"A blank the tool admits is worth more than a detail the tool made up. Honest gaps beat confident fiction."
The Temptation to Fill In
When Carve normalizes a rough plan, it constantly hits values it cannot determine: a date that could be one of two days, a hotel named only as "the place near the station," a transport leg with no time. The tempting behavior — the one most AI tools default to — is to guess the most plausible value so the output looks complete and confident. Carve refuses this. Facts that are present are preserved verbatim; values that are unknown surface as an explicit TBD marker. The scaffold is allowed to have honest holes. What it is never allowed to do is invent precision that was not in the source.
Why a Visible Gap Beats a Confident Guess
The reason is about trust and about what a plan is for. A plan with a visible TBD: hotel tells you exactly what still needs deciding — it is a to-do, surfaced. A plan where Carve quietly guessed a hotel looks finished but is a trap: you might book against a detail the tool fabricated, or simply never realize a real decision was silently made for you. False precision is more dangerous than an admitted gap precisely because it hides. A blank asks to be filled; a fabricated fact asks to be trusted, and does not deserve to be.
Uncertainty Is Information
This connects back to the agency principle from the consultation track: Pippa's job is to supply counsel with its uncertainty attached, not to launder guesses into facts. When Carve marks a TBD, it is doing the honest version of its job — telling you where the source ran out and a human decision is still needed. An import tool that never shows uncertainty is not more capable; it is less honest, and it moves the risk from a visible marker onto you, silently. The mark of a trustworthy normalizer is that it is comfortable saying "I don't know this one" in writing.