# "Bees sting." = [1, 2, 3, 7, 8, 9, 5, 6, 4]

If

"Bill mops." = [1, 2, 3, 4, 5, 6, 7, 8]

and

"Bees sting." = [1, 2, 3, 7, 8, 9, 5, 6, 4]

then

"These make up words like feathers make birds." = [?, ?, ?, ?, ?, ?, 1, ?, ?, ?, ?, 37, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?]

and

"For persons ???'?? reshelving ?????, this ????? helps them ???? ???? homes." = [13, 35, 43, 41, 3, 44, 49, 36, 32, 50, 61, 16, 37, 45, 4, 46, 5, 51, 17, 6, 26, 60, 23, 33, 15, 56, 38, 28, 7, 52, 57, 18, 24, 53, 39, 47, 1, 8, 48, 19, 9, 27, 42, 54, 58, 20, 10, 29, 14, 25, 34, 2, 59, 21, 11, 30, 22, 40, 31, 12, 55]

and

"??????? ??-????? ????? ????? ??? ????? ????? ?????? ???? ?? ??????." = [55, 52, 17, 32, 43, 20, 7, 11, 51, 50, 21, 49, 12, 38, 44, 37, 47, 39, 45, 36, 48, 22, 35, 40, 30, 33, 46, 29, 23, 27, 26, 53, 25, 24, 19, 1, 9, 18, 2, 15, 16, 13, 41, 10, 34, 31, 14, 5, 54, 8, 3, 6, 4, 28, 42]

(a nonsensical but grammatically all right sentence designed to make this easy to solve)

and finally

"????? ????? ??? ??? ??? ?????? ????? ???? ??????? ?? ??????." = [49, 8, 22, 6, 1, 25, 17, 12, 14, 39, 11, 18, 45, 40, 15, 9, 20, 30, 4, 13, 36, 31, 44, 41, 16, 34, 23, 2, 27, 38, 43, 10, 37, 46, 35, 42, 19, 5, 21, 24, 47, 32, 28, 26, 33, 29, 7, 3, 48].

(a more sensical but harder-to-solve sentence that appears somewhere on this page: http://clagnut.com/blog/2380/ . Visit it for a hint as to what kind of sentence it is. (The preceding nonsensical one is the same kind of sentence.))

This puzzle works by

labelling all the as in the sentence sequentially, then all the bs, then all the cs, and so on. In other words, the numbers tell you the position of their corresponding letter if you sorted all the letters alphabetically.

You can figure out the question-marked letters by

using the given letters to bound their ranges in the first one; for the others, using the fact that they are pangrams lets you place uncommon letters in their likely positions.

To break into the second, I saw that ZY must be the starting two letters. The last letter being very early on lets you determine the word. And for the last, it was easy to just find the sentence on the linked page given the letter pattern.

The solutions are:

[34, 15, 7, 30, 8, 23, 1, 19, 9, 36, 26, 37, 25, 27, 5, 31, 22, 17, 20, 10, 14, 11, 2, 35, 16, 12, 28, 32, 24, 3, 21, 13, 4, 18, 29, 6, 33]
"For persons who're reshelving tomes, this order helps them find them homes."
"Zygotic ex-wives trust quips, not milky jihad gaffes done by cabals."
"Zelda might fix the job growth plans very quickly on Monday."

• This is excellent, and of course all correct; but I do feel a little like you skirted the last puzzle, or peeked at the answer, rather than solving it. :) Sep 25 at 17:10

Here's how I solved the last sentence, with some notes about puzzle design.

I realized that when deducing the text from the numbers, if you scan through the numbers in numerical order, you know for certain that every time you have to go backwards, you have a letter-boundary. (In the last sentence, there is such a boundary between 3 and 4, for example.) For the penultimate sentence, as a gentle introduction, I wanted to create a sentence that had 25 definite word boundaries -- because then every letter of the alphabet could just be slotted in. So I wrote a pangram, but quickly realized that this didn't guarantee explicit boundaries: if your first "z" comes after your last "y", then no reversal of direction occurs. So I set out to write a pangram as much as possible in reverse alphabetical order, so that the first instance of every letter appeared before the last instance of its predecessor. Thus "Zygotic ex-wives trust quips, not milky jihad gaffes done by cabals."

For something a little harder, I took a batch of pangrams from the cited webpage, wrote a little program to select one at random (so I could try to solve it without having seen it), convert it to its "sortpher" or sorted-cipher, and count how many letter-boundaries appeared. I chose to work on one with the highest number of boundaries: 20. That meant (since I knew it was a pangram) there were 5 more boundaries to place, and since these could go in any 5 of 28 remaining positions, there were about 98,000 possibilities (I think). That seemed a few too many to laboriously scan through, so instead I wrote a little program to place those five boundaries at random, then show the solution using the alphabet so divided. This gave me results like these:

 zflea njhiv gjy vif kpc htpxvi rmbou wfty svjdkmy qo nqoeby
zfnea ojhiv gky vif lrd htrxvi snbpu wfty svkdmny rp orqecy
zfnda olhju glx ukf mqc isqwuk rnapt vfsx rulcmnx qp oqpeby
zdkca mhfgt ehx ugd job frowug pkans vdrx quibjlx on moncay
and eventually, after only a few tries:
zdlca might fjx the kob growuh qlans very qujbkly pn mpncay
From there it was pretty easy to work out the answer.