The triangular distribution is typically used for reasonably well understood tasks in which there may be both quantity uncertainty and a limited amount of scope uncertainty. An upper limit can be specified which is unlikely to be exceeded.
A triangular distribution has the highest probability at the base or mid-range cost, and decreases linearly away from this cost to the range limits.
Modelling a cost as $10,000 -10%/+20% with a triangular distribution indicates a belief that the actual cost is going to be in the range $9,000 to $12,000 with a most likely cost of $10,000. The probability of any other cost will decrease linearly on either side of this value, finally reaching zero probability at $9,000 and $12,000 respectively.
Using this distribution also indicates a belief that under no circumstances will the actual cost be outside the specified range.