how to interpret very short effective length in Salmon output

Hello
I am interested in a specific transcript in the quant.sf output file of Salmon. This transcript has length of 153 nt and, according to salmon, an effective length of 13-15 nt, depending on the sample. The transcript have very low counts of 26 TPM and 1 read in a sample. It seems like a big difference between the actual length and the effective length. What is the cause of such a difference and how does it effect the TPM and count output?
Thank you