Sure, but I'm trying to measure HOW suspicious. Or making a distinction between risky and not-so-risky txs.

A tx without opt-in RBF flag, one confirmed input, two non-dust outputs, and a fee of 100 satoshis per byte, is way less suspicious (or perhaps I should say: way less likely to be replaced or never confirm) than a tx WITH opt-in RBF, spending lots of unconfirmed UTXOs from multiple transactions, lots of dust outputs, and a fee of 2 satoshis per byte.

It all depends on the amount of money we're talking about, but I would feel quite safe accepting the first kind of tx for typical retail purchases up to a few hundred dollars.

Thanks for all your feedback.

Is there some generally agreed upon boundary for "enough" or "too low" fees? (in satoshis per byte, I know it's not about the total fee amount)

Similarly, is there a standard limit for dust? When I mentioned dust outputs I was just referring to outputs with very small amounts, but is there universally accepted threshold for when to consider something dust?

Also, do these dust and fee criteria depend on the current network congestion (size of memory pool or total number of unconfirmed txs)? And does the Bitcoin days destroyed measurement play a role in this?

When looking at Bitcoin unconfirmed transactions, I'm trying to distinguish between txs that are likely to confirm (whether that be in the first next block or in 2 hours) versus txs that are likely to be replaced, or never to confirm at all.

Obviously, txs with the opt-in RBF flag fall in the latter category. But I guess there are other criteria, such as:

• Having an extremely low fee, or even zero fee.
• Having lots of dust outputs.
• Depending on unconfirmed inputs.

   
Anything else I should take into consideration? What other factors could make a tx less likely to end up confirmed? I guess criteria that cause a tx to be considered "spam" are closely related to this.

For example, besides dust output, do dust (i.e. many small) inputs make a tx less sure to confirm? (other than more inputs causing larger tx data, thus resulting in a lower fee when measured in satoshis per KB)

P.S. I realize there's no single way of doing this, and there's never certainty about this anyway. I'm just looking for some indicators that give a reasonable estimate.