The Psychology of Completion: Why Unfinished Games Nag You
Updated August 5, 2026
The Zeigarnik effect is the claim, first published in 1927, that people remember tasks they were interrupted during better than tasks they finished, and it is the standard explanation for why an unfinished game keeps surfacing in your head. The best available evidence says that explanation is wrong. A 2025 meta-analysis of 59 publications found the pooled ratio of interrupted to completed tasks recalled was 0.99, which is no memory advantage at all. What did survive was a different effect from the same era: across 21 publications, people resumed an interrupted task 67.00% of the time. Unfinished things are not remembered better but returned to, and that is a different mechanism with different design consequences.
What did Zeigarnik actually find?
Bluma Zeigarnik’s 1927 paper “Das Behalten erledigter und unerledigter Handlungen” appeared in Psychologische Forschung. A scan of the original is public, and it is a German-language image scan, which matters for what follows.
Almost every popular retelling attaches two things to it. The first is a cafe anecdote in which Kurt Lewin watches a waiter recall unpaid orders perfectly and forget them the moment the bill is settled. The second is a number, usually phrased as interrupted tasks being recalled about twice as well, or roughly 90% better. We went looking for a primary source for either and did not find one, and the English-language articles quoting the ratio do not cite a page. We are not repeating either as fact.
The part that is verifiable is what happened afterwards. Romain Ghibellini and Beat Meier’s meta-analysis in Humanities and Social Sciences Communications pooled 59 publications: 38 examining the Zeigarnik effect, 20 examining the Ovsiankina effect, and one covering both. The weighted ratio of interrupted to completed recall came out at 0.99, and it stayed at 0.99 when Zeigarnik’s own study was removed. Across 13 publications, interrupted tasks made up 49.16% of everything recalled, which is a coin flip. The pooled effect size across the 8 publications that reported one was a Cohen’s dz of 0.15, which the authors call a small effect. They report no confidence intervals and say plainly that the limited number of studies behind each measure needs to be taken into consideration when interpreting the results.
Their explanation for the original result has nothing to do with memory. Experimenter authority and situational demand were far stronger in a 1920s European university than in a modern lab, and an interrupted participant then had reasons to stay gripped that a participant today does not.
Where is the popular understanding wrong?
The Zeigarnik effect is sold as a memory tool: cliffhangers, stopping a study session mid-sentence, leaving an advertisement unresolved. The memory claim is precisely the part that failed to replicate. If you have used it to argue that a player will remember your game better because you cut them off, the evidence does not support you.
There is a second, quieter error. The effect is usually described as pleasant, a productive itch. The strongest modern evidence for anything Zeigarnik-shaped comes from occupational health research, where unfinished tasks are studied as a stressor. A 2026 meta-analysis by Johannes Wendsche, Oliver Weigelt and Christine Syrek in Anxiety, Stress and Coping combined 17 between-person samples (N = 2,473) with 14 within-person samples (N = 12,129). Unfinished work tasks were positively related to work-related thoughts during off-job time, at rho = .382 between persons and rho = .247 within persons, with affective rumination showing the strongest associations.
The replicable footprint of an unfinished task is intrusive thought while you are trying to do something else. These are correlations rather than demonstrations of cause, but the mechanism keeping a player thinking about their half-upgraded building looks like the one keeping a worker thinking about an unsent email. That is the lever a designer is pulling.
If it is not memory, what is pulling you back?
Maria Ovsiankina’s 1928 dissertation “Die Wiederaufnahme unterbrochener Handlungen” was published in Psychologische Forschung, volume 11, pages 302 to 379, out of the same German research programme. Her question was what people go back to.
This is the effect that held up. In the same meta-analysis, the pooled resumption rate was 67.00% across 21 publications including her original, and 66.79% across the 20 publications excluding it. Removing the founder’s own data moved the result by about two tenths of a percentage point. That is exactly the test the Zeigarnik effect failed, and the Ovsiankina effect passed it while being the one almost nobody cites.
Two details matter. Resumption is measured in setups where the interrupted task is still available and the participant is given free time afterwards, so the pull is observed under conditions that leave room to act on it. And a third of people never resumed. Two out of three is a tendency, not a compulsion.
Why does effort only pay off once the task closes?
Michael Norton, Daniel Mochon and Dan Ariely’s IKEA effect is usually summarised as labour increasing valuation. The working paper, later published in the Journal of Consumer Psychology (volume 22, issue 3, pages 453 to 460, in the July 2012 issue and online in 2011), contains a finding that gets left out.
In their first experiment, participants who assembled an IKEA “Kassett” box bid a mean of $0.78 for it against $0.48 from participants offered an identical pre-assembled box, t(50) = 2.12, p < .05. In their third experiment, 39 participants either assembled a box or were stopped before the last two steps. The incomplete builders had every piece in front of them and only trivial effort left. Completers bid a mean of $1.46. Incomplete builders bid $0.59, t(37) = 2.35, p < .05. In a Lego experiment, builders bid more for their own set than their partner’s, t(39) = 3.08, p < .01, but if they were made to take their creation apart first, that difference became non-significant, t(39) = 1.20, p = .23.
Effort converts into value at the moment something closes. Those are separate experiments with separate samples, so the dollar figures should not be compared across them, and none of them tested a half-finished item against one the participant never touched. What each experiment shows on its own is that the premium attaches to completion.
That is the line between this post and why restoring things in games feels good, which covers the endowed progress effect and the satisfaction of visible change. That post is about the reward. This one is about the state before it, and the IKEA results show the two move separately. Work goes in continuously. Value only lands when something closes.
Why are management and building games so dense with unfinished states?
