It really depends on how large the amounts are in your wallet (i.e. dust or bags?)  It also depends greatly on what you want to do with the funds - buy one big ticket item, or a few smaller items.

Specifically, if you want to buy an item that requires two or three of your deposits, ask yourself if you want to consolidate all transactions at the same time (and probably move to new wallet addresses while you are at it), or just pay the bare minimum and pay for the item and have one even smaller transaction in your wallet.

No. It depends: if a small amount arrives in my wallet, I sometimes consolidate it instantly, and sometimes just let it be. I also receive weekly payments to a cold wallet. I've consolidated those a few times (at fees higher than the current fees), but it's a lot of work to reboot into offline mode to sign the transaction. So I don't do it very often.

You can still do it when fees are 1 sat/byte. But as long as fees stay this low, waiting a week saves you a few sats. If you need the funds soon and expect fees to go up, then weekly consolidations at 1 sat/byte would be a good choice.

Inputs are larger than outputs, so consolidating 2 legacy inputs 4 times costs 4 times 340 bytes (1360), while doing it all at once takes only 784 bytes. It's always a trade off between saving bytes on transactions and future fee expectations. And at the moment you know for sure fees can't get any lower, so it can't get any better.

The number of unspent outputs you have should not be considered when deciding if you want to consolidate. You should consider:

• the value of your unspent outputs
• how high you anticipate transaction fees being when you need to spend your unspent outputs
• the value of the transactions you anticipate spending when fees are higher (including the cost of the tx fee)

For example, if you have a bunch of unspent outputs valued at 0.002, but never anticipate spending more than 0.001 in a single transaction (inclusive of anticipated fees), you should not consolidate any of your unspent outputs, and should not consolidate anything until you have unspent outputs totaling 0.001 (including tx fees to consolidate), but once you have enough to consolidate into at least 0.001, you should not consolidate additional unspent outputs, unless you have enough to consolidate into an additional multiple of 0.001, in which case your new outputs will each be 0.001.

If you anticipate spending varying amounts, you can use a similar principle. For example, if for every three 0.002 transactions you will have, you will have one transaction valued at 0.005, you can consolidate into new outputs valued at 0.002 three times and 0.005 one time (in any order).

When deciding if it is economical to consolidate, you should add the cost to create a consolidation transaction to the cost of a transaction with one input and one output (plus change if applicable) at the fee rate you believe fees will rise to when you have to actually spend your coin. You should compare this cost to the cost of creating a transaction with however many inputs you are consolidating, and one output (plus change, if applicable). If consolidating results in a lower total fee paid, you should consolidate, otherwise, you should not.

The above ignores any privacy implications to consolidating, which you should also consider. 

It has occurred to me that something else can be done as well. Simply wait to consolidate, for example, to the 8-10 legacy inputs that poyaa87 said, and if I have to make a transaction, use coin control and use the large input. At the slightest hint of rising fees, consolidate, but just the small inputs just in case the transaction gets stuck in the mempool. If I have to spend, use the large input. Even if it would take too long to confirm the transaction to consolidate, I wouldn't care, because I could keep spending. I think that when the mempool was fuller I waited about a month. In this case I would not consolidate all the inputs, as I would finally have 2 inputs, the large original one and the second with an origin of 8-10 small inputs consolidated into one. I wouldn't have all the inputs consolidated into just one but I would save on fees as well.

I don't know if I am making myself clear, and what do you think, because little by little I am learning but I still have a lot to know.

I use bech32 adresses but I never talked about consolidating into the same (old) address. I was thinking something like:

I have 1BTC in one input in address1.

In address2 I receive 0.1BTC every week, so after 4 weeks I have 4 inputs for a total of 0.4BTC.

I could consolidate the 4 inputs of address2 into a new address3 but let's suppose that the mempool hasn't cleared for the past couple of weeks. I can sent the transaction with just 1 sat/byte as fee and if the transaction takes a while to get confirmed and I need to spend I can still use address 1.

This was a hypothetical scenario because the most normal thing is that if I had to consolidate I would consolidate everything, all the inputs from all directions, using a rate that would prevent the transaction from being stuck in the mempool and I would not be able to use the funds. For this I could use fee calculators or this thread.

1. This time you didn't mention if address2 is legacy or not. I'd advise to consolidate to bech32.
2. 0.1BTC is not small input, it can stay as it is. If you spend less than 0.1BTC (normal), then only one input is used from address2; if you spend more than 0.1BTC at a time you won't care that much if you pay a bit more for the fee, even if the fees are not low.
3. If you have one bigger consolidated input already which you can spend when you need it, it's no rush to consolidate address2 inputs and you better wait until there will be more of them (and smaller too).

But since it's a hypothetical scenario, maybe the addresses are actually bech32, maybe the inputs are actually much more and much smaller and then the scenario would be pretty much OK.

By consolidating everything into one address, you give up privacy (and possibly safety). Not only can anyone you've transacted with find some of your other transactions, they can also see how much Bitcoin you have. When you buy a coffee, you don't want them to know you own a small fortune in Bitcoin.

When buying something, it's unlikely to have an input of exactly the right size. Sending one input and one output works well when depositing to for instance an exchange.
It is indeed worth trying to avoid small change outputs. At low fees it could be worth adding another input and consolidating it into a larger change address. Sometimes I send the change to a custodial LN wallet, or sending it to an exchange would also mean it's no longer your problem. Even better if you can pay several services at once (that works well for payments you can plan, such as hosting, subscriptions or VPN).

---

---

In retrospect, most of the consolidations I ever did were at fees >1 sat/vbyte. And, looking back, I could have had a lower fee if I would have waited. But the total "damage" isn't that bad, and well worth the possible risk of high fees later on. Half a year ago I didn't expect fees to get this low for so long.

Doesn't consolidating hurt privacy? Maybe consolidate, then mix? What's your opinion on this?

If you have reused addresses, you have already negatively affected your privacy.

Yes, one "cost" of consolidating unspent outputs is a reduction in privacy. This should be considered when deciding if you want to consolidate your unspent outputs. The value of privacy is not infinite.

I don't think consolidating unspent outputs and subsequently mixing the resulting coin would provide much privacy benefit. When you consolidate unspent outputs from two previously unconnected addresses, an observer would know that the two addresses likely belong to the same entity. Mixing your consolidated coin does not change this fact. If you want to maintain your privacy, you will need to move your unspent outputs through a mixer before you consolidate your unspent outputs, or you could send your coin to multiple deposit addresses to a mixer, and have the mixer send you a single output. If you engage in the above, an observer may still be able to draw a link between the inputs and outputs in the mixing service. You will also need to trust the mixing service to not log, nor intentionally leak information about their customers.