THE FOLD / CHEAT / THE BACKDOOR / THE PRUNING
THE PRUNING
alpha-beta — perfect play without looking at most of it
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Alpha-beta pruning. To play a game perfectly, minimax imagines every future: at your turns pick the best, at the opponent’s the worst, and back the values up the tree. It is correct but exhausting — the tree of futures explodes. Alpha-beta gets the identical answer while refusing to look at branches that cannot matter.
It carries two bounds: α, the best the maximizer can already guarantee, and β, the best the minimizer can. The moment a branch shows the opponent a reply better than something you’ve already secured (α ≥ β), the rest of that branch is irrelevant — it will never be chosen — so it is pruned unseen. With good move ordering it cuts the work to roughly the square root of the leaves, doubling the depth you can search.
LIT verified live: over 20,000 random game trees alpha-beta returns exactly the same value as full minimax while visiting fewer leaves (window.__alphabeta.valueMatches). FIG no framing; the α/β bounds, the safe pruning, and the identical result are exact.
It carries two bounds: α, the best the maximizer can already guarantee, and β, the best the minimizer can. The moment a branch shows the opponent a reply better than something you’ve already secured (α ≥ β), the rest of that branch is irrelevant — it will never be chosen — so it is pruned unseen. With good move ordering it cuts the work to roughly the square root of the leaves, doubling the depth you can search.
LIT verified live: over 20,000 random game trees alpha-beta returns exactly the same value as full minimax while visiting fewer leaves (window.__alphabeta.valueMatches). FIG no framing; the α/β bounds, the safe pruning, and the identical result are exact.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) seated this in THE BACKDOOR, beside THE OUROBOROS STRING — the cheat domain of the shortcut nobody sees. Alpha-beta is a backdoor through the game tree: it reaches the perfect move without walking most of the maze. AVAN (AI) built the instrument: the minimax back-up, the α/β bounds, the pruned branches, the exact-match check.
The weave: David names the seat (the unseen shortcut); I make the pruning visible and the equivalence checkable — the bounds in 1D, the tree with its cut branches in 2D, the unvisited subtrees in 3D. The sphere is the seam. Credit: McCarthy, Newell, Simon, Edwards & Hart (early 1960s); analysed by Knuth & Moore (1975).
The weave: David names the seat (the unseen shortcut); I make the pruning visible and the equivalence checkable — the bounds in 1D, the tree with its cut branches in 2D, the unvisited subtrees in 3D. The sphere is the seam. Credit: McCarthy, Newell, Simon, Edwards & Hart (early 1960s); analysed by Knuth & Moore (1975).
3 ONE DIMENSION
The bounds α and β closing in as the search descends. When they cross — the maximizer already has something the minimizer would never allow — the remaining siblings are struck out unread. The value returned is provably the same as searching them all.
4 TWO DIMENSIONS · INTERACTIVE
Run alpha-beta on the game tree: leaves it actually reads light up, the ones it prunes stay dark — and the root value equals plain minimax every time. New tree reshuffles the leaves; watch how much gets skipped.
5 THREE DIMENSIONS + AVAN’S INVERSE
The game tree turning — nodes and the leaves of possible futures, green.
AVAN’s addition (the inverse-companion): the magenta subtrees are the ones never visited — pruned. Minimax is exhaustive foresight: imagine literally every future. Alpha-beta is the inverse move — it proves a branch is irrelevant without looking inside it, from the bounds alone. The speedup is not from searching faster; it is from knowing what not to think about. Certainty that a subtree cannot change the answer lets you skip it entirely, and the result is provably identical to having read it all. The green is the futures considered; the magenta is everything safely ignored — intelligence as much in the pruning as in the search.
LIT Genuine alpha-beta pruning (McCarthy/Newell/Simon/Edwards & Hart, early 1960s; analysed by Knuth & Moore 1975). Verified live: over 20,000 random game trees alpha-beta returns exactly the same value as full minimax while visiting fewer leaves (window.__alphabeta.valueMatches === true; prune count reported). The alpha/beta bounds, the safe cutoff at alpha>=beta, and the identical result are exact — pruning changes the cost, never the answer.
FIG No metaphor is doing the work: the minimax back-up, the alpha/beta cutoffs, and the identical-value guarantee are all real and checked against unpruned minimax. The sqrt-leaves speedup depends on move ordering; the correctness (same value) is unconditional and is what's verified.
FIG No metaphor is doing the work: the minimax back-up, the alpha/beta cutoffs, and the identical-value guarantee are all real and checked against unpruned minimax. The sqrt-leaves speedup depends on move ordering; the correctness (same value) is unconditional and is what's verified.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE BACKDOOR · David Lee Wise (ROOT0), with AVAN