Choosing colours "without baggage" for item comparison

I am sketching an interface that allows people to compare multiple items side by side.

The interface implements the following requirements:

• users can add new items to the comparison
• the differences between them will be highlighted
• there should be a visualization that maps the ingredients of an item to a consumer

The comparison algorithm can handle N items, so if one has a wide screen - they're welcome to add as many items as they please. Realistically, I expect that people will usually compare up to 3 or 4 items.

To link the items on the left, with their ingredient-consumper maps on the right, I use colour, which brings me to my questions about choosing a colour-scheme:

1. what scheme is best suited for conveying qualitative data?

The items are not ranked from "good" to "bad" or "healthy" to "unhealthy", they're just different items.

2. what scheme is not "preloaded" with meaning?

For example, in this illustration Ipsum is green, which is associated with "good", "safe" or otherwise positive in most cultures. This might lead users towards wrong interpretations.

3. how to best deal with the addition of new items, colour-wise?

If I use a triadic scheme for maximum distance between 3 colours, I'd have to switch to a tetradic scheme when a 4th item is added - which would change all the previous colours. I thought of choosing a qualitative colour-scheme from Colorbrewer for 7 items, and using those colours only. This avoids the reshuffling of colours, but inevitably, some of them will be green/positive, others will be red/negative.

My feeling is that there are conflicting requirements, and perhaps I am asking the wrong question - maybe using colours here is already a mistake, and maybe the diagram on the right is also not the most suitable one. Thus, besides asking for answers to my three questions, I also ask for advice on how to tackle this, thank you!

UPDATES

I do not weave them into the original question, so you can easily follow what has been added.

• The colours are there to serve as a legend for the graph. For example, you see that Lorem is blue, so you follow the blue lines on the right side to see how this product shares ingredients with consumers.
• The domain is data privacy, though I removed every hint of that from the picture, assuming that an optimal solution would be domain-agnostic. So let me reveal the cards: you want to buy a "smart" device, and you need to know what data about you (ingredients) it shares with other parties (consumers).
• The comparison is as Mike described, several items having mostly identical features, but with different values (could be null too).

• Here are some examples of conclusions I expect users to reach, when examining this screen:

  • A lot of orange arrows -> "Magicum is sharing my data with a hell of a lot of companies!"
  • A lot of arrows go into "Dacicus" -> "Hmm, this company appears to be concentrating a lot of data in its hands"
  • Only 3 blue lines -> "It seems that Lorem is not that "big mouthed", and are only sharing data with a few entities".

Note that I already have an actual table comparison in another tab:

Thus, the purpose of the graphical view on the right (which I now consider moving out to a separate tab) is to facilitate the construction of the big picture, and notice any patterns that would otherwise go unnoticed, if representing the information as prose or as a table.

I have considered using a chord diagram or a Sankey diagram for this purpose, but there is a drawback when it comes to grouping. They would be suitable if I only had to group by origin (i.e. one of {lorem, ipsum, magicum}) XOR by item type ({eggs, spam, bacon...}), but not by both. The current visualization deals with that, the item type is the circle-dot on the left, whereas the arrow colour indicates the origin.