Treasure Hunter

You start at the top-left cell of a dungeon with 5 health. Move orthogonally through spaces. X is a wall, a digit is a trap that deals that much damage, H is a potion that restores 10 health, and T is the treasure.
A trap or potion applies only the first time that path enters its cell. Health must remain positive. Return [minimumSteps, maximumRemainingHealth], maximizing health only among minimum-step surviving paths. Return [-1, 5] when no path reaches a treasure.

Function
treasureHunter(dungeon: String[]) → int[]

Examples
Example 1
dungeon = [" X","2T"]
return = [2,3]
The two-step path triggers damage 2 and reaches the treasure with health 3.
Example 2
dungeon = [" H"," XX"," 9T"]
return = [8,6]
The shortest surviving route first takes the potion, then survives the damage-9 trap with health 6.
Example 3
dungeon = [" X","XT"]
return = [-1,5]
Walls isolate the treasure.

Constraints
1 <= rows, columns <= 12.
The grid contains spaces, X, digits, H, and zero or one T.
The top-left cell is a space.
At most 15 trap and potion cells appear.

The trick is state, not the grid. Plain BFS on (row, col) fails because a trap or potion only applies the first time a path enters it, so two paths to the same cell can have different histories. Index each trap and potion cell 0 to 14, and run BFS on (row, col, mask). Health is determined by the path, so store the best health seen per state, and only revisit a state if health improves at the same step count. Process level by level. When any level reaches T, return that step count and the max health among those arrivals. The common pitfall is applying damage on every re-entry, or keeping a plain visited set that throws away a better-health path. Also check health stays above zero after each trap. With a 12x12 grid and 2^15 masks, this is fine. If you blank on the mask idea during the live OA, StealthCoder is the hedge.

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