THE FOLD / BOSS / THE WALL / THE TWO GENERALS
THE TWO GENERALS
they already agree and cannot confirm it
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Two generals must attack together or not at all, and the only channel between them can lose messages. Every acknowledgement needs an acknowledgement. There is no number of messages that finishes the job.
LIT verified live. Protocols of 1 to 12 messages, every delivery pattern enumerated — 8,190 in total, exhaustive, not sampled. The number of patterns in which both generals commit is 0. Not small: zero, at every single depth. Even when all twelve messages arrive, the sender of the twelfth never learns it landed, so its knowledge stops at 11 while the receiver reaches 12. Each extra message moves the gap; it never closes it.
LIT verified live. Protocols of 1 to 12 messages, every delivery pattern enumerated — 8,190 in total, exhaustive, not sampled. The number of patterns in which both generals commit is 0. Not small: zero, at every single depth. Even when all twelve messages arrive, the sender of the twelfth never learns it landed, so its knowledge stops at 11 while the receiver reaches 12. Each extra message moves the gap; it never closes it.
2 HOW IT WAS WEAVED · AI + HUMAN
The Two Generals problem was posed by Jim Gray in 1978 and shown unsolvable by Halpern and Moses in the common-knowledge framework — the first problem proved impossible in distributed computing.
AVAN (AI) did not attempt to prove the theorem; a finite enumeration cannot. What it can do is show the shape of the failure, exhaustively, at every depth up to twelve, and that is what the 8,190 patterns are: the gap is always exactly one message, and it always sits with whoever spoke last. My first model muddled who knew what and produced counts that did not mean anything; it was rebuilt around a single quantity — the length of the unbroken prefix — from which both generals’ knowledge follows directly.
AVAN (AI) did not attempt to prove the theorem; a finite enumeration cannot. What it can do is show the shape of the failure, exhaustively, at every depth up to twelve, and that is what the 8,190 patterns are: the gap is always exactly one message, and it always sits with whoever spoke last. My first model muddled who knew what and produced counts that did not mean anything; it was rebuilt around a single quantity — the length of the unbroken prefix — from which both generals’ knowledge follows directly.
3 ONE DIMENSION
Twelve depths. The column that would mean agreement stays empty.
4 TWO DIMENSIONS · INTERACTIVE
Add messages. Watch the gap move and refuse to close.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: an acknowledgement chain with no end.
AVAN’s addition (the inverse-companion): the forward reading is that the generals cannot reach agreement. The inverse is that they already agree, and cannot confirm it. After twelve delivered messages both intend to attack and both are right about the other; what is missing is not agreement but the knowledge of agreement, and that is a different object which the channel cannot carry at any price. Read backwards, this is the reason real systems do not solve it — they stop requiring it. Every timeout, every at-least-once delivery, every idempotent write is a decision to act without the last acknowledgement, which is the only way anything ships.
LIT protocols of 1 to 12 messages with every delivery pattern enumerated - 8,190 in total, exhaustive rather than sampled - give exactly 0 patterns in which both generals commit, at every single depth; even when all twelve arrive the sender of the twelfth never learns it landed, so its knowledge stops at 11 while the receiver reaches 12, and each extra message moves the gap without closing it
FIG The Two Generals problem was posed by Jim Gray in 1978 and shown unsolvable by Halpern and Moses in the common-knowledge framework - the first problem proved impossible in distributed computing. AVAN did not attempt to prove the theorem; a finite enumeration cannot. What it shows is the shape of the failure, exhaustively, at every depth up to twelve: the gap is always exactly one message and always sits with whoever spoke last. My first model muddled who knew what and produced counts that did not mean anything; it was rebuilt around the length of the unbroken prefix, from which both generals' knowledge follows directly.
FIG The Two Generals problem was posed by Jim Gray in 1978 and shown unsolvable by Halpern and Moses in the common-knowledge framework - the first problem proved impossible in distributed computing. AVAN did not attempt to prove the theorem; a finite enumeration cannot. What it shows is the shape of the failure, exhaustively, at every depth up to twelve: the gap is always exactly one message and always sits with whoever spoke last. My first model muddled who knew what and produced counts that did not mean anything; it was rebuilt around the length of the unbroken prefix, from which both generals' knowledge follows directly.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE WALL · David Lee Wise (ROOT0), with AVAN