of anything that failed, including the detail a message box does not show. Help > About gives the exact version to quote. Both are what a useful bug report is made of.

24Glossary

Every term this guide uses in a particular way, in one place.

Uncertain Input

ONE distribution: the thing that gets sampled, that appears in the tornado, and that you can freeze or correlate. A cell can hold more than one.

SimuXL formula cell

A cell whose formula contains one or more Uncertain Inputs. =SXL.YesNo(0.05) * SXL.Triangular(10, 20, 50) is one cell and two Uncertain Inputs.

Call

One SimuXL function call inside one cell's formula. =SXL.YesNo(0.05) * SXL.Triangular(10, 20, 50) is one cell and two calls, so it appears as two rows on every listing that numbers them. The number is the order the calls FIRE when Excel calculates the cell - which for an ordinary left-to-right formula is the order they appear in it, and is not the same thing when a call is nested inside another's argument. That is why editing a formula can change which call is which, and why SimuXL checks a cell's saved Freeze and Correlate settings against it before applying them. The column is Call (from 1), on the Find Inputs listing, on the SXL Trial Run sheet and in both run reports. On a model where each input sits in a cell of its own, every row of that column reads 1 - which is what it should read, not a sign that the column is broken.

Counting from 0 or from 1

Every column SimuXL numbers with a #, and the two call-ordinal columns, state in their own heading where the counting starts, because they do not all start in the same place and a reader comparing two sheets needs to be told. Sample # (from 0) is a POSITION IN AN ARRAY rather than an ordinal - it is the slot in the seeded random stream that input draws from, so the number on the sheet is the number the sampler uses. On the Find Inputs listing, an input SimuXL never assigned a slot shows a dash there rather than a number, because there is none to print - and a zero could not say so, since sample 0 is a real slot. Ordinal (from 0), on the handle listing, is the ordinal SimuXL hands out internally, written unchanged - so the same call is numbered one LOWER there than in the Call (from 1) column on the Find Inputs listing, which writes that ordinal plus one. That disagreement is real and deliberate, and both headings say so. Call (from 1), Output # (from 1), Control # (from 1) and Made input # (from 1) are ORDINALS - the first, the second - and every other surface that numbers them starts at one, so starting a listing at zero would trade a disagreement inside one sheet for a disagreement across the product. No figure changes because a heading says this; only the claim the heading makes.

Output

A cell whose distribution you want measured. Defined once, by wrapping the cell's own formula in SXL.Output, so the definition travels with the cell.

Trial

One pass: draw a value for every Uncertain Input, recalculate your whole workbook, record every output.

Seed

The number that determines every draw in a run. The same seed under the same sampler reproduces a run exactly, on any machine.

Sampler

How the draws are laid out - plain Monte Carlo or Latin Hypercube. A seed alone does not identify a run; the sampler is the other half.

Latin Hypercube

A sampler that splits each distribution into equal slices and takes exactly one draw from each, so the draws cover the distribution evenly instead of clustering by luck.

Design

One complete, independent Latin Hypercube layout. A run is made of several, which is what lets Convergence put a real band around a percentile.

Harvest

The single recalculation SimuXL does before a run to find out what your model contains. Find Inputs is that same scan, reported, and run whenever you ask for it.

Freeze

Holding one input at its average for a whole run, to see what the answer looks like without it. The formula is not touched and no other input's draws change.

Ceiling

The highest rank correlation a pair can reach when one of them takes only a few distinct values - it is a property of the pair, not of either input, and two coarse inputs cap each other harder than either caps a continuous partner. A Yes-No at 5% tops out near 0.38 however strongly you correlate it. No column is headed with this word on its own, because three surfaces show it and they are measuring three different things. Ceiling (from p) in the Correlation Editor is computed from the probability you DECLARED, before any run exists. Ceiling (measured) on the run report is measured from the run's own draws, against the PARTNER INPUT. The tornado does not use this word at all, because it is not measuring the same thing: its Highest |ρ| possible column is the largest |rho| an input could reach against the OUTPUT, which is a different pair entirely - which is why that figure differs from the other two by more than rounding.

Percentile direction

Whether P90 means the value 90% of trials fall at or below (Below, the default) or the value exceeded 90% of the time (Above). Stored per workbook.

Base value

A deterministic reference figure drawn as a dashed rule on one output's histogram, in that output's own units. You supply it, or nominate a cell for it.

Noise floor

How large a shift an input entirely unrelated to the output would show anyway, by chance, in a tail table of this size. Rows marked ~ are inside it.

Parameters set

How an input's boxes are read: as the distribution's own parameters, or as values at percentiles - 10%, 50%, 90%, say. A set other than the distribution's own is kept in the formula, after the name - =SXL.Triangular(80, 100, 130,, "P10,P50,P90") - and SimuXL works the distribution's own parameters out from it every time the cell calculates.

Fit record

The block added to the SXL Fit sheet when you press Use this distribution in Fit… and then OK in the Input dialog: the input, what it was fitted to, when, how, and every family's result. The newest record for a cell is what the Fitted column of Find Inputs names.

Joint outcome

One of the four ways a trial can fall against two outputs' thresholds - both met, one met and the other missed (either way round), or both missed. Joint Outcome counts the trials of each from the two values each trial produced together, never by multiplying the two separate shares.

Run record

The block of facts identifying a run - date, trials, seed, sampler, status, elapsed time and the outputs measured. It is on every artefact SimuXL writes.