**Participants.** 174 Prolific workers (age mean = 27.1, range 18 to 49; 80 males, 93 females, 1 other) completed three online sessions across three consecutive days in return for monetary compensation. All participants reported normal or corrected vision, no current or past psychiatric or neurological diagnosis (see SI). The study protocol was approved by the Research Ethics Council of Tel-Aviv University and all participants signed informed consent before participating in the study. **Procedure.** In the first session, 200 participants performed 120 trials of a working memory capacity measurement. In the second and third sessions 178/174 participants respectively, completed in total 300 trials of a reinforcement learning task under three working memory load conditions. **Reinforcement learning task.** Participants completed a reinforcement learning bandit task interleaved between the memory array and test array stages of a working memory task (Figure 1). This design allowed us to examine outcome-irrelevant learning processes under different working memory load conditions. Trials started with a memory array stage where participants had to memorize a visual array (i.e., colored squares). Next, participants made two decisions on two sequential offers of a reinforcement learning bandit task where they had to choose one of two cards to gain monetary rewards. Finally, the trial ended with a test array stage, where participants were asked to report whether a newly presented colored square was part of the initial visual array or not. We will now describe the memory array, multi-armed bandit trials, and test array stages that were included on each trial with more details: The memory array stage of the working memory task included the brief presentation (200 ms) of a visual array (i.e., colored squares) which the participants had to remember until the test array stage. To manipulate working memory load, we included three types of visual arrays ( (a) no-load (a fixed color for the entire block of trials), (b) low load (one colored square, with its color being randomly selected each trial by the computer), and (c) high load (four colored squares, with all colors being randomly selected by the computer each trial without replacement out of 9 possible colors). The three conditions (i.e., no load, low and high load) were manipulated between blocks. The location of the square/s was further randomized each trial by the computer (see SI for further details). ![trial_sequence_rl][2] *Figure 1. Trial sequence in the reinforcement learning task, which was performed under varying working memory load. Participants were asked to first memorize a visual array (memory array stage), then make two choices across two card offers (reinforcement learning bandit task stage), and finally report whether a target was the same or different compared to the visual array that was memorized at the trial initiation (test array stage). Working memory load was manipulated between blocks by including in the memory stage either four random colored squares (high load), one random colored square (low load), or a fixed color squared (no-load).* **Reinforcement learning bandit task** included two offers for each trial (interleaved between memory array and test array stages), which were designed to allow us to estimate credit assignment to the outcome-irrelevant response keys. In each trial, the computer allocated four cards (without return) to the two offers (i.e., first or second). Therefore, each offer allowed individuals to chose one of two cards, and these two cards were further randomly allocated to the right or left sides of the screen (see Figure 1). Offers in the reinforcement learning bandit task started with a fixation (900ms), followed by the presentation of two cards in the right/left location. Participants then choose a card freely using a right/left corresponding response-key press (‘s’ or ‘k’ keys in a QWERTY keyboard; until response with 6sec deadline). After making a choice the unchosen card disappeared and the chosen card remind to allow choice feedback (500ms). Cards led probabilistically to reward (£0 or £1, play-pound coins; outcome presented for 750ms) according to a true expected value (which slowly drifted across trials according to a predefined random walk; see SI). Participants were asked to do their best to make choices that will maximize monetary return. To further ensure participants’ motivation, we rewarded them with a monetary payment bonus at the end of the study according to their wins in the task. Importantly, only the cards predicted reward, but not locations or response-keys used to report cards’ selection. This fact is important since it renders any credit assignment to the location/response key as outcome-irrelevant learning (Shahar et al., 2019, 2021). Participants were told explicitly during the instruction phase that only cards predicted reward and not the response-keys used for their selection (see SI). Before starting the task, participants were asked to complete a short quiz (9 questions long) to ensure they read and understood the instructions (see SI), which also included a specific question regarding which task features predict a monetary outcome. Participants had to show 100% accuracy in the quiz, and if they scored lower, they were prompted to the beginning of the instructions phase and were asked to retake the quiz. **Test array stage** included a target screen in which one colored square was randomly selected by the computer in a color that was either the ‘same’ or ‘different’ color as the square that appeared in the same location in the memory array stage. Participants were asked to respond ‘same’ or ‘different’ by pressing ‘s’ or ‘k’ response-keys (keyboard mapping was counterbalanced between participants; until response with a 6sec deadline). Following the target screen. Participants were presented with feedback indicating whether their response was correct or incorrect, which was then followed by a fixation screen (inter-trial interval, 500ms fixation). **Working memory capacity task.** To measure working memory capacity participants were asked to maintain and retrieve visual information (Balaban et al., 2019; Luck & Vogel, 1997; Pashler, 1988). On each trial, a visual array of colored squares appeared (set size of 4 or 8 squares; see SI). Squares in each array had distinct colors and were evenly spread across the screen. Each trial started with the presentation of the squares array (i.e., memory array phase, 200 ms), followed by a fixation cross (i.e., retention phase, 900ms) and then a target (i.e., test array phase, until response with a 6sec deadline). The target screen included one colored square randomly selected by the computer in a color that was either the ‘same’ or ‘different’ color as the square that appeared in the same location in the memorized array. Participants were asked to respond ‘same’ or ‘different’ by pressing ‘s’ or ‘k’ response-keys (keyboard mapping is counterbalanced between participants). Following the target screen participants saw feedback indicating whether their response was correct or incorrect, followed by a fixation screen (inter-trial interval, 500ms fixation). Each participant completed 120 trials aimed to identify individual differences in working memory capacity. ![trial_sequence_wm][1] *Working memory capacity estimation was conducted using the change detection task (Balaban et al., 2019). Participants had to remember four or eight colored squares and then were asked whether the shown square is in the same color as the one previously presented at that location. Accuracy in both conditions was averaged and used to measure capacity according to (Cowan, 2001).* **Self-report data** for exploratory reasons all participants completed the OCI-R, and the ICAR test. Data was collected during 2021 (part of Ido Ben Artzi's Master's thesis) A component with the code converting raw json is also included (not visible for the public) - email us for access if needed. [1]: https://files.osf.io/v1/resources/6tfdj/providers/osfstorage/61f92874026ee607bcb5094b?mode=render [2]: https://files.osf.io/v1/resources/6tfdj/providers/osfstorage/61f92a702082b806924dfe96?mode=render