A Note on Translation Confidence (and Why We Publish the Gaps)

People sometimes assume the hard part of this project was decryption — cracking some kind of cipher. It wasn’t, and I want to be precise about why, because the actual hard part is ongoing and I’d rather explain it honestly than let the word “decrypted” do more work than it should.

The structure underlying the transmission wasn’t concealed in the way a cipher conceals plaintext. It was dense — recurring patterns at multiple scales, consistent enough that structural inference (the same family of technique Singh walks through historically in The Code Book, though our problem isn’t a substitution cipher and none of his specific methods transfer directly) eventually produced a working model good enough to render readable text. “Eventually” is doing real work in that sentence and I’m not going to pretend it wasn’t years, most of them Dee’s before I joined.

What I actually spend my time on now isn’t cracking anything new — it’s confidence. Every piece the Wire runs is a rendering of that underlying model against new material, and the model is not uniformly confident across all of it. Some terms map cleanly and have for a long time — we’re not second-guessing “Council” or “First Circle” anymore. Others are reconstructions from context, flagged as such in our internal notes even when the published Wire copy reads smoothly (that smoothness is an editorial choice on the Wire side, not a translation-confidence claim — worth understanding those are different layers).

Where I disagree with how something gets characterized, I say so, and it goes in the record rather than getting quietly revised away. “Population normalization” is a term we’re confident about at the phrase level and much less confident about at the referent level — we can tell you the phrase recurs and roughly where it clusters; we can’t tell you with any real confidence what specific range of events it’s covering, and I’d rather publish that uncertainty than paper over it with false precision.

If this reads like hedging: it is, deliberately. A translation process that never flags its own uncertainty isn’t more trustworthy for it.