Build an Opening Repertoire That Trains the Moves You'll Actually Face

Most opening prep breaks in the same place: you memorize a neat engine line, sit down at the board, and your opponent plays something the engine never bothered to mention — because it wasn't "best," just common. You're on your own by move 6.

Kingside's new repertoire trainer on the analysis board is built around that exact problem. Instead of a single sterile line, it grows a repertoire tree with two different engines playing two different roles — and it's honest about where the tree ends.

Two engines, two jobs

The asymmetry is the whole idea:

• Your side plays Stockfish's best move — one strong line. You learn the correct move, not a menu of options you'll never remember.
• Your opponent branches with Maia — the human-like neural engine. Maia doesn't answer with the objectively best reply; it answers with the moves real players actually choose at a given level. So the tree fans out along the replies you'll genuinely run into over the board, weighted by how often they occur.

The result is prep that matches reality: your responses are sharp and principled, and the branches you study are the ones a human opponent is likely to throw at you — not rare computer novelties.

It has a visible end

The honest problem with "just build the whole tree" is that it never stops — it grows wide and deep forever and becomes unusable. This trainer draws its border from the meaning of the repertoire itself, not an arbitrary node count:

• Target depth — how many moves deep you want to learn (e.g. "up to move 10").
• Response coverage — only the opponent replies that occur often enough to matter; the rare tail is trimmed and clearly labelled, never silently dropped.

Because both limits are things you understand, the size is predictable and estimated before you start: "repertoire for White up to move 10, ~50–100 positions." You see the border before you press go.

Presets, not homework

Pick a preset by plain values:

• Concise — 8 moves deep, frequent replies only (tens of positions).
• Standard — 10 moves, the balanced default (~50–100 positions).
• Detailed — 12 moves, wider coverage (more).

Building runs most-common-lines-first, so useful prep appears immediately and a progress bar counts toward a known total — "built X of ~Y positions," not an endless ribbon.

Every branch tells you why it ended

Each leaf of the tree carries a reason it stopped, so nothing is a mystery:

• End of repertoire — you hit your target depth.
• Theory complete — the position is stable and there's nothing left to drill.
• Refuted — the opponent erred and your reply reaches a decisive edge (shown with a couple of converting moves).
• Rare line — trimmed — a reply too infrequent to be worth memorizing, marked visibly so you know it was set aside (and can reveal it by raising the preset).
• Transposition — the line merged into one you already have.

Watch it play out on the board

Every move is animated on the board and written straight into your game tree as variations. It exports to PGN, reloads, and persists — so your repertoire is real study material, not a throwaway. Undo and Redo let you reshape it as you go.

Try it

Open the analysis board, set up your opening, pick a preset, and let it build. You'll get a finite, realistic repertoire — your best moves against the replies you'll actually meet — in the time it takes to read this page.

The repertoire tree is built entirely by Stockfish and Maia running in your browser — no per-day analysis limit on building it.