What I've Been Building

These three metrics break down each interaction into its specific motor components, search, approach, and commit, following the same targeting framework often used in esports performance research. The sharp peak below 500ms in the acquisition time histogram shows that most of the targeting is fast and automatic as this measures how quickly the player locates and orients to a target. Aim efficiency is the measurement of how accurately a player can aim at a target with values closer to 1.0 indicating aiming straight at the target with no correction. The 0.84 median value from this dataset indicates that the players are reasonably direct, but still make some corrective movements, which is consistent with a 3D environment where the target shelf isn't always immediately visible. Dwell time is the final aim metric calculated with longer dwell times suggesting a player is deliberating the placement, confirming the correct shelf before placement. The median value of 834ms for dwell time vs 283ms median acquisition time suggests that confirming the correct placement of an item demands more cognitive effort than locating the shelf in the first place. In an esports context, these same three metrics directly measure a player's motor fluency under pressure. The difference is that research labs typically derive them manually from gameplay that exports data to be analysed whereas LudoSense computes them automatically from the SDK at capture time.

These two visualisations capture complementary aspects of input behaviour: the interval distribution shows how fast a player issues consecutive inputs, while the transition matrix shows which action follows which. Together, they give a complete picture of input tempo and sequencing. The mass sitting between 200ms and 800ms indicates rapid, rhythmic input during active gameplay, with a long right tail where the player pauses to deliberate. This is a granular measure of motor fluency that maps directly to actions-per-minute in esports research. The transition matrix reveals the dominant gameplay loop in Alchemy Shoppe Chaos: Interact → Move → Interact, corresponding to picking up an item, moving to a shelf, and placing the item where it belongs. In a sorting game such as this, this is the expected pattern and is consistent across players. In games with more complex input mechanics, such as fighting game combos, MOBA ability rotations, or FPS strafe-shoot patterns, the same matrix computation would surface skill-differentiating sequences. LudoSense computes both from the same stream of key events captured at the SDK level.

These four curves track performance across a single 12-minute session of gameplay, converting what would otherwise be static summary statistics into a temporal profile. Actions-per-minute measures raw input activity, tracking how actively a player is issuing inputs across the session. Angular velocity captures visual scanning intensity with higher values indicating active visual scanning and lower values demonstrating the player has locked onto a target or paused to deliberate. Items shelved measures productive throughput with rising values indicating the player is learning the shelf layout and declining values demonstrating possible fatigue or increasing difficulty. Lastly, error fraction measures the overall accuracy of play during a session with any value above zero meaning the player made at least one wrong-bookcase, wrong-shelf, or wrong-ordering error in that minute. Taken together, this temporal profile reveals whether performance in a session is stable, improving, or degrading over time. These patterns map directly to learning curves, attentional fatigue, and engagement states in the performance psychological literature.

Where the per-minute curves show how one player's session unfolds over time, these bars compare the same four metrics across all seven sessions. This is the between-subjects view with the same game and same mechanics, just with different players on different days. Any differences that emerge here could reflect individual skill, familiarity with the game, play style, or even hardware setup. At this sample size, the bars are descriptive rather than inferential, but they illustrate the kind of cohort comparison that LudoSense enables automatically from its per-session aggregation. Filters can be applied that allow for comparison between specific types of players, such as those who use a trackpad vs those who use a separate mouse. A built-in player survey as well as a hardware snapshot will allow for these types of comparisons.

The colour coding above matches the game's own ghost feedback system: green for correct, blue for wrong volume order, yellow for wrong shelf but correct bookcase, and red for incorrect overall. The near-total green across all seven sessions tells a clear game design story: players understood the category system almost perfectly, with zero red errors across most of the sessions. The interesting signal is in the blue colour, the volume ordering errors, which are concentrated almost entirely in one of the illustrated sessions. That pattern would suggest the tutorial may not have landed properly for that hypothetical player in terms of explaining the within-shelf ordering by volume. This is exactly the kind of targeted, actionable insight that distinguishes analytics from raw data. My experience with psychological research allows me to pinpoint not just that players made errors, but the specific mechanics that led to them.

For gaming studios, knowing which settings players change, and what values they change them to, is one of the simplest ways to improve the out-of-the-box experience for gamers. If most players immediately adjust a setting to the same value, such as camera sensitivity, that value should probably be the default for the game itself. The settings that don't appear in the chart are just as informative as they were left untouched by all seven players, which validates those default values. Departures that cluster around a single alternative value demonstrate a strong signal to change the default. Where they scatter across multiple values suggests more of a genuine personal preference rather than a miscalibrated default. LudoSense tracks this passively with the SDK capturing the settings at launch and any changes made during the gameplay, requiring no additional instrumentation from the developer.

Upgrade purchase patterns reveal what players perceive as most valuable and when they feel each constraint. Hold Capacity appears as the first purchase in four of the seven sessions, suggesting inventory limit is the first friction point players tend to encounter. Auto-fly dominates later purchases and frequently closes out the session as the final purchase, demonstrating that once basic carrying needs are met, gathering items more quickly becomes the priority. Item highlight and storage highlight are purchased less frequently, indicating players are comfortable with their ability to identify items and shelves without assistance, which makes sense for the size of this playtest. The fact that 35 of 37 purchases left the player with 0 upgrade points shows that players spend their points as soon as they can afford to, demonstrating that they fully understand the upgrade flow and are treating the purchases as quality-of-life improvements. For game designers, this kind of purchase sequencing data surfaces which mechanics create the most friction and whether the upgrade economy is paced correctly.