Many economic decisions involving private information are made in social environments rather than in isolation. We study dishonest reporting in a controlled dice-rolling experiment in which 522 participants complete the task in groups, never alone, allowing social context and beliefs about others’ behavior to play a central role. Under honest reporting, outcomes should follow a uniform distribution with an expected value of 3.5. In the first reporting stage (Q1), reported outcomes deviate significantly from this benchmark. A chi-square goodness-of-fit test rejects uniformity (χ²(5)=13.91, p=0.016), and the average reported outcome equals 3.70, significantly above the honest benchmark (t(521)=2.69, p=0.0075). Using a mixture approach, we estimate that even under conservative assumptions—allowing dishonest participants to report arbitrarily—at least 26% of observations must originate from a non-honest reporting process to rationalize the observed distribution. After the first reporting stage, we elicit participants’ beliefs about the distribution of reports in the group. Average beliefs across outcomes deviate by at most 1–2 percentage points from 1/6 and do not reflect the substantial excess mass at high values observed in actual reports. In particular, participants do not systematically expect others to overreport the highest outcomes, even though such behavior is prevalent in the data. Reported dishonesty is stronger in the second reporting stage (Q3). The distribution of reports is heavily skewed toward high values, with uniform reporting decisively rejected (χ²(5)=23.29, p<0.001). The mean reported outcome increases to 3.80, remaining significantly above the honest benchmark (t(521)=3.83, p<0.001). A corresponding mixture analysis implies that at least 19.5% of reports in Q3 cannot be explained by honest reporting alone. Although a paired t-test fails to detect a significant increase in the average individual report between Q1 and Q3, a Kolmogorov–Smirnov test rejects equality of the two distributions (p<0.001), indicating a substantial reallocation of probability mass toward higher outcomes once more. Reported outcomes are unrelated to age, gender, or education, indicating that dishonest reporting in this setting is largely independent of observable individual characteristics.