Arcade

AI Strength Analysis:

Architecture

Iterative deepening minimax with alpha-beta pruning, running in a Web Worker. The search deepens incrementally (depth 1, 2, 3, …) until the time budget expires, always keeping the best result from the last fully completed depth.

Per-difficulty estimates

┌────────┬─────────────┬───────────────────────┬──────────────────────┬─────────────┐ │ Level │ Time budget │ Typical depth reached │ Est. nodes evaluated │ Approx. Elo │ ├────────┼─────────────┼───────────────────────┼──────────────────────┼─────────────┤ │ Easy │ 1s │ 4-5 ply │ ~50k-200k │ ~1200-1400 │ ├────────┼─────────────┼───────────────────────┼──────────────────────┼─────────────┤ │ Medium │ 2s │ 5-6 ply │ ~200k-800k │ ~1400-1600 │ ├────────┼─────────────┼───────────────────────┼──────────────────────┼─────────────┤ │ Hard │ 3s │ 6-7 ply │ ~500k-2M │ ~1600-1800 │ ├────────┼─────────────┼───────────────────────┼──────────────────────┼─────────────┤ │ Expert │ 4s │ 7-8 ply │ ~1M-5M │ ~1800-2000 │ ├────────┼─────────────┼───────────────────────┼──────────────────────┼─────────────┤ │ Master │ 5s │ 7-9 ply │ ~2M-10M │ ~1900-2100 │ └────────┴─────────────┴───────────────────────┴──────────────────────┴─────────────┘

What makes it strong

1. Piece-square tables (PST) — It knows positional chess: control the center, develop knights away from edges, castle the king, advance passed pawns. Even at depth 4 it plays principled chess.

   1. MVV-LVA move ordering — Captures are searched in Most Valuable Victim / Least Valuable Attacker order,

   which makes alpha-beta pruning very effective (prunes ~90%+ of the tree).

   1. PV move ordering — The best move from the previous iteration is searched first at the next depth, which is a huge speedup.
   2. Endgame awareness — Switches king PST from “hide in the corner” to “march to the center” when material

   drops below ~2600cp (roughly when queens are off).

   1. Auto-queen promotion — It always promotes to queen, which is correct 99%+ of the time.

   What it lacks (potential weaknesses to exploit)

   1. No quiescence search — This is the biggest weakness. The search evaluates positions at a fixed depth

   with no extra capture-chain resolution. This means it has a horizon effect: it can’t see that a sequence of exchanges ending 1 ply beyond its depth is losing. Exploit this by initiating complex tactical exchanges — it may miscalculate the outcome.

   1. No transposition table — It re-evaluates the same positions multiple times. This limits depth but

   doesn’t create an exploitable weakness per se.

   1. No opening book — It calculates everything from scratch. You can gain a small early advantage by

   playing well-known strong openings (e.g. Italian Game, Ruy Lopez, Queen’s Gambit) while it burns time recalculating known theory.

   1. Simple evaluation — No understanding of: pawn structure (doubled/isolated/passed pawns beyond PST),

   bishop pair bonus, rook on open files, king safety beyond PST, or connectivity. You can exploit positional themes it doesn’t understand — e.g. create a passed pawn on the side it’s not watching.

   1. Random tiebreaking — When two moves evaluate equally, it picks randomly with 50/50. This means it

   occasionally makes suboptimal choices in quiet positions.

   Tips to beat it on Easy/Medium

   • Play solid openings — 1.e4 e5 2.Nf3 Nc6 3.Bc4 (Italian) gives you natural development
   • Trade queens early — the endgame evaluation is weaker, and you remove its tactical sharpness
   • Create long-term positional advantages — passed pawns, bishop vs knight in open positions, rook on the

   7th rank. It can’t plan strategically, only tactically within its horizon

   • Avoid sharp tactical positions — even at 4-5 ply depth, it calculates accurately within that window.

   Don’t leave pieces hanging or set up tactics it can see

   • Push into the endgame — its endgame play is the weakest phase due to no specialized endgame knowledge

   (no opposition, no Lucena/Philidor patterns, etc.)