What the tools cannot import correctly

First, coverage. Every tool has a list of supported chains and venues, and history from anything outside that list will not load automatically. Check whether the services you used are supported before you commit to a tool. What is not supported has to go in as a generic CSV you assemble yourself, or by hand.

Second, DeFi. Providing liquidity, borrowing against collateral, claiming rewards, closing a position — on-chain, each of these appears as several token movements. A tool infers the event type from that sequence, but telling a deposit from a swap, or a receipt token from a newly acquired asset, is genuinely hard to do mechanically. Misread as a swap, the action produces a gain or loss that should not be there; misread as a mere transfer, something that should have been recognised slips past.

Third, newer transaction forms. Support for a new protocol or a new product structure always arrives after the fact, so activity from the intervening period sits as unclassified or gets rounded into the nearest type. If you know you use leading-edge services, approaching the tool's output with suspicion is about the right posture.

Fourth, assets received through airdrops or forks. An asset that simply appears in a wallet looks to a tool like a receipt with no cost basis, and how it should be handled depends on the nature of that receipt. Tokens you never asked for also turn up, and whether those should be booked as an acquisition at all needs separate thought — see tax-airdrop-received.

The basic response is to know explicitly what did not come through, and to write it down: 'this protocol's activity could not be imported automatically, so N transactions were classified by hand from explorer history'. Where the classification is a judgement call, take it to a tax accountant rather than settling it yourself. A transaction the tool did not output is still a transaction that happened.