Algorithm Overview

A  compressed "library file" of prime numbers is created and retained.
It allows any-time rapid access to any of 203,280,221 primes (all that fit in 32 bits).

Background Information
The arithmetic process of division often yields a fraction that can be expressed as a repeating
series of digits.  (If "repeating series" includes "000000...", then that is always true.)  A repeating
series often contains different digits, such as "423432432..." or even "115115115..." -- both of
those sequences has a "period" of 3 digits-that-repeat.  If the divisor happens to be a prime
number P, the length of the period is often P-1 digits-that-repeat.  (When different bases than
the common Base Ten are used, every prime divisor P has in some Base a period of length
P-1.  Meanwhile, no composite divisor C ever has in any Base a period of length C-1.)  This
means that a large prime divisor -- for example greater than 100,000,000 -- can easily generate
100,000,000 digits before the sequence begins to repeat.  While the digits within that sequence
do not quite qualify as being "pseudorandom", they are irregular enough that if combined with
the sequences of other prime divisors, the result can qualify as being pseudorandom enough
for encryption purposes.  This division and combination process is also computationally simple
and fast., and the more sequences are involved, the higher-level the encryption will be.

Part One: Bootstrap Preparation
A "key file" of any type (graphic, executable, audio, zipped, etc.) is chosen by the user.
It is opened in Binary Mode for reading only.
The key file is given a "minimal randomness" test.
If 90 different values can be derived via "exclusive-or" (XOR), the key file is not rejected.
The 90 different values are pseudorandomized, per defaults built into the program.
(See Part Two for a description of the pseudorandomizer.)
Some of the resulting values are used to select Nth primes from the library.
(If N is greater than 203,280,221, it is halved until it is small enough for the library.)
The selected primes replace the four default values built into the program.
Next, 8 data bytes at a time from the key file are pseudorandomized.
Pairs of the scrambled bytes are combined using OR (not XOR).
They make a 32-bit N to select a (biased large) prime from the library.
The  longer the key file, the more primes can be chosen, up to 65,536 of them.
Duplicate primes are mostly prevented (allowed among first 1/50 of selected primes).
The primes are placed within an array of "division scratchpads" in the memory.
They are left in the array as placed, in unsorted order.
(Examination of the source code will reveal a sorting section.)
(It exists to pull many primes from the library fast and efficiently.)
(The primes will still end up out-of-order in the main scratchpad array.)
Division by one of the initial 4 primes yields 8 more OR'd bytes.
They become a scratchpad "dividend" to be divided by a prime.
The minimum key file length is 8300 bytes (for approx. 1000 primes).
Toward the end of the key file, 108 bytes are reserved.
A minimum-length 8300-byte file can yield 1024 primes if no duplicates occur.
When as many scratchpads are filled as the file-length allows,
the 108 reserved bytes are pseudorandomized and used to
again replace the 4 initial primes.

File encryption or decryption may now begin.

