Other tools also... https://www.grc.com/haystack.htm http://www.passwordmeter.com/
For the passphrase, 30 characters chosen from 73 is 185.7 bits (log(2,73)* 30). Even this isn't correct though, because the characters aren't randomly chosen. As the author says: "Not so for human-created passwords, where a sea of associations and language rules greatly reduces the search space." In other words, knowing that the first word is "I" limits the possibilities for the next word, and so on.
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.
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:
The actual randomness of such a password is a factor when using some intelligent means to walk the keys pace, such as Markov chains.
While many sites would think this is a good password "x5tRwz3m" Length: 8 Strength: Weak. Entropy: 36 bits. Charset Size: 62 characters Is impossible for me to remember. A easy to remember mixed (seemly, but not random) password like that is much easier to generate and use in a off-line dictionary attack.
"I like Microsoft's login page." Length: 30 Strength: Very Strong. Entropy: 147.3 bits Charset Size: 73 characters
is a sentence and people are pretty good at remembering them.
Add in a tiny bit of complexity "I like M?icrosoft's login p!age."
Length: 32 Strength: Very Strong. Entropy: 163.5 bits. Charset Size: 83 characters.
It is exceptionally unlikely that a cracking attempt will reveal that password and that some other type of flaw will be needed to reveal the password or gain access to the account.