Suspicious indeed since it repeats in various places. The only match I could find online of "The journey inward begins here" is in an Amazon review of the album "A Passage In Time" by "Dead Can Dance":
http://www.amazon.com/Passage-Time-Dead-Can-Dance/product-reviews/B000S5DY00?pageNumber=2

Perhaps we need to use track titles, lyrics or the review text somehow, e.g. as decryption key?

Or cross-reference track titles or lyrics with the other hints, e.g. with “The dragon guards the key.”?

TL;DR: me rambling on possible directions for the Aztec code. Conclusion: try a lot of things randomly and hope the ECC saves us.

HOW WAS THIS AZTEC CODE POTENTIALLY GENERATED?

Suppose that the resulting code can be generated either by
A) rotating the Aztec code 180 degrees, and then overlaying it on the original;
B) rotating the Aztec code 180 degrees, and setting the result using any commutative logical operation (i.e. f(x,y)=f(y,x)) to get a symmetric result (there are 8 possible truth tables that satisfy that requirement);
C) taking half the original Aztec code and adding it's mirror to get a complete one.
D) other?

option A
based on the missing orientation marks, we can cross off option A, or we would have seen them in the result.

option B
the 8 possible truth tables are
* the resulting code is not fully white or fully black codes, so possibilities a and h can be ignored.
* overlaying the bottom right orientation marker on the top left orientation marker (https://en.wikipedia.org/wiki/Aztec_Code#/media/File:Aztec_Code_with_desc.png) means 11->1, given the result (same for the bulls-eye on top of the bulls-eye), hence possibilities c, e and g can be ignored.

This leaves
* the top right orientation marker on top of the bottom left orientation marker means 10->0, 01->0, 00->0, given the resulting code. Therefore, possibilities d and f can be dismissed.

This leaves only
* This means that in the resulting code only those pixels that have a mirrored black pixel remain visible. However, that would mean that the top left orientation code would disappear, except for one pixel....which it did not. Hence possibility b can also be dismissed.

Conclusion: no commutative function was used to generate a symmetric Aztec code from the original Aztec code and the original aztec code rotated 180 degrees.

option C
This option is possible if you take a subset of the pixels (not necessarily half) of the original Aztec code, and then overlay a 180 degree rotated version of that. This is the option analyzed further below.

option D
Not sure what other ways there would be to generate a symmetric code. Or perhaps it's not an Aztec code in the first place... 


REMOVE THE SYMMETRY, BY REMOVING ONE HALF?

Ignoring the 2 extra pixels and the bulls-eye, there is 65 pairs of mirrored pixels and therefore 2^65 ways of removing the symmetry and removing 1 pixel from each pair, and 3^65 possibilities if we also consider the option of keeping both pixels of the pair. Therefore, it's not feasible to go through all possibilities unless a big percentage of those possibilities decode successfully due to the error correction capability of the Aztec code.

Therefore, if removing the symmetry and reading out the result is the right approach, we try assuming a few things to limit the possibilities:

1) that the bulls-eye remains intact

2) a more structured way to leave out half the pixels by requiring that the halves are contiguous areas of pixels (with one of the possibilities given above). Any path from the bulls-eye to the edge of the code, together with its mirrored path defines two such halves that are mirrors of each other.

3) we can further try assuming that the mirrored half contains two correct orientation markers (as shown in https://en.wikipedia.org/wiki/Aztec_Code#/media/File:Aztec_Code_with_desc.png). Therefore, it must be a half that contains the top-left and bottom-left orientation marker (i.e. the top-right and bottom-right orientation markers are wrong in the Aztec code) that is picked and rotated.

4) the orientation marks of the missing half can be filled in correctly

However, even with these assumptions there are still many ways to do split up the code in two halves and remove one of the halves. We might get lucky though if the error correction level of the original Aztec code is high enough and the exact path does not really matter. Either way, it is probably something we would want to do programmatically.

Alternatively to removing one half completely (which corresponds to flipping all the 1s to 0s in that half), we could randomly guess the missing half of the code and hope the ECC will allow us to read the code .


REMOVE THE SYMMETRY, BY EXPLORING CODES WITHIN A CERTAIN EDIT DISTANCE?

Finally, in case we don't make the assumption of the halves being contiguous areas, we could still explore randomly the codes within a certain edit distance of the given code, and hope the ECC will allow us to read the code successfully. So, assume a correct bulls-eye with correct positioning markers, randomly flip up to n pixels in the given code, and attempt to read the result.


ARE WE MISSING SOMETHING?

Still:

- the 2 non-symmetric pixels are really bothering me
- we have not used the markings in any way yet.

So, are we missing something here? Did we miss a clue? Or are we just failing to see the forest for the trees?

Another observation about the QR codes, after going through http://www.thonky.com/qr-code-tutorial/data-masking


That said, it could just be a coincidence of course.

Great summary mirth23! A few things to add:

* The phone substitution made me realize that 'space' could be a possible substitution as well.

