Why we pivoted then
Five experiments in a row in family 1 had reverted (0001-0005), which at the time forced a change of family under the program's rule. The direct periodic attacks had run their course: the period sweep left only lengths 27-29 and 53 and up, and none of the surviving settings forced a readable key.
What happened before
- 0001 set the baseline: the KRYPTOS keyword alone gets 2 of 24 known letters.
- 0002 and 0003 ruled out ciphertext autokey at every lag and plaintext autokey at lags 1-11 and 30-52.
- 0004 ruled out K1-K3 as a running key at every offset.
- 0005 swept the table with separate plain and cipher alphabets: periods 1-26 and 30-52 are dead under every convention, and the surviving periods force key fragments that match no word.
Family 1 kept open bounds: plaintext autokey at lags 12-29 and 53-96, and the progressive key.
The change we believed was needed
Sanborn said he used "matrix codes" and then "went back to this table" (C1). If a move came first, a single repeating key could never fit the known letters where they sit on the copper. The next step was to put a transposition in front of the table: columnar routes, K3's own rule and grid turns, each followed by a periodic key (0006-0008).