The investigation

1. The 395 printed digits are 196 two-digit coordinates once the trailing nulls come off. Odd positions are always 6-7-8-9, even positions always 1-2-3-4-5. That is a 5 × 5 square with the row digit offset by five, which is also why a lone 0 appears — it is row five.

2. The units carry no English structure in their order. The best substitution fit is indistinguishable from the cipher's own shuffle.

3. Their repeat profile settles it. That statistic cannot be changed by any one-to-one substitution — letters, syllables, nomenclator, homophones — only by reordering. The cipher sits dead centre of its own shuffle and far from any language. A transposition happened.

4. And it was total: every unit moved on its own. Swapped printed lines or groups would have left most of the order standing, and the measurement would have seen it.

5. Every transposition in reach has been tried and rejected — keyed columnar at every width, digit-level by exhaustion, 614 unkeyed routes, irregular columnar, nulls, real words and dates as keys, the whole thing run backwards, the opening scored on its own, and double transposition in all four direction combinations at 139.8 million key pairs. 222 million keys. Nothing above −1.32, where real English is −0.74.

6. Which leaves what the published record already suspected: d'Agapeyeff made a mistake. And a 2017 finding, verified here against his own printed example, gives the mechanism — he did not understand which direction his transposition ran.