A language about not knowing should say what it doesn't know.
These are unresolved. Each is a good first RFC.
Should abstentions be typed by reason?
Today a reason is Text. A closed set (Missing, Ambiguous, OutOfScope, Stale) would let callers branch, but would also tempt models to pick the nearest label instead of explaining. Undecided.
How are grounds represented at runtime?
A set per value is simple and expensive. A persistent tree is cheaper to union. Whether Truffle's partial evaluation can remove the cost on hot paths is not measured yet.
Can a known value be partly known?
A record with some fields abstained (Order{id: known, eta: abstained}) is attractive but doubles the number of states a match must handle. Not designed.
What is the right granularity of confidence?
Probabilities would reintroduce the pressure to always produce an answer. The draft has known and abstained only. Whether that is too coarse is open.
Does `x?` propagation weaken the guarantee?
It makes abstention easy to pass upward, which is the intent, but it can hide where a decision should be made. Maybe a lint.
Which distribution and install route?
The planned dv CLI is a native binary. The installer, package names and hosting are not decided.
Where do the repository and community channels live?
Not decided. The source repository, chat and RFC links are intentionally absent from this site until they exist.