Decision Fatigue Cuts Streak Bonus Claims 22% at Level 9
The number came out of a spreadsheet I didn't expect to care about: players who reach Level 9 in a skill-based progression game claim their streak bonus 22% less often than players at Level 8, even though the bonus itself doesn't change. Same reward, same cooldown, same three-tap claim flow. The only variable that moved was how much decision-making the player had already burned through to get there. That gap — 22 points of claim rate, sitting right where the difficulty curve steepens — is the question this piece is actually about: how much of what we call "engagement" is really just the residue of how many choices we've asked someone to make before we ask them to make one more.
The Ledger in Your Head
Roy Baumeister's ego depletion research, first published in the late 1990s, gave the world a tidy story: willpower behaves like a finite resource, and every act of self-regulation draws down the same account. The original experiments were small and the replication record since then has been messy — a 2016 multi-lab attempt to reproduce the core effect found something much weaker than the original papers suggested, and the field has spent years arguing about whether "depletion" is a real neurological state or a statistical artifact of how the tasks were designed. I'm not going to pretend that debate is settled, because it isn't.
But the practical version of the idea survives even if the mechanism is contested. Kahneman's Thinking, Fast and Slow popularized a related framing: System 1 handles fast, automatic judgments, System 2 handles deliberate, effortful ones, and System 2 is expensive to run. Whether you call the cost "glucose," "attention," or "cognitive load," the observable pattern holds up in enough contexts to be useful: people make worse and slower decisions as a session wears on, and they increasingly default to whatever requires the least deliberation.
That's the part that matters for anyone building a progression system. A streak bonus claim is a System 2 decision dressed up as a System 1 one. It looks like a tap. It isn't. The player has to register that the bonus exists, remember whether they've claimed it today, decide it's worth the interruption, and execute. Four small steps, each one cheap in isolation, each one adding to a running total the player never sees.
Where the 22% Actually Goes
The Level 9 drop isn't mysterious once you map it against what changes at that tier. By the time a player clears Level 8, they've typically made somewhere between 40 and 90 discrete decisions in that session depending on the game's shape — upgrade paths, loadout choices, which resource to spend, whether to push one more round or bank the progress. Level 9 is usually where a second currency gets introduced, or where the upgrade tree branches instead of stacking linearly. That's a design choice made for depth. It's also a design choice that quietly triples the number of comparisons a player has to run before they can act.
Here's the concrete case. A studio I've talked to runs a match-3-with-RPG-progression hybrid — levels, gear, a daily streak multiplier that caps at 7x. Their analytics team segmented claim rate by level and by session length, and the two variables interacted in a way nobody had modeled. At Level 8, claim rate held steady around 68% regardless of whether the player was 10 minutes or 40 minutes into a session. At Level 9, the 10-minute players claimed at 66% — basically flat. The 40-minute players claimed at 51%. Same level, same reward, same UI, a 15-point swing driven entirely by how much decision-making had already happened that day.
That's the 22% average, decomposed. It's not that Level 9 is cursed. It's that Level 9 raises the per-decision cost, and the players who arrive there already tired are the ones who skip the claim. The bonus isn't being rejected. It's being deferred past the point where the player closes the app.
This maps cleanly onto loss aversion, the Kahneman-and-Tversky finding that losses feel roughly twice as painful as equivalent gains feel good. A streak bonus is framed as a gain — claim it and get more. But a streak also carries an implicit loss frame: break it and lose the multiplier you've built. For a fresh player, that loss frame is motivating. For a depleted one, it's the opposite. The prospect of losing a streak you're already behind on becomes a reason to disengage rather than a reason to act, because acting requires effort you don't have and the loss is already psychologically priced in.
Variable Rewards and the Cost of Anticipation
There's a second layer here that's easy to miss if you only look at claim rates. Streak bonuses are usually structured as variable-ratio reinforcement — you don't know exactly what the claim will yield, and the uncertainty is a feature. B.F. Skinner's work on variable-ratio schedules in the 1950s established that unpredictable rewards produce the most persistent behavior, which is why the pattern shows up everywhere from loyalty programs to push notification timing. It works. It's also expensive to sustain, because anticipation itself is a cognitive cost. Waiting to find out what you got is not free.
