Table 1.

Design Table

	Question	Hypothesis	Sampling plan (e.g., power analysis)	Analysis Plan	Interpretation given to different outcomes	Outcome
Analysis Step 1: task effect replication	Are exploitation and exploration horizon-modulated?	1.1.a: Less exploitation in the long horizon: High-value bandit frequency: SH > LH 1.1.b: More value-free random exploration in the long horizon: Low value bandit frequency: SH < LH 1.1.c: More novelty exploration in the long horizon: Novel bandit frequency: SH < LH	Pilot data lowest effect size: 0.410. To detect with 95% power, a similar effect size requires a sample of N = 83.	Paired samples t-test (or Wilcoxon signed-rank test if the Shapiro normality assumption is violated).	1.1.a: No effect: no evidence that our horizon manipulation modulated overall exploitation. Opposite effect: overall exploitation is increased in the long horizon. 1.1.b: No effect: no evidence that our horizon manipulation modulated value-free random exploration. Opposite effect: value-free random exploration is increased in the short horizon. 1.1.c: No effect: no evidence that our horizon manipulation modulated novelty exploration. Opposite effect: novelty exploration is increased in the short horizon.	1.1.a: Hypothesis confirmed: Less exploitation in the long horizon. 1.1.b: Hypothesis confirmed: More value-free random exploration in the long horizon. 1.1.c: Hypothesis confirmed: More novelty exploration in the long horizon. Answer to research question: Yes, exploitation and exploration are horizon-modulated.
	Is exploration beneficial for participants?	1.2.a: Lower initial reward in the long horizon: Reward of first sample: SH > LH 1.2.b: Higher reward overall in the long horizon: Reward averaged over samples: SH < LH	Pilot data lowest effect size: 0.835. To be detected with 95% power, a similar effect size requires a sample of N = 21.		1.2.a: No effect: no evidence that participants sacrificed a higher initial outcome in the long horizon. Opposite effect: participants optimised initial reward in the long horizon. 1.2.b: No effect: no evidence that participants took advantage of the information gained in the long horizon. Opposite effect: information gain negatively impacted reward.	1.2.a: Hypothesis confirmed: Lower initial reward in the long horizon. 1.2.b: Hypothesis confirmed: Higher reward overall in the long horizon. Answer to research question: Yes, exploration is beneficial for participants.
	Do participants use exploration heuristics?	1.3: Average BIC score: complex model + $ϵ$ + $\eta$ > other models	Pilot data effect size: 1.304. To be detected with 95% power, a similar effect size requires a sample of N = 10.		1.3: No effect: no evidence that participants combine complex models with heuristics. Opposite effect: Participants are not using complex models with both heuristics.	1.3: Hypothesis confirmed: The average BIC score was higher for the complex models with both heuristics compared to other models. Answer to research question: Yes, participants use exploration heuristics.
	Are exploration heuristics used more in the long horizon?	1.4.a: Value-free random exploration is used more in the long horizon: $ϵ$-greedy parameter: SH < LH 1.4.b Novelty exploration is used more in the long horizon: Novelty bonus $\eta$: SH < LH	Pilot data lowest effect size: 0.446. To be detected with 95% power, a similar effect size requires a sample of N = 71.		1.4.a: No effect: no evidence that $ϵ$ was modulated by the horizon. Opposite effect: value-free random exploration is increased in the short horizon. 1.4.b: No effect: no evidence that $\eta$ was modulated by the horizon. Opposite effect: novelty exploration is increased in the short horizon.	1.4.a: Hypothesis confirmed: Value-free random exploration is used more in the long horizon. 1.4.b: Hypothesis confirmed: Novelty exploration is used more in the long horizon. Answer to research question: Yes, exploration heuristics are used more in the long horizon.
Analysis Step 2: impulsivity	Is impulsivity linked to value-free random exploration?	2.1: Value-free random exploration is positively associated to BIS: $ϵ$-greedy parameter and low-value bandit frequency correlates positively with the BIS total score.	Previous study22 correlation for a similar measure: R = 0.26. To be detected with 95% power, a similar effect size requires a sample of N = 190.	Bivariate and partial (correcting for age and IQ) Pearson correlation and a repeated-measures ANOVA with within factor horizon and between participants variable [impulsivity/ADHD-symptoms]	2.1: No effect: no evidence for an association between value-free random exploration and general impulsivity as measured by the BIS. Opposite effect: Value-free random exploration is negatively associated to BIS.	2.1: Hypothesis confirmed: Value-free random exploration is positively associated to BIS. Answer to research question: Yes, impulsivity is linked to value-free random exploration.
	Are ADHD symptoms linked to value-free random exploration?	2.2: Value-free random exploration is positively associated to ASRS: $ϵ$-greedy parameter and low-value bandit frequency correlates positively with the ASRS total score.			2.2: No effect: no evidence for an association between value-free random exploration and ADHD as measured by the ASRS total score. Opposite effect: Value-free random exploration is negatively associated to ASRS total score.	2.2: Hypothesis confirmed: Value-free random exploration is positively associated to ASRS. Answer to research question: Yes, ADHD symptoms are linked to value-free random exploration.

Summary of preregistered hypotheses from our Stage 1 Registered report (the full protocol can be found at: 10.6084/m9.figshare.14346506.v1) with an additional ‘outcome’ column describing the observed effect. SH: Short horizon condition, LH: Long horizon condition.