Keep the most importance under the 1048-token cap.

Each turn can be kept whole, summarised (fewer tokens, less signal), or dropped. A bulky-but-useful turn often wins as a summary where keeping it whole wouldn't fit.

Practice — no rating pressure. Get as close to the optimum as you can; each play teaches the pattern.

Importance kept

Tokens

0 / 1048

each option shows importance i / tokens t

turn 1

turn 2

turn 3

turn 4

turn 5

turn 6

turn 7

Memory Compaction — practice the decision, scored against the optimum

What does the Memory Compaction puzzle teach?

Under a context cap, summarising a bulky-but-useful turn can beat keeping it whole or dropping it.

Why do interviews ask about this?

What to keep vs summarize is the long-context / agent-memory interview: compact to fit the window without dropping the signal.

How is it scored?

The engine computes the provably-optimal answer for the exact instance you played (dynamic programming, shortest-path, or exhaustive search — never a heuristic) and scores your attempt as a percentage of it. No login needed to see your score.

← Arena

Practice0:00

Memory Compaction · D5

Keep the most importance under the 1048-token cap.

Each turn can be kept whole, summarised (fewer tokens, less signal), or dropped. A bulky-but-useful turn often wins as a summary where keeping it whole wouldn't fit.

Practice — no rating pressure. Get as close to the optimum as you can; each play teaches the pattern.

Importance kept

Tokens

0 / 1048

each option shows importance i / tokens t

turn 1

turn 2

turn 3

turn 4

turn 5

turn 6

turn 7