2003 · Lee Sedol's broken-ladder experiment
Lee Sedol runs a ladder he knows is broken: a 31-move whole-board ledger
Textbooks say not to play a bad ladder. From Black 67 to 97 against Hong Chang-sik, Lee deliberately pursued stones that could escape. Each turn collected a forcing profit elsewhere, turning the board into a diagonal ledger.
Lee Sedol (Black) vs Hong Chang-sik (White) · 2003.04.23 · Black won by resignation · 211 moves
A bad ladder is not a misread ladder
A ladder is a forcing diagonal chase. If the fleeing side meets support on its path, the capture fails, giving beginners a sound rule: do not start an unfavorable ladder.
Lee knew the endpoint. He valued each chase-and-escape pair as a forcing whole-board exchange, changing the question from whether one string dies to whether every intermediate profit adds up.
Thirty-one moves are settled two at a time
The diagonal looks insane on one numbered diagram. At every region, however, Black needs an extra gain—forcing contact, suppression, thickness, or reduction—or the project becomes a gift.
A static diagram shows spectacle. A clickable board isolates each exchange, then zooms mentally back to the distant ladder breaker.
Lee prices the path, not only the ladder's endpoint
Ordinary ladder reading asks captured or escaped. Here intermediate states have value: repeated forced replies can damage territory or build influence even if the target survives.
Black's win cannot prove Black 67 correct. One non-forcing White reply can break the ledger, so TreeView audits which responses are necessary.
AI compares starting cost, then audits every exchange
After White 66, the played D12 is preserved with precomputed best replies. D12 is also a top-three model candidate, so the same point appears once; other branches show whole-board play without starting the ladder immediately.
Follow the played branch to evaluation jumps, return one node, and compare candidates to locate the unpaid item.
One broken ladder and two ways to decline it
D12 lowers Black from about 46.4% and -0.3 to 41.1% and -0.6: a cost, but nothing like an amateur collapse.
L6 dominates with 7,524 visits; N8, N9, L8, then E18 handles the lower fight before using the ladder condition globally.
B5 receives only 85 visits and about 39.2% for Black, an exploratory lower-left use that search barely trusts.
Principles are priced conditions, not commandments
The game does not tell beginners to ignore ladders. Only accurate ordinary reading makes it possible to know when path value outweighs the local result.
TreeView is essential: it turns a 31-move marvel into auditable forcing relations. Understanding Black 67 means demanding a receipt for every move.