The moment hard begins
Every hard puzzle here still opens with singles. You'll place eight or ten digits without much trouble, and then the grid will go quiet — no square with one option, no digit with one home. That silence is the actual start of the puzzle.
What breaks it is a technique that operates on possibilities rather than placements. To use one you need the possibilities visible, which means candidates written into the squares. Our hard puzzles are verified to require at least one such technique, and verified not to need the grid-spanning patterns that expert demands.
Writing candidates without drowning in them
Fill every empty square with every digit still legal there, and a hard grid becomes a wall of small numbers. Two things make it manageable.
First, don't fill them in at the start. Solve the singles first — every digit you place removes candidates you'd otherwise have written and then crossed out. The board above has a Fill button that writes them all in one step, and the right time to press it is when the singles have run dry, not before.
Second, read for squares with exactly two candidates. Bivalue squares are the raw material of almost every technique above this level, and once you start noticing them the wall resolves into a much smaller set of interesting places.
Subsets: when squares club together
Suppose two squares in the same row both hold exactly the candidates 3 and 7. You don't know which takes which. You do know that between them, they will consume both digits — so no other square in that row can be a 3 or a 7. Cross them off. That's a naked pair, and it extends: three squares between them holding only three digits behave the same way.
The mirror image catches more people out. If, in a row, the digits 3 and 7 can only appear in two particular squares, then those two squares are the 3 and the 7 — whatever else they were listing can be deleted. That's a hidden pair, and it's the same relationship read from the other end.
Locked candidates: the workhorse
This one probably unlocks more hard puzzles than everything else combined, and it needs no arithmetic at all.
Look at a box and pick a digit. If every square in that box that could hold the digit sits in the same row, then the digit is definitely somewhere on that row inside that box. You don't know which square. You don't need to — it means the digit cannot appear on that row anywhere else, outside the box. Cross it off along the rest of the row.
Run it backwards too: if within a row a digit can only sit inside one box, remove it from the rest of that box. Same idea, opposite direction. Between them these two account for most of the "how on earth was I supposed to see that" moments at this level.
Two mistakes worth avoiding
Stale candidates. If you write marks by hand and then place a digit, you have to remove that digit from every square it now sees. Miss one and you'll make a deduction from information that isn't true any more. The board here does this for you.
Guessing to get moving. A hard puzzle never requires it, so a guess here is always a shortcut past something findable. If you're properly stuck, the hint will name the technique that applies and show you the squares — which is a far better trade than picking a digit and hoping.
Both of these get easier with drilling rather than reading. The technique trainer hands you a position where a naked pair is the only move on the board, and does the same for hidden pairs and locked candidates — the three that turn hard puzzles from a grind into a routine.
The step after this
At hard, every technique is local: it lives inside one row, one column, one box. Expert is where patterns start spanning the whole grid and you have to look at four squares in different corners at once. That's a genuine change in what you're doing with your eyes, not just a harder version of this.