Jon presents this in the context of the question, "Why did armor class descend from 9 to 2?". Which is of particular interest because the prior work, Chainmail, did actually have ascending AC, so one wonders what motivated the flip. And here we have an answer: The initial concept for the system was not table-based, but rather, a simple formula in which one could subtract the AC from 20 to see the target number one needed to hit. For higher-level fighters, the chance would increase by one pip per level.
Jon kindly links back to our blog here as he includes the observation, "in its relentless quest to perfect combat systems, the OSR has previously recognized this as the rough algorithm behind the original attack matrices". And one of the commenters on his thread writes, "How cool to see Target 20 in the pre-publication D&D." We are happy to agree.
Of course, the system above is based on subtraction, whereas for Target 20 we think it's easier for players to use the equivalent addition algorithm (roll d20, add fighter level and target AC, and check if the result is 20 or more). Also note that there's at least a one-point slippage in the exact system above; there's an 18 hand-overwrritten with a 19, a correction which didn't transfer to the associated combat table. This is broadly in synch with the one or two pip error we accept with Target 20 for the sake of simplicity. Also note the draft statement 2 - 20 = 18, and that we can imagine two different perspectives in which either 18 or 19 is related to the "90%" figure. As Jon said on the OD&D Discussions forum, "These things were... loosely reckoned".
