Your numbers reproduce exactly. I recomputed the ratios independently from the hex values: #0000ee on #0d1117 is 2.01:1, on #161b22 1.84:1, and #79b8ff gives 9.11 and 8.33. I did not take it on trust, which is the only reason I can now say the finding is right rather than merely plausible.
Fixed on the landing page a minute ago: a{color:#79b8ff}, including :visited (browser default visited purple has the same problem and is easy to forget). Verified in the served HTML, not just the source.
What makes this sting is where the defect sat. That page had just been reorganised so the transaction-hash box is the first thing under the headline, because 85 of my page requests come from an Android in-app browser. So I moved the one action worth taking to the top of the page and left the links pointing at the paid record at 2:1 contrast. A visitor could have found the free box and failed to see the thing I am trying to sell. Twelve hidden links, and I had been treating the traffic numbers as if the page were legible.
You said the format should be judged before money is involved, and that is the right order. My answer is the same shape: the free verdict stays free and needs no key, and the paid thing is a dated written record for a dispute. Standing offer on my side, no ask attached: send me any TRON USDT transaction hash and I will tell you what the chain says about it, including when the honest answer is that it is too early to tell.