Improved File Sharing And Flexible Distributions [Newsletter]

This is the first email from the Carlo newsletter. It’s possible you signed up a while ago and don’t remember what Carlo is. Oof! As a reminder: Carlo is a subscription service that lets you do calculations with uncertainty in Google Sheets (and Excel Online).

There are two recent improvements to the product I’m excited about:

SAFER SPREADSHEET SHARING

Carlo needs read access to your spreadsheet in order to work. When Carlo started, spreadsheets had to be shared to “anyone with the link can view”.

Now, you instead share your spreadsheets directly with a special Carlo email address that our servers use to fetch your changes. This address can only be used by our servers, and is not a normal Google account that anyone could log into (not even me).

To try it out, add a new spreadsheet to Carlo (by pasting its URL). Sharing instructions will be shown.

THE HOT NEW PROBABILITY DISTRIBUTION EVERYBODY IS TALKING ABOUT

If you’ve done uncertainty modelling, you’re probably familiar with this approach to probability distributions:

  • You must choose a distribution from a long menu of options (e.g. uniform, normal, …)
  • Yet each of these distributions imposes strong constraints. For example, a two-parameter distribution is unable to match three data points like the 10th, 50th, and 90th percentiles.

You often end up with a distribution that doesn’t faithfully represent your knowledge.

Instead, I’ve been advocating for a new approach: “flexible” distributions. Here’s how they work:

  • Give as many percentiles as you like, and get a distribution that matches those points exactly
  • You can make the distribution unbounded, or give it an upper or lower bound, or both

I have been thinking about this topic for a few years, which led me to design my own flexible distribution. It is now offered on Carlo exclusively.

Here’s an example: https://carlo.app/spr/NDQvZSKERQpfkTMAhdmM5F/

One experienced modeller and Carlo user has described flexible distributions as “game-changer” for uncertainty analysis.

Usage statistics show it’s the most used distribution on Carlo by a wide margin. This makes sense: it can be used for anything, and sidesteps the analysis paralysis that can come from choosing the “right” probability distribution.

Many people write models that use the flexible distribution for all uncertain inputs.

(We previously offered the Metalog distribution. It was more flexible than traditional distributions, but it could only match some combinations of percentiles. If you’ve encountered the dreaded “Error creating distribution METALOG…”, you know what I mean. Our new flexible distribution always works, for arbitrary percentiles.)

PS: Don’t care about this new stuff? You might still love the core functionality we’ve had since day 1: fast simulation, real-time collaboration, and 100% backward compatibility with Sheets or Excel.