* A correction to QR code 11: http://textadventures.co.uk/games/view/5craldfdmkkzngf1davsna/generation-a-days-destroyed
'Generation A' is a book by Douglas Coupland (https://en.wikipedia.org/wiki/Generation_A) and according to that page, the name 'Generation A' is based on a commencement speech by Kurt Vonnegut (http://versailles1.tripod.com/syracuse.html). No idea yet how this relates to 'days destroyed' or 'bitcoin days destroyed'.

* note the different quotes around "Mechanica Mathematica". This indicates that in case we try doing something with the quoted parts of the text, this should not be treated in the same way as the actual quotes.

* 181 aztec pixels corresponds roughly to the length of a bitcoin address in binary (?)

* The hi-res image seems to be a scan of the artwork. The low-res map image from the text adventure seems to be a photo of the artwork hanging from a wall (see the shadows). The different lighting makes it easier to distinguish some of the red squares in the low-res version than in the hi-res version.


* Good point! I'd agree that it means something, and would guess that it indicates that for level 1, we need to use the QR codes.

* Besides 4 or 6 gold dots, there are the 4 possible orientations (=possible encoding of 2 bits) for the QR codes, and potential mirroring (= possible encoding of 1 bit). Together with the dots, that gives 4 potential bits encoded per QR code.

* I've been trying to recreate the QR-codes after reading them out, with varying success: some are identical, some are partly identical, some are different altogether. My thinking was that with the error correction level L in the QR codes it would be possible to flip up to 7% of the bits and still read it out successfully. Those flipped bits could potentially be used to encode letters or something else directly into the pixels of the QR code. These could then be recovered by taking the difference of the pristine and the altered QR code.

Here is a possible segmentation of the image by content:
http://imgur.com/iylMHIc

This animation illustrates possible depths of the image content and might help to see which pipe goes where. (here it assumes QR on top of AZTEC CODE AND DARK PIPES on top of YELLOWISH PIPES on top of LIGHT PIPES on top of RED GRID)
http://imgur.com/V5Gfo3I

GIMP file with image content split out to different layers:
https://mega.nz/#!6gYB2K7D!7WD3zHx0RfQ0y3po6Ij3SYSAWV7ATDkJkX9AerybMOI

Disclaimer: another red-green colorblind person here, so do not blindly trust this segmentation. It probably needs corrections and needs to be split out further still.

Stuck and unsure which direction to go at this point and what parts of the available information are relevant for this stage (no luck with any of the approaches so far):

- use content of the qr codes?
- anagrams?
- use the markings?
- use the pipes?
- decode the aztec as something else? (since no proper orientation markers, and apparent improbable symmetry suggesting it might not be an aztec code)
- select appropriate pixels from the aztec using markings or pipes, and decode entire code? (the aztec core does not contain proper orientation markers though)
- select appropriate pixels from the aztec using markings or pipes, and decode part of the code? (the aztec core mode message indicates 2 layers and 7 data codewords, so presumably the keypad code is <= 7 characters)

Any opinions or hints to narrow it down?

Checked it again, and you're probably right. It's really hard to see and be sure, so I'll change it back to 2b and will add a note:

I like sheets! Added a tab to your sheet where you can easily switch between multiple substitutions.

working on a new direction now: anagrams of letters from the QR messages  (e.g. first or last letters of message or quote within message), using:

http://anagram-solver.net/ (note the option for anagrams using only part of the letters)
https://www.arrak.fi/en/ag/

a few nice hits, so far:

Thanks for corrections! Here is the updated grid (with the 6 corrections so far indicated by underscores):

Us assigning letters to the triangles is arbitrary, as is the ordering of the triangles (i.e. we can assign 1,2,3 to a,b,d in 6 different ways). Assuming a, b, and d are interchangeable, that leaves 6 ways to assign A-Z to the pairs like:


For frequency analysis, none of this matters though, since each of the pairs is replaced by a single letter. In that case, only the mapping between pairs and letters matters, as well as the order in which the result is read from the grid (left-right, up-down, spiral, etc). So, if each pair stands for a single unique letter, we should be able to break it by reading out the pairs from the grid in some logical way, and then attack the result as a substitution cipher with frequency analysis.  Using the above substitution:


Haven't had any luck attacking this as a substitution cipher. Frequency analysis will fail though if we are using the wrong way to read out the pairs from the grid, or if we only need a subset of them (e.g. only the pairs that are attached to the dark green pipes).

Another direction would be that the aztec code is the result of a superposition of multiple aztec codes, and it can be demultiplexed by the markings. However, then the whole core of the aztec code should have the same marking.

A final thought on "the journey inward": the map seems to consist of a number of layers on top of each other: QR on top of AZTEC on top of DARK GREEN PIPES on top of LIGHT GREEN PIPES on top of a RED GRID. So maybe Level 1 of Days Destroyed would only need the QR, level 2 only the AZTEC, and so on.

Here is a corrected version of the earlier posted grid (4 corrections marked with underscores):

Some thoughts for directions to decode the aztec code that might not be an aztec code:

• interpret the empty spaces instead: “When in your house black crows give birth to white doves, then you will be called wise”?
• interpret blocks of it as bitstrings and then convert?
• interpret blocks of it as braille?
• interpret blocks of it as some other script, like http://www.omniglot.com/conscripts/ancients.htm ?

None of those really account for the 2 pixels that don't match the symmetry though, nor does it use the markings...