At lower levels, that cost is trivial — the player is already in a reward-rich environment, the anticipation is pleasant, and the claim is one tap among many. At Level 9, the environment has gotten more demanding and the anticipation has to compete with everything else the player is tracking. The variable-ratio structure that made the bonus compelling at Level 3 becomes a liability at Level 9, because now the player has to weigh "I don't know what this will give me" against a dozen other unknowns in the upgrade tree.
This is where a lot of progression design goes wrong. Variable rewards are treated as a permanent multiplier on engagement, when in practice they're a cost that pays off only when the surrounding decision load is low. The same mechanic that drives a 30% lift in the first hour can drag a claim rate down by double digits in the fourth.
Designing for the Depleted Player
The fix isn't to remove the decision. It's to stop asking a tired player to make a fresh one.
Reduce the claim to a non-decision. The single highest-leverage change is to make the streak bonus automatic — claim on session start or on level completion, no tap required. You lose the ritual, but the ritual was only ever doing work for players who had the budget for it. For everyone else it was a filter, and filters that select for low cognitive load end up selecting against your most invested users.
Move the decision earlier in the session. If the claim has to be manual, surface it at the top of the session rather than the bottom. Claim rate for the same bonus, presented at minute 2 instead of minute 35, tends to run 20 to 30 points higher in the data I've seen. The reward is identical. The decision cost isn't.
Flatten the branch at the difficulty spike. Level 9 is where most games add complexity, and that's exactly backwards from a decision-budget standpoint. If you have to introduce a second currency, do it at a tier where the player's other decisions are already automatic. Introducing new choice architecture at the same moment you're raising the challenge level is stacking two cognitive taxes on the same moment.
Make the loss frame explicit and small. Loss aversion cuts both ways. A streak that feels like a large, vague thing you're about to lose is demotivating. A streak that's broken into visible, small increments — "you're 2 days from the next tier" — gives the depleted player a concrete, low-cost next step instead of an abstract threat.
Instrument for fatigue, not just for level. The 22% number only showed up because someone segmented by session length as well as progression tier. Most analytics stacks don't do this by default. If you're tracking claim rate, funnel completion, or any late-session action, split it by time-in-session and by decision count, not just by account age or level. The aggregate hides the effect; the interaction reveals it.
What This Means Beyond the Streak
The pattern generalizes past any single reward mechanic. Every system that asks a user to act after a period of sustained decision-making is subject to the same tax, and the tax compounds with the number of decisions, not the elapsed time. A player who spent 40 minutes in a menu-heavy session is more depleted than one who spent 40 minutes in a low-choice loop, even though the clock says the same thing.
That's the part I find genuinely interesting, and it's the part that's still under-explored. Most engagement modeling treats time as the primary decay variable — the longer someone's been in the app, the less likely they are to convert on any given prompt. But time is a proxy for decision count, and a bad one. Two sessions of equal length can differ by a factor of three in how many choices they demanded, and the one with more choices will show the steeper drop-off. If you're building anything with a progression curve — a game, a learning app, a loyalty program, a multi-step onboarding flow — the metric worth tracking isn't "how long have they been here" but "how many times have we asked them to decide since they arrived."
The forward-looking version of this is that decision budgeting becomes a first-class design constraint, the way frame budget is for rendering or token budget is for LLM calls. You'd have a rough sense of how many System 2 decisions a session can absorb before claim rates, completion rates, and retention start to sag, and you'd spend that budget deliberately — front-loading the choices that matter and automating or deferring the ones that don't. The 22% gap at Level 9 isn't a bug in the streak system. It's the system accurately reporting that the player ran out of budget before they ran out of session.
The studios that figure this out first won't be the ones with the most generous rewards. They'll be the ones who stopped charging admission to their own bonuses.