Decision 04 · Week 4 · 11 August 2026

The Best Augment Is Not Always the Strongest Augment

Why population-level augment strength fails to settle an individual decision

Augment rankings exist, they are easy to find, and they are largely correct about the thing they measure. This week is about the gap between the thing they measure and the decision in front of you.

Note first what kind of week this is. Eleven of the twelve take something the mode removed and ask what is still decidable inside the hole. This one does the opposite. Week 3 dealt with the reroll, the only choice the mode hands back before the game starts; augments are the choice it hands back during one, and they are the largest genuine selection this mode contains — a real menu, offered three times, with no wrong-looking options on it.

That is why it is the hardest week rather than the easiest. Everywhere else the constraint does half the analytical work: no-pick narrows the question to what can this champion be made to do, and a narrow question is a tractable one. Here the constraint lifts, the option space opens, and the reader is handed back exactly the kind of decision the rest of the course never has to make — which is also the kind a ranking is most confident about, and most confidently wrong about.

Two different questions#

A ranking answers: across a large number of games, which augment is associated with the most wins?

You are being asked: of these particular options, offered to me, on this champion, alongside what I already have, which is best?

The two questions coincide only when your game resembles the average one. The ranking is computed over every champion that was offered the augment, every composition it was taken into, and every player who took it. Yours is one draw from that population, and the features that make it yours are exactly the ones the average discards.

This is not an argument against rankings. It is an argument about what kind of object a ranking is: a prior, strong enough to start from and too coarse to finish with.

Four things the ranking cannot see#

Your champion’s damage profile. An augment that multiplies one large hit and an augment that adds to many small ones are not interchangeable, and their population averages are taken over champions of both kinds.

What you already have. Augment value is conditional, not additive. A second source of the same effect is worth less than the first, sometimes much less, and the ranking scores each augment as though it were the only one you owned.

Your team. An augment that rewards fighting alongside allies is worth one thing on a team that groups and another on a team that cannot.

The enemy. An augment that solves a problem you do not have is a wasted selection, regardless of how well it performs against the problems other people had.

Redundancy and marginal value#

The single most common error is treating augment selection as pick the strongest rather than pick the one that adds the most.

These differ whenever you already own something. A durability augment taken by a champion who is already unkillable in the fights they are actually in converts gold-equivalent value into a statistic nobody will notice. The same augment on the same champion, two rounds earlier, might have been the pick of the set.

The question is always marginal: what does my ability to affect this game look like with this, minus what it looks like without it? Rankings answer the absolute question. Nobody has ever needed the absolute question answered.

Risk, and when you want it#

Augments differ in variance as well as in expectation. Some produce a reliable increment; others produce a large effect conditional on something happening.

The reliable one is usually correct, and this is where rankings and good play agree most often. But the correct amount of variance is not fixed — it depends on whether the reliable outcome is good enough. A team that is behind does not need a small consistent improvement to a losing position; it needs the distribution widened, because the left tail is already where it is going. Week 11 develops this properly.

For now the shape of the rule: take variance when the expected outcome is a loss, and consistency when it is a win. Neither is a property of the augment. Both are properties of the game you are in when you are offered it.

A case#

You are Vladimir, on a team that groups well, against a composition with no obvious way to burst you down. You are offered a strong damage augment, a sustain augment, and a high-variance one.

The ranking may well put the sustain augment first, and against most compositions it would deserve that. Against this one you are already surviving the fights — the reason you are not winning them is that nothing on your team converts a won fight into dead enemies. The augment that adds the most is the one that solves the problem you have, which is not the problem the population had.

The counterargument#

A player without the knowledge to make this read will do worse deviating than following the list. That is true and it is the honest cost of this week: taught badly, it produces confident deviation on no evidence, which is worse than compliance.

The defence is procedural. Deviation requires a stated reason of the form the ranking assumes X, and X is false here. If you cannot name X, take the ranking’s answer. It is a good prior and you have no evidence against it.

Takeaway#

Rank the options against your game, not against the population. The list is where the decision starts and it has never once been where it ends.