Because their entire interface is a list of started, unfinished actions. A room grid with empty cells. A tenant roster with vacancies. An upgrade tree showing three of five levels bought. A collection screen of silhouettes. In a shooter an ammo counter is a resource readout; in a management sim a “3/5 rooms upgraded” line is an interrupted action rendered as a permanent UI element, and there are dozens on screen at once. Which of those open states is real progression and which is only a gate is a separate sorting job, and we did it by type in 10 games with the most satisfying progression systems.
The mechanism suggests a design rule that is not obvious. If the pull comes from resumption, the design variable is how cheap the open state is to act on. Ovsiankina’s paradigm measured resumption in exactly that situation: the task still available, and unpressured time in which to return to it. A game that surfaces twelve open tasks and then buries the means to advance any of them behind a long reload, an unclear next step, or a resource the player cannot see how to get produces the intrusive thought without the return. That is the stressor half of the research with none of the payoff.
How many players actually finish?
Steam publishes global unlock rates for every achievement, which gives a rough public measure of how far players get in games built around completion.
| Game | Studio | Year | An early achievement | A full-completion achievement |
|---|---|---|---|---|
| Stardew Valley | ConcernedApe | 2016 | Greenhorn, earn 15,000g: 62.7% | A Complete Collection, complete the museum: 7.1% |
| Two Point Hospital | Two Point Studios | 2018 | Cure 1 patient: 94.2% | The New Gold Standard, earn 54 gold stars: 2.3% |
| PowerWash Simulator | FuturLab | 2022 | Starting Out, earn 5 stars in career mode: 82.0% | As It Was Foretold, complete career mode: 19.4% |
Studio and year as credited on each game’s Steam store page. Achievement percentages are Steam global unlock rates, checked August 2026.
Stardew Valley is the clearest case: its store page frames restoration as the spine of the game, offering the choice to repair the old community center or join forces with Joja Corporation instead. Local Legend, for restoring the Pelican Town Community Center, sits at 22.5%. The museum, the most literal unfinished-collection interface in the game and a wall of empty slots you can stare at from hour one, is completed by 7.1%.
One caveat: Valve’s own Steamworks documentation explains how to request global achievement percentages but does not document which population the denominator covers. Treat the decimal places as noise and read the shape. The shape is consistent across three games from three studios. Nearly everyone starts. Between 2.3% and 19.4% reach the end. Most players who see the empty grid never fill it. On the difference between a game that ends and a game that merely stops, see Unpacking versus PowerWash Simulator.
Where does our own game sit in this?
Grandma Left Me a Building is our cozy apartment management sim, and it is unreleased. We have no player data, no reviews and no reception to point at, so nothing below is evidence that any of it works.
The design has no time pressure and lets you plan at your own pace, which removes almost every source of tension a management game normally has and leaves incompleteness carrying most of the load. The open states are literal: rooms with nobody in them, an applicant sitting in a drag-and-drop roster until you place them, and 20+ building upgrades on a multi level track where owning one level shows you the next. The research turns that from a vibe into a bet we can name. The measurable footprint of an unfinished state is intrusive thought, and the IKEA results say effort only converts into value at the moment something closes, which is why placing one tenant and buying one upgrade level both have to land as finished on their own. Our tenant design is written up in designing tenants that feel alive. Whether the balance lands is not something we can know before people play it.
Frequently asked questions
Is the Zeigarnik effect real?
The memory version has not held up. Pooling 38 published studies gives a recall ratio of 0.99 between interrupted and completed tasks, meaning no advantage, and the pooled effect size across the studies that reported one is small at dz = 0.15. The related resumption tendency, measured in the same analysis, is robust at around 67%. So something real is happening with interrupted tasks, and it shows up in what people go back to.
What is the difference between the Zeigarnik and Ovsiankina effects?
Zeigarnik measured recall: whether people remember interrupted tasks better than finished ones. Ovsiankina measured behaviour: whether people go back to an interrupted task when nothing forces them to. They come from the same 1920s German research programme and are constantly conflated, but only the second replicates, and only the second matters for design. Games care about whether a player returns.
Does a progress bar make people finish a game?
The public evidence suggests it holds attention far better than it produces completion. Across Stardew Valley, Two Point Hospital and PowerWash Simulator, achievements marking full completion sit between 2.3% and 19.4%, while early achievements in the same games run from 62.7% to 94.2%. Treat a progress interface as a reason to come back. It will not carry most players to the end.
Should designers deliberately leave things unfinished?
Deliberately interrupting a player is riskier than it looks, because the strongest modern evidence for unfinished-task effects comes from stress research where they are associated with rumination during time off. The safer reading: make open states visible and resumption cheap. An open state the player cannot act on is the cost without the benefit.
Key takeaways
- The memory claim is dead. A 2025 meta-analysis of 59 publications puts the interrupted-to-completed recall ratio at 0.99, and removing Zeigarnik’s own 1927 data does not change it.
- The resumption claim is alive. Pooled resumption sits at 67.00% across 21 publications and 66.79% with the founding study removed, which makes the Ovsiankina effect the better-supported and far less cited of the pair.
- Unfinished states show up as intrusive thought. The strongest modern evidence on unfinished tasks comes from work-recovery research, where they are associated with rumination during off-hours at rho = .382 between persons.
- Effort only pays on completion. In the IKEA effect experiments, builders stopped two steps short of finishing bid a mean of $0.59 against $1.46 from those allowed to complete the box, and making people take a finished Lego set apart left them valuing it no more than a stranger’s.
- Completion interfaces hold attention more than they produce completion. Full-completion achievements in three management and cleaning games sit between 2.3% and 19.4%, against early achievements between 62.7% and 94.2%.
Grandma Left Me a Building is in development and you can follow it on its Steam page.
