So You Want to Eat M&Ms Like Christine

So you want to eat M&Ms like Christine.

Excellent.

First, obtain some plain milk-chocolate M&Ms.

Not peanut M&Ms. I’m allergic to peanuts, so if your goal is to eat M&Ms like me, those are disqualified before the experiment begins. Same goes for the peanut butter variety. Might as well exclude all M&M varieties that require a flavor qualifier to distinguish them from plain.

And not a custom color mix or one of the approximately fourteen seasonal color-mix varieties that appear to exist exclusively to complicate grocery-store shelves, either.

For purposes of this theory, that leaves plain, standard, milk-chocolate M&Ms only. Different fillings, sizes, shapes, or color assortments introduce variables I have not tested and will not pretend to understand.

I did not derive these rules by writing down what I thought I did and declaring the matter settled. I tested them instead, and changed the theory when the predictions were wrong.1

The standard six-color population is:

🟤 brown | 🔴 red | 🟠 orange | 🟡 yellow | 🟢 green | 🔵 blue

Sort them by color.

Yes, all of them.

If your population is well behaved, the rest is easy. I normally eat exactly two M&Ms at a time, preferably in same-color pairs, in this order:

🟤 → 🔴 → 🟠 → 🟡 → 🟢 → 🔵

A quick notation note: when I show a population, the number after each color is the number of M&Ms of that color. In an eating sequence, a color by itself represents that color’s remaining run, eaten in same-color pairs unless otherwise specified. A slash means those colors are eaten together in the same bite.

Brown toward the beginning. Blue toward the end.

That is the baseline.

There is one more thing you need to know before proceeding: an intact even run is indivisible.

If you have six 🟢 M&Ms, you have three stable green pairs. You do not remove one green to solve a parity problem somewhere else.

I’m not saying you can’t.

I’m saying you’d be wrong.

At this point, you are ready to encounter an actual handful of M&Ms, which will almost certainly refuse to cooperate.

Hard Constraints

Before this gets more complicated, one distinction matters: preference versus constraint.

The canonical order is a preference.

Some things are law.

Two mixed-color combinations are prohibited within the same bite:

🟤 + 🔵: never

🔴 + 🟢: never

This does not mean those colors can never appear next to each other in the overall eating order. It means I will not eat them together in the same bite.

Other mixed-color bites range from excellent to merely tolerable.

Blue has additional protections.

🔵 is the only color I will eat as a same-color triple.

🔵 is also never individually disposed of.

I did not begin this research knowing that was a hard rule. I discovered it when a predicted solution required getting rid of a single blue M&M and my immediate reaction was: absolutely not.

Subsequent cases held.

A blue M&M may remain uneaten if I reject the entire population.

It may not be singled out as the problem.

The distinction is meaningful.

Resolving Odd Populations

Eventually, one of your colors will have an odd count.

Do not panic.

More importantly, do not immediately discard the extra M&M.

An odd run can often be repaired with a mixed-color boundary pair.

Three 🔴 M&Ms, for example, do not necessarily mean one red must go. They may become one red pair plus one red paired with a compatible color.

Some mixed bites are especially useful:

🟤 + 🟠: great

🟡 + 🟢: great

🟠 + 🔵: great

🟢 + 🔵: great

Others are fine.

Some are limited.

Two remain crimes.

The quality of the mixed bite matters because merely making every color even is not enough.

The goal is not just to make the numbers work. The whole eating sequence still has to feel right.

Consider:

🟤2 | 🔴2 | 🟠7 | 🟡1 | 🟢5 | 🔵0

The obvious visual anomaly is the single 🟡.

I initially predicted that I would dispose of it.

Incorrect.

The yellow pairs beautifully with the extra green. The extra orange is the expendable piece.

The resulting sequence is:

🟤 → 🔴 → 🟠 → 🟡/🟢 → 🟢

with one 🟠 removed.

This is why you cannot simply look for the loneliest M&M and declare it guilty.

You have to consider the population as a whole.

When the Canonical Order Moves

The baseline order is:

🟤 → 🔴 → 🟠 → 🟡 → 🟢 → 🔵

It is useful.

It is not absolute.

Once the population gets complicated, the canonical order stops being enough. Now the colors have to negotiate with each other.

One force is numerical weight.

A very large run can pull toward an endpoint or change which color feels right next to it.

In one test I had:

🟤11 | 🔴4 | 🟠2 | 🟡5 | 🟢4 | 🔵4

The correct cleanup was to remove one 🟤 and one 🟡, leaving even runs.

The resulting order began:

🟤 → 🟠 → 🔴

The ten remaining brown M&Ms created enough weight that orange felt better directly beside them.

Canonical red-before-orange lost.

Missing colors matter too.

If a color is missing, it does not retain an honorary seat in the sequence.

There is no conceptual 🔴 between 🟤 and 🟠 if there are no red M&Ms.

The remaining colors become actual neighbors.

That can change both compatibility and position.

In one real-world case, there were no brown, red, or blue M&Ms:

🟠7 | 🟡2 | 🟢5

The order was:

🟡 → 🟢 → 🟢/🟠 → 🟠

The heavy orange run pulled toward the exposed terminal end.

