The dots in between the squares...
... are in groups of three. Taking top/left = 2, middle = 1, bottom/right = 0, reading as ternary, and then converting A=001, B=002..., gives the following:
top row: SNAKE DOCK BOAT
middle row: POT NAKED BURNS
bottom row: CATCH UPSET TEN
left column: ROLL IN JOB UNIT
middle column: METAL OR ENERGY
right column: BUG BOTCHED TAN
These can be interpreted
as cryptic clues. Each leads to a three-letter word: BOA(-t), _URN_, NET<, _BUN_, OR+E, ANT*. And these words are a word square, with the words placed in the locations of their clues.
BOA
URN
NET
There are also some other ways to extract similar arrangements of letters:
The nine large squares each have the same color as exactly one of the colored triplets of pixels. We can either extract the letter from that triplet, or assign the triplets A-Z (since there are 26 of them).
The nine large squares also each are missing a pixel - this is always one of the first 26 pixels. So this gives another 3x3 square of letters.
The data collected:
CCs is the square from the cryptic clues; A-Z is taking the corresponding triplets' positions; dec is taking the corresponding triplets' letters; gap is the position of the hole in the large square.
CCs A-Z dec gap
--- --- --- ---
BOA YLN NAP DTQ
URN HOL COA XBP
NET ILA KAS AMJ
The next step needed heavy hinting for me to find it:
If you use the gap position to Caesar-shift the letters from the word square, you get FIRSTDORD. This is likely a reference to "Dord", a fake dictionary word accidentally added when a note asking to add "D or d" to mean density was misinterpreted. So the answer phrase may just be FIRST DENSITY?
Unused information [hinted to be a clue to the step I did first]:
It seems pretty clear that this is the data that needs to be extracted - I'm just not sure how to get an actual answer out of it.
The only remaining information in the puzzle is the particular colors used. They don't seem to be notable in any way - some are fully saturated, some not, and the RGB values used don't seem to have any particular pattern.
Of the second and third squares (both using the colors), only one is probably intended. I'm not sure which, though - both look equally plausible as letter sets.
It's strange that one color is repeated three times - this makes me think that the "color triplet position" rather than "color triplet decoded" 3x3 is intended (because there were multiple other A's that could have been chosen from the decoded text).
I've used all the tags, except possibly steganography and anagram (depending on how you interpret those - do the color matching or the anagram in the cryptic clue count?).
Other uses of steganography: There are no slightly-off-color pixels - all the large regions really are a single color each.
Other uses of anagram: I've tried anagramming all four of the squares I found - nothing interesting appears. I've also tried anagramming the letters in the same position in each square, and again, nothing.
