Random Picks Excluding Recent Values

You are given an array of distinct integer values, an exclusion length k, a requested number of accepted picks picks, and recorded random candidate indices draws.
Process draws from left to right. A candidate index selects values[draws[i]]. Reject that candidate when its value appears among the most recent k accepted values. Otherwise, accept it, append it to the result, and update the recent-value history.
Rejected candidates do not change the recent-value history. Stop as soon as picks values have been accepted and return them in acceptance order.
When k is 0, every candidate is accepted. The input guarantees that draws contains enough acceptable candidates to produce the requested result.

Function
pickWithoutRecent(values: int[], k: int, picks: int, draws: int[]) → int[]

Examples
Example 1
values = [10,20,30]
k = 1
picks = 4
draws = [0,0,1,1,2,0]
return = [10,20,30,10]
Accept 10, reject the repeated 10, accept 20, reject the repeated 20, then accept 30 and 10.
Example 2
values = [1,2,3,4]
k = 2
picks = 5
draws = [0,1,0,2,1,3,0]
return = [1,2,3,4,1]
After accepting 1 and 2, the next 1 is rejected. After accepting 3, the next 2 is still recent and is rejected.
Example 3
values = [5,7]
k = 0
picks = 4
draws = [1,1,0,1]
return = [7,7,5,7]
With k = 0, no accepted value is excluded, so every recorded candidate is returned.

Constraints
1 <= values.length <= 10^5.
All values are distinct and satisfy -10^9 <= values[i] <= 10^9.
0 <= k < values.length.
1 <= picks <= 10^5.
picks <= draws.length <= 2 * 10^5.
0 <= draws[i] < values.length.
The recorded candidates contain enough accepted values to produce exactly picks results.

The trick is keeping two structures in sync: a queue (or the result array itself) holding the last k accepted values, and a hash set for O(1) membership checks. For each draw, look up values[draws[i]]. If it's in the set, skip it. Otherwise append it to the result, add it to the set, and if the window now exceeds k, evict the oldest accepted value from the set. Total work is O(draws). The common pitfall is updating history on rejected draws, which the statement explicitly forbids. Another is the k = 0 case: if you add then evict, make sure the set stays empty, or just short-circuit and accept everything. Since values are distinct, checking the value or the index is equivalent. Stop the moment you hit picks accepted results. If the window logic slips under pressure, StealthCoder is the hedge for the live OA.

StealthCoder is the hedge for the one pattern you didn't drill. It runs invisibly during the screen share.