Or he might read HN, and see that your preferred passphrase method is composed of english sentences with common substitutions.
Maybe a website you're using leaks its password DB, and your "padding method" is exposed, reducing the search space back to what it was before the padding. I'm going to have to disagree with GRC here. "D0g.........." is a very bad password, especially if you use ".........." as your padding on every website.
Even without this, common substitutions like '3' for 'e' or '!' or 'i' are accounted for in everything but the most naive cracking tools.
Perhaps less likely, a timing attack in the auth system exposes the first few characters of your password, but not the rest.
Why put up with these issues, when something like xkcd's 4 random words works well, is easy to remember and type, and calculates entropy in a uncontroversial way?
(http://world.std.com/~reinhold/diceware.html)
And substituting some symbols for letters doesn't add much security. Depending on the mode of attack there are precomputed sentences with a variety of substitution.
Be careful with GRC. Some people don't have a high opinion of his analyses:
How would an attacker know what the first letter is? Yes an attacker can guess if sentence passwords are used that (a|A),(the|The),and (i|I) would be very common starters but after that the bigger search space complicates things.
https://www.grc.com/haystack.htm
Read the sections on entropy and padding. Just using sentences alone does present attackers the ability to use an encyclopedia attack, but a few randomly placed symbols (padding) turns the endeavor back in to an exhaustive search.