A modified backward-facing step flow with a large expansion ratio of five (5) was modeled by 19 teams without benchmark solutions or experimental data for validation in an ISHVAC-COBEE July 2015 Tianjin Workshop, entitled as "to predict low turbulent flow". Different computational fluid dynamics (CFD) codes/software, turbulence models, boundary conditions, numerical schemes and convergent criteria were adopted based on the own CFD experience of each participating team. The largest coefficient of variation is larger than 50% and the largest relative maximum difference of penetration length is larger than 150%. The predicted non-dimensional penetration lengths as a function of the Reynolds number (1-10,000) are found to be significantly diverse among different teams. Even when the same turbulence model or even the laminar model is used, the difference in the predicted results is still notable among different teams. It indicates that the combined effects of a lack of general turbulence model, and possible errors in multiple decisions based on users' experience may have caused the observed significant difference. Prediction of transitional flows, as often observed in building ventilation, is shown to be still a very challenging task. This calls for a solid approach of validation and uncertainty assessment in CFD "experiments". The users are recommended to follow an existing guideline of uncertainty assessment of CFD predictions to minimize the errors and uncertainties in the future.