Skill Badges Beat Global Ranks on Persistence in 900 Runs

Why does a leaderboard position fail to motivate a player who has just lost, while a small badge for "ten clean landings" keeps them returning? The question matters beyond games: it concerns how feedback design shapes persistence under uncertainty. A recent analysis of 900 anonymized play sessions — drawn from a puzzle-platformer where players attempt procedurally generated obstacle courses — found that players shown skill-specific badges returned for more attempts after failure than players shown a global percentile rank. The finding sits at an unusually productive intersection: behavioral psychology's account of reinforcement, decision research on how people evaluate uncertain outcomes, and the design literature on competitive play.

The Two Feedback Conditions

The 900 runs were split into three conditions. In the first, players saw a global rank after each attempt: a percentile placing them against every other player who had attempted that course. In the second, they saw skill badges — discrete, named achievements tied to specific competencies, such as completing a section without touching a checkpoint, or landing a sequence of jumps without a mid-air correction. The third condition showed both. Persistence was measured as the number of retries within a session and the probability of returning within 48 hours.

The badge condition produced roughly 40 percent more retries than the rank condition, and the combined condition did not outperform badges alone. That last detail is the interesting one. If rank were simply an additional motivator, stacking it on badges should have helped. It didn't. Something about the global comparison appears to actively suppress the effect of the more granular feedback.

Why a Percentile Is a Strange Number

A global rank is a single scalar summarizing an entire population's performance against a single course. It is precise, comparable, and almost entirely uninformative about what the player should do next. A player at the 34th percentile learns that they are below average and nothing else. The number is technically feedback, but it carries no actionable structure — it describes a position, not a behavior.

There is a second problem, which decision research has documented extensively. A percentile is a relative measure, and relative measures invite social comparison. Leon Festinger's work on social comparison theory established that people evaluate their abilities against others precisely when objective standards are unavailable or ambiguous. A percentile rank makes the comparison explicit and unavoidable. For a player already struggling, that comparison is rarely flattering, and the cost of confronting it rises with each failed attempt.

Reinforcement Schedules and the Shape of a Badge

The persistence gap is easier to understand through the lens of reinforcement schedules. B.F. Skinner's operant conditioning research distinguished between ratio schedules (reinforcement after a number of responses) and interval schedules (reinforcement after elapsed time), and between fixed and variable versions of each. Variable-ratio schedules — where reward arrives after an unpredictable number of responses — produce the highest and most persistent response rates. This is among the most replicated findings in behavioral psychology, and it is routinely invoked to explain compulsive engagement with unpredictable reward structures.

Badges do not fit that pattern cleanly, and that is worth noting. A skill badge is not a random payout; it is a deterministic, contingent reward for a specific behavior. Yet the path to a badge is variable: a player attempting "five consecutive clean landings" does not know on which attempt the sequence will finally hold. The badge converts an unpredictable process into a legible, bounded goal. The player is not chasing a slot-machine outcome; they are chasing a named competency whose conditions they understand.

That distinction matters for how we read persistence. Persistence driven by variable-ratio reinforcement is often described as compulsive — the behavior persists despite the absence of genuine skill acquisition. Persistence driven by a skill badge is closer to what Anders Ericsson's work on deliberate practice describes: effort directed at a specific, well-defined sub-skill, with feedback tight enough to support correction. The badge is doing pedagogical work, not just motivational work.

The Loss Aversion Asymmetry

Daniel Kahneman and Amos Tversky's prospect theory supplies a complementary explanation. Losses loom larger than equivalent gains — a well-established asymmetry in how people weigh outcomes relative to a reference point. A global rank establishes the reference point at the population median. A player below that point is, in effect, sitting in a loss frame on every attempt, and the psychological cost of each retry includes re-confronting that deficit. A badge resets the reference point to the player's own prior attempt: the question becomes whether the landing sequence held this time, not whether the player is better than strangers.

Kahneman's later work with Angus Deaton on the distinction between life evaluation and experienced well-being is a useful analogy here, though it was not about games. The point is that global, comparative measures and local, experiential measures answer different questions — and people respond differently depending on which one is in front of them at the moment of decision.

What the Combined Condition Suggests

The finding that badges plus rank underperformed badges alone deserves more attention than the headline result. One plausible reading is attentional: the percentile is a salient, easily computed number, and it may crowd out processing of the badge criteria. A player who sees "34th percentile" has a ready-made summary of their standing and little reason to parse the finer-grained achievement list. The badge's instructional content is present but not attended to.

A second reading is motivational. The rank may reassert the social comparison frame that the badge had temporarily displaced, pulling the player back toward evaluating themselves against others rather than against the specific skill. If the badge's advantage comes from localizing the reference point, then reintroducing a global reference point undoes part of that work.

Neither explanation is settled by 900 runs. The sample is small, the game is one genre, and the analysis was observational in part. But the direction is consistent with a broader pattern in the feedback-design literature: granular, behavioral, self-referenced feedback tends to outperform aggregate, comparative feedback for sustained engagement with difficult tasks.

Designing Feedback That Survives Failure

The practical implication is not that competition is bad. Competitive play clearly motivates a large population, and the rank condition did produce persistence — just less of it. The implication is that global comparison is a fragile motivator in the presence of repeated failure, and that it should not be layered on top of skill feedback without care.

Three design moves follow from the findings. First, sequence the feedback: show skill criteria during the attempt and immediately after, and defer the global rank to a separate, less frequent review screen. Second, make badge criteria behavioral and bounded — "three consecutive clean landings" rather than "improved aerial control" — so the player can tell whether they succeeded. Third, treat the reference point as a design variable. Every feedback element implicitly tells the player what to compare themselves against, and that choice has measurable consequences for whether they try again.

The forward-looking question is whether these effects hold outside games. The same structural problem appears in education dashboards that show class percentile instead of specific skill mastery, in fitness apps that rank users against a global cohort, and in workplace performance systems built on relative ratings. In each case, the aggregate comparison is easy to compute and easy to display, and the granular skill feedback is harder to design. The 900-run result is a small piece of evidence that the harder design is the one that keeps people coming back — and, more importantly, the one that tells them what to practice next.