Tower View strategy: pair the views, then cross the candidates.
Use all six two-sided candidates, filter them at crossings, and test a seven-entry branch that looks locally clean but has no completion. Includes a different fixed board and a complete replay.
A line can match both outside clues and still be the wrong line for the board. The useful habit is to keep every surviving ordering, then let the crossing rows and columns remove options. Here we build a complete candidate list for a two-sided view, apply it to a new fixed board, and test a tempting seven-entry branch that has no immediate diagnostic error but cannot be completed.
Load the paired-views board
Use Practice this example beside the article, or Load strategy board in the game. Confirm Edits 0 and the following outside clues. This is the second arrangement in the existing five-board collection, different from the first-view lesson. It does not add a new game or enlarge the collection. Rows run top to bottom and columns left to right. Read each clue inward from its own edge.
TOP: 3 4 2 1
LEFT 3 · · · · 1 RIGHT
LEFT 2 · · · · 2 RIGHT
LEFT 1 · · · · 3 RIGHT
LEFT 2 · · · · 2 RIGHT
BOTTOM: 2 1 2 2
The task here is to explain each elimination, rather than copy a completed grid. Leave row 4 undecided while working through the other crossings. You can enter values with the visible height buttons or select a cell and type its digit. Check towers diagnoses the current entries; it does not search for a future completion.
Enumerate both views before choosing a row
Each full line is one of 4! = 24 permutations of heights 1–4. For a left clue of 2 and a right clue of 2, height 4 cannot stand at either edge: that edge would see only one tower. Put 4 second or third, then test the remaining order from both sides. Exactly these six lines survive:
1 4 2 32 1 4 32 4 1 33 1 4 23 2 4 13 4 1 2
This list is exhaustive, not a set of suggestions. When 4 is second, the last two heights must increase from left to right; otherwise the right edge would see three towers. When 4 is third, the first two heights must decrease; otherwise the left edge would see three. Each case gives three permutations. For example, 3 1 4 2 shows 3 then 4 from the left, and 2 then 4 from the right. It fits the two clues but is not automatically a valid row in this particular grid.
Use one fixed column as a filter
Column 2 has top clue 4 and bottom clue 1. Seeing all four heights from the top forces the column to read 1 2 3 4 downward. Enter R1C2=1, R2C2=2, R3C2=3 and R4C2=4. This is four edits with twelve blank cells. Those values let us filter whole row permutations without committing to a convenient-looking row.
Row 1 has views 3 from the left and 1 from the right. Its complete list before using column 2 is 1 3 2 4, 2 1 3 4, and 2 3 1 4. Only the middle candidate has 1 in column 2, so row 1 must be 2 1 3 4. Enter the three remaining cells in that row.
Row 2 has the six 2/2 candidates above. Only 3 2 4 1 has 2 in the second position. That fixes R2C1=3, R2C3=4 and R2C4=1. Notice the method: a crossing value removes complete alternatives; you are not simply putting the largest available height into a cell.
Row 3 has views 1/3. Its three candidates are 4 1 3 2, 4 2 3 1, and 4 3 1 2. Column 2 requires 3, leaving 4 3 1 2. After entering its other three cells, thirteen entries are fixed and only row 4 is unfinished.
Keep row 4's three options until the crossings decide
Do not confuse the six initial 2/2 candidates with the three that remain after R4C2=4. At that point row 4 could locally be 1 4 2 3, 2 4 1 3, or 3 4 1 2. Matching the second position alone does not distinguish them.
Now look at column 1 above it: heights 2, 3 and 4 are already used. Row 4 therefore needs R4C1=1, choosing 1 4 2 3. The second option would repeat 2 in column 1 and 1 in column 3. The third would repeat 3 in column 1, 1 in column 3 and 2 in column 4. The correct line is determined by the crossings, even though all three lines independently satisfy the row's two outside clues.
A locally clean branch can already be impossible
To reproduce the counterexample, choose Retry this board and enter only column 2 as 1 2 3 4. Then set R4C1=2, R4C3=1 and R4C4=3. You have made seven edits, not entered the other forced rows. The partial grid is:
R1: · 1 · · R2: · 2 · · R3: · 3 · · R4: 2 4 1 3
Before selecting Check towers, predict the blank count, repeated-row count, repeated-column count and number of mismatched complete-line clues. Also decide whether this partial grid can be completed.
Counterexample answer: no immediate error is not a completion proof
The diagnostic counts are nine blanks, zero rows with repeats, zero columns with repeats, and zero mismatched complete-line clues. Row 4 and column 2 are the only complete lines, and both match their two views. The board stays editable and does not win. Yet there are zero complete solutions extending these entries: row 1 is still forced to 2 1 3 4 by its views and second-position 1, which conflicts with the 2 already placed at R4C1. The row-3 crossing would produce another conflict with R4C3=1.
The checker only tests duplicates and complete-line visibility in the current board. It does not promise that an unfinished, locally clean board has an extension. Retry removes this experiment on unchanged clues. Reapply the candidate deductions instead of treating the zero error counts as permission to keep the branch.
Use the edit checkpoints to audit your solve
For a clean replay, load or retry the strategy board and follow the deductions in this order. Select a cell before using a height button. Entries already present in column 2 do not need to be entered again.
- Fill column 2: four edits, twelve blanks.
- Complete row 1: seven edits, nine blanks.
- Complete row 2: ten edits, six blanks.
- Complete row 3: thirteen edits, three blanks.
- Complete row 4: sixteen edits, no blanks.
Full answer and outside-view audit
R1: 2 1 3 4 R2: 3 2 4 1 R3: 4 3 1 2 R4: 1 4 2 3
Left views are 3,2,1,2 and right views are 1,2,3,2. Top views are 3,4,2,1 and bottom views are 2,1,2,2. Each row and column uses every height once. Filling the last cell is not an automatic win: use Check towers to verify and lock the answer. Undo last edit reopens a checked win with fifteen entries; Retry clears all sixteen entries on the same clues. A larger edit count is valid too; sixteen is this replay's count, not a required minimum score.
What this strategy proves, and where it stops
The candidate method is sound when you retain every ordering compatible with the clues and known crossings. A value shared by every surviving candidate is forced. Choosing the first valid candidate is not sound. On this board, filtering by column 2 and then the completed crossings solves everything; harder layouts can leave several candidates in every line. That alone proves neither impossibility nor multiple solutions. Further deductions or a carefully tracked assumption may be needed. The unique answer here was separately checked against all 576 four-by-four Latin grids; uniqueness does not make every locally valid move safe.
As a small self-check, ask why row 4 initially has six candidates, then three after the column-2 anchor, then one after the other rows are fixed. If you cannot name the constraint responsible for each reduction, revisit that step before trying a different puzzle. Load strategy board restores this same arrangement; New puzzle moves on to the next prepared layout, and refreshing the strategy address reloads the clues without preserving entries.
Rule reference
[1] Simon Tatham, Portable Puzzle Collection, Towers manual, consulted November 14, 2026. It supplies the common visibility rules. This board, the candidate analysis, counterexample, implementation and explanatory text are original; no affiliation is claimed.