Look along the four straight directions
The rule checks the source’s own row and column. A diagonal resident is not in that scope merely because it looks close on the board. Likewise, a direction is not an exact address: several addresses may lie along the relevant line.
Start by tracing the row and column from the Blood Hound’s position. Compare only candidates on those lines using the rule’s distance condition. Do not let an unrelated nearby diagonal candidate change your reading of the announcement.
Barking and sniffing answer different questions
In an illustrative grid with a reliable normally good Blood Hound in the center, put one eligible evil one space above and another one space to the right. They are equally near, so the Demo tie rule calls for barking. The sound does not name one of those addresses or tell you the total number of evil roles elsewhere.
Sniffing is the no-eligible-target result. Under a reliable normal clue model, it says the checked row/column has no found evil target; it does not clear the entire board. If you are unsure whether the source is reliable or whether a special role changes counting, preserve that uncertainty.
Turn a direction into a smaller question
If the direction points right, compare the surviving addresses to the right in the source’s row. A further clue about one of those addresses can distinguish them. If there is a tie, choose information that separates the tied candidates rather than assuming the first visible card is the answer.
An evil truthful source, such as a Conman disguise, creates a particular self-counting question not clearly resolved by the read materials. This page does not assert the output for that case. Keep it separate from the normal scope and tie rules above.