Present IGS state-of-the-art bias products are focused on differential code biases (DCBs). DCBs originate in hardware delays in theory and are useful in aligning various pseudorange for unbiased solutions. It is usually presumed that such hardware delays, especially those from GNSS satellites, are quite stable over time; empirically, their daily estimation will more than suffice for precise enough DCB products. However, this estimation strategy is not necessarily applicable to phase biases, despite their origination in hardware delays as well. Relevant PPP (precise point positioning) studies have shown that phase biases within narrow-lane observables normally vary up to several tenths of cycles over 24 hours, whereas contradictorily those within the Melbourne-Wübbena combinations stay almost constant over days to weeks. Narrow-lane phase biases have thus to be computed at a sub-daily interval, e.g., every 15 minutes, to ensure the ultimate precision of ambiguity-fixed PPP. It is still unresolved why those phase biases behave diversely over time, and whether super stable narrow-lane phase biases are possible in practice. In this study, we elucidate the “enigma” behind those time-varying phase biases, and develop a method to compute one narrow-lane phase bias per day for each satellite, while still maintaining its compatibility with IGS clock products. Such phase bias products will not only simplify their archive, but also facilitate PPP ambiguity fixing. We computed 10 years (2006-2016) of these phase bias products and stored them in the bias-SINEX format at Wuhan University IGS analysis center (AC) for free public access; an open-source software package has been adapted to use these products and demonstrate how they can improve PPP. We find that our new phase bias product can further reduce the differences of daily positions from IGS weekly solutions by 10-19% on average for the east component, as compared to traditional 15-minute phase biases in terms of RMS statistics. Finally, while our new phase biases conform to the favorable “rule” of stable hardware bias products, this study also suggests a rigorous route to combine phase bias products from various sources or ACs, if IGS is going to release such products in the form of bias-SINEX files.