Transcript Window Metrics

Transcript events arrive in order. Event i has user userIds[i] and transcript length lengths[i].
For every contiguous window of exactly windowSize events, return one row [maximumTranscriptLength, distinctUserCount]. Rows must follow window order.

Function
transcriptWindowMetrics(userIds: String[], lengths: int[], windowSize: int) → int[][]

Examples
Example 1
userIds = ["a","b","a","c"]
lengths = [4,7,5,9]
windowSize = 3
return = [[7,2],[9,3]]
The first window has users a and b; the second has a, b, and c.
Example 2
userIds = ["u","u","v"]
lengths = [2,8,1]
windowSize = 2
return = [[8,1],[8,2]]
The maximum stays 8 while the distinct-user count changes.
Example 3
userIds = ["x","y"]
lengths = [0,3]
windowSize = 1
return = [[0,1],[3,1]]
A one-event window always has one distinct user.

Constraints
1 <= windowSize <= userIds.length == lengths.length <= 100000.
0 <= lengths[i] <= 1000000000.
User IDs are nonempty strings.

The edge case that breaks the naive solution is size. Rescanning every window costs O(n * windowSize), which is up to 10^10 operations. Split the work. For distinct users, keep a hash map of counts. Add the incoming user, decrement the outgoing one, and delete the key when its count hits zero. Map size is your answer. For the max, use a monotonic deque of indices with decreasing lengths. Pop from the back while the new value is bigger, pop from the front when the index leaves the window. Front is the max. Common pitfall: forgetting to delete zero-count keys, which inflates the distinct count. Also don't use a plain running max, because it can't shrink when the big value exits. Lengths go to 10^9, so int is fine, but keep the output as an int[][]. StealthCoder is your hedge if the deque logic slips live. Total time is O(n).

The honest play: practice the pattern, and have StealthCoder ready for the one you didn't see coming.