Part Two: The Pseudorandomizer
As indicated above, everything depends on divisions performed in "scratchpads".
In Base Two, division consists of doubling the dividend and subtracting the divisor.
A one-bit output results if the bit-shifted (doubled) dividend is larger than the divisor.
A zero-bit output results if the dividend is smaller than the divisor.
The result of the subtraction becomes the dividend for the next division operation.
A "divide" subfunction exists to obtain N quotient bits from one scratchpad.
The divide function can operate on any (pseudorandomly) specified scratchpad.
The first time a scratchpad is divided, a "random" dividend may be larger than divisor.
If so, the dividend is halved multiple times, until it becomes smaller than the divisor.
Then the doubling (via bit-shift) and subtraction process begins, for N quotient bits.
Every scratchpad retains its last dividend result for use when called-upon again.
(The saved dividend is always smaller than the divisor, so initial halving never recurs.)
At the start of the process, each byte we want to scramble is copied to a holding variable.
The 4 initial primes built into the program allow some divisions to occur "at once".
These scratchpads control the combining of quotient-data for pseudorandomization.
One specifies how many times an overall scrambling loop will be performed on our byte.
Two data bits from it yield 0,1,2, or 3 -- modified to cause 1,2,3, or 2 overall loops.
Inside each overall loop, another initial scratchpad provides two more bits of data.
Values 0,1,2, or 3 are modified to cause 2,3,4, or 5 individual scrambling operations.
For each operation, the third initial scratchpad provides 16 bits of data.
The quantity N of scratchpads in the main array is now an important factor.
If N is zero, then 8 bits of those 16 are going to be used as scramble data.
Otherwise Modulus N of the 16-bit data yields a main-array index value.
(That is, as soon as any main-array scratchpads are filled in, they start getting used.)
The scratchpad at that index is tapped to provide 8 bits of scramble data.
For each scramble operation, the fourth initial scratchpad provides 3 bits of data.
The complete group of scrambling-data is saved in two local arrays.
The "Encryption" process pulls/uses the data from these arrays first-to-last.
The "Decryption" process pulls/uses the data from these arrays last-to-first.
At the start of each actual-scrambling operation,  Modulus 7 of the 8-bit data is saved.
For the "Encryption" process:
The 3 bits of data specify one of the following scrambling operations to our data-byte:
0 - Use the saved Modulus 7 number to perform 1,2,3,4,5,6 or 7 LEFT bit-rotations.
1 - Same as 0, but also perform a NOT (flip all bits) operation afterward.
2 - Add the 8-bit scramble-data.  (If result exceeds 255, byte "odometerizes".)
3 - Add the 8-bit scramble-data, and then perform a NOT operation.
4 - Subtract the 8-bit scramble-data.  (Yes, this can odometerize backwards.)
5 - Subtract and then perform NOT.
6 - Combine the 8-bit scramble-data using XOR.
7 - Combine using XOR, and follow with NOT.
For the "Decryption" process:
The 3 bits of data specify one of the following scrambling operations to our data-byte:
0 - Use the saved Modulus 7 number to perform 1,2,3,4,5,6 or 7 RIGHT bit-rotations.
1 - Same as 0, except also perform a NOT (flip all bits) operation before the rotation.
2 - Subtract the 8-bit scramble-data.
3 - Perform a NOT operation, then subtract the 8-bit scramble-data.
4 - Add the 8-bit scramble-data.
5 - Perform NOT, and then Add.
6 - Combine the 8-bit scramble-data using XOR.
7 - Perform a NOT, then combine using XOR.
As previously indicated, every data byte receives two-to-five scrambling operations.
Only very rarely will a single scratchpad provide two consecutively-used scramble bytes.
This makes it very difficult to deduce what dividends/primes yielded scrambling-data.
There is also no clue as to how many scratchpads were used, from one-to-65 thousand.
(And how many different ways can one select thousands of primes from 200 million?)
All those different ways to multiply-scramble a byte don't help the cryptanalyst, either.
The overall scrambling loop is now done for the first time.
If that loop is about to repeat, then the holding variable is used to restore our data byte.
Only the last overall scrambling loop actually provides the scrambled byte that we save.
The deliberate purpose:  It breaks up the actually-used sequence of quotient-bits.
About half of all quotient-bits are pseudorandomly "wasted".
It becomes even more difficult to determine the dividends/primes used to encrypt data.

When thousands of bytes are to be encrypted, every scratchpad in the main array is
probably going to be pseudorandomly selected to help provide scramble-bytes.  It is
currently thought that an encrypted file will provide nearly zero insight into what key
file data was used to bootstrap itself to provide the primes that were used to encrypt
the file.  Thus the likeliest way to break this encryption scheme appears to be the
brute-force method of trying all possible key files.  That is where the "Extra Randomizer"
mentioned elsewhere comes into play.  It allows the user to specify that up to ten million
scramble-events will take place, wastefully, before any are actually used to encrypt data.
This means that every possible key file must be tested up to ten million times, before
the next key file can be tested.  And that can work only if the encrypted file was the first
one specified to be encrypted by "cryption.exe".  Instead it could have easily been the
second or the hundredth file of a single session; it could be multiply encrypted in a
single session (or in separate sessions with different key files), and equally large
Extra Randomizers could have been specified in-between every file-to-encrypt of a
single session.  Finally, there may be more than a trillion files accessible via the
Internet, to be used as keys.  Enjoy!  Because so far, it looks like the cryptanalysts
probably won't.