That suggested that when one of the usual endpoint colors is missing, another color may start behaving like the endpoint instead.

A later test showed that this does not happen automatically.

So the current finding is narrower:

Endpoint inheritance has been observed.

Its governing conditions remain under investigation.

Mixed bites can pull on the sequence too.

A mixed pair is not always just a local parity fix. Sometimes it affects where the rest of its color belongs.

Take:

🟤6 | 🔴2 | 🟠5 | 🟡4 | 🟢0 | 🔵1

The single 🔵 cannot be disposed of.

🟠 + 🔵 is an excellent pairing, so the sequence ends with an orange/blue bite.

But the remaining orange pairs do not stay ahead of yellow.

Instead:

🟤 → 🔴 → 🟡 → 🟠 → 🟠/🔵

The orange run follows its mixed unit toward the end.

I think of this as structural gravity.

This is not an official mathematical term.

Yet.

This is why I think of the canonical order as precedent rather than statute.

It governs until the facts of the case give me a good enough reason not to follow it.

A Brief Note on Triples

The normal unit is two.

Three requires justification.

There are two established exceptions.

First, 🔵 may be eaten as a same-color triple.

No other color receives this privilege.

Second, three different colors may occasionally be eaten together when doing so resolves multiple anomalies at once.

A heterogeneous triple is not something I do for variety.

It is a normalization mechanism.

Once you permit three M&Ms without adequate controls, the entire enterprise deteriorates rapidly.

Disposal

Disposal is always on the table.

What that looks like depends on where the M&Ms came from.

If you took a random handful from a larger container, you can return a piece.

If the population came from a closed package, you might give an unwanted piece to someone else.

Or you might literally throw it away.

Returning one to the bag does not feel the same as throwing one away. But either way, that M&M is no longer part of the problem.

Eating every M&M is not automatically worth making the whole sequence worse.

Removing one piece is unremarkable.

Two can be fine.

Three has been demonstrated to be acceptable.

Four is where the system starts objecting.

In multiple tests, a clean solution that required removing four individual pieces pushed me toward a different strategy. Either I found a more complicated way to eat the population intact, or I rejected the handful and started over.

Apparently, three is cleanup.

Four is a referendum on the entire population.

I did not set that threshold in advance.

The M&Ms did.

The most extreme available remedy is to reject the population itself.

That is not psychologically equivalent to removing several individual M&Ms.

I know how that sounds mathematically.

Nevertheless.

Whole-population rejection has happened often enough that I know it belongs in the system.

I am not comfortable pretending I have a complete rule for when it should happen.

Sometimes there are too many local corrections. Sometimes the population is just small and awkward enough that I stop negotiating with it.

I don’t think there is a clean rule beyond that yet.

You look at the population.

You consider the available remedies.

And sometimes the correct conclusion is simply:

No.

Further study is warranted.

The Model Can Also Correct Me

There is one more complication.

Sometimes I can tell you what I historically did and still conclude that there was a better solution.

A real handful contained:

🟤7 | 🔴5 | 🟠2 | 🟡4 | 🟢3 | 🔵1

At the time, I disposed of one 🟤 and one 🔴, then ate the remaining runs.

When I later tested the population, the proposed solution was:

🟤 → 🟤/🔴 → 🔴 → 🟠 → 🟡 → 🟢 → 🟢/🔵

Every M&M could be eaten.

And I preferred that solution.

To be clear, that does not make the prediction historically correct.

It is something stranger.

The model I built from my own rules had found a more Christine solution than Past Christine had.

I have apparently developed a system for identifying errors in my own prior M&M jurisprudence.

Good for science.

Less good for Past Christine.

At that point, calling this a decision tree gets hard to defend.

The Part Where This Stops Being a Decision Tree

At the beginning of this project, I assumed there was a sequence of instructions somewhere in my head. Something like:

If this, do that.

If odd, pair this.

If singleton, remove that.

That is not what I found.

Instead, every handful gives me a bunch of things to balance at once. Some rules are absolute. Some are preferences.

I’m balancing which colors can share a bite, how much of each color there is, where the heavy runs want to sit, and whether I can keep a run together.

I also care about what a move leaves behind, how many M&Ms I have to sacrifice, and whether the whole solution feels unnecessarily convoluted.

And sometimes there is more than one valid answer.

What I actually seem to have is a constrained optimization problem with a few non-negotiable rules.

In practice, I am doing all of those calculations against each other until something resolves into a solution that feels right.

Most of this happens almost instantly when I am actually holding M&Ms.

Writing it down took considerably longer.

So You Still Want to Eat M&Ms Like Christine

Fine.

Here is the closest thing I can give you to an algorithm:

Sort the M&Ms. Leave intact even runs intact.

Find the odd runs and true singletons.

Apply the hard constraints.

Look for good mixed-bite repairs. Protect blue. Pay attention to weight, endpoints, and whatever weird new adjacencies the missing colors have created.

Use triples only when justified. Dispose sparingly.

If you are doing too much work to save the handful, consider whether the handful deserves to be saved.

Take the solution that feels cleanest.

Accept that sometimes more than one answer works.

And remember: the canonical order is precedent, not statute.

If you have done all of that and still cannot determine the proper course of action:

When in doubt, eat them in the dark.