Decision 10 · Week 10 · 6 October 2026
The Correct Build Should Change Mid-Game
Re-evaluation, abandoned plans and the cost of sunk-cost thinking
The claim
Consistency with an earlier plan becomes a fault once the information it rested on has changed.
The decision: when the plan stops being right
Constraints this week is about
- no recall
Week 5 was about choosing with almost no information. This week is about what happens when the information arrives.
The interesting problem is not what should I change to — that is week 5’s question asked again, with better inputs, and most players can answer it if prompted. The problem is that they are not prompted. Nothing in the game interrupts to say the plan has expired. The build keeps completing itself, item by item, on assumptions that stopped being true several minutes ago.
So this week is about detection.
Why revision does not happen#
Three reasons, in ascending order of how much they explain.
Nobody is checking. The build was decided once and became a queue. Purchases are executed from the queue whenever gold allows, which under the no-recall constraint is at respawn, which is exactly the moment a player is least inclined to re-analyse anything.
Revision costs gold already spent. Selling is lossy. Pivoting means the components in your inventory were partly wasted, and that feels like an admission. It is an admission. It is also cheaper than the alternative.
The plan was never written down, so there is nothing to notice the expiry of. A player who cannot state the assumption their build rests on cannot detect that it has failed. This is the real cause, and the other two are symptoms.
Triggers#
The practical content of the week is a short list of observable events, each of which should cause a re-read rather than a specific purchase.
- An enemy buys a resistance itemThat your damage still scales the way you assumed.
- An enemy buys anti-healYour team's sustain plan, immediately.
- A threat emerges from a champion you had discountedYour priority target, and where you stand.
- An enemy augment or item changes their range or mobilityWhether you are actually safe where you are.
- Your team's frontline is dying first, repeatedlyWho the fight is actually being fought around.
- You stopped being able to kill somebody you used to killThe whole build, usually.
Unnumbered on purpose: these do not arrive in an order, and numbering them would claim a sequence the game does not follow.
None of these prescribes an item. Each says the model you have been operating is now known to be wrong in a specific way. What to do about it is week 5’s procedure, run again with the new evidence.
A case#
You are Miss Fortune on a team of Miss Fortune, Orianna, Amumu, Thresh, Darius, against Rammus, Leona, Vex, Twitch and Anivia.
Early on the read was straightforward: their front is durable but their back is not, the team has enough setup to open a fight, and your job is damage into whatever the setup catches.
By the middle of the game two things have happened. Rammus and Leona have both bought armour, and their Vex has begun reaching your position rather than Orianna’s. Neither event is announced. Both invalidate a different part of the plan — the first says your damage no longer converts against the people you have to shoot through, the second says the safe distance you have been standing at is no longer safe.
A player who does not check will continue completing a build that answers a question the game stopped asking, and will describe the result afterwards as their team not following up.
Sunk cost, precisely#
The components already bought are gone whether you pivot or not. They cannot be recovered by continuing to build toward them; that only spends more gold in the same direction. The correct comparison is never what have I already put into this but of the gold I have not yet spent, what does the most work from here.
This is easy to state and genuinely hard to do, because the alternative feels like admitting the earlier decision was bad. Usually it was not. It was correct on the information available and has since been overtaken. Week 9’s distinction applies here too: a decision is judged against what was knowable when it was made, and revising it later is not a confession.
The counterargument#
Constant revision is its own failure. A player who re-evaluates after every death converges on nothing, buys components toward four different builds, and finishes with an inventory that is good against everything and sufficient against nothing. Under the no-recall constraint, thrashing is expensive.
The resolution is that revision should be triggered, not continuous. The list above is short on purpose. If none of those has happened, the plan stands and the queue is fine. Discipline here is knowing which events deserve a re-read, not re-reading constantly.
Takeaway#
Write the assumption down, even if only to yourself, so there is something capable of being falsified. A plan you cannot state is a plan you cannot notice the death of.
Teaching team
- Spike — Convener — Game Systems & Decision-Making Lecturer