Grounding: why you can trust an entry
Last reviewed: 16 Sep 2026
An understanding you cannot trust is worse than none: an agent that builds on a wrong entry does confident, wrong work. Grounding is what makes an entry trustworthy, and it is visible everywhere on the Understanding page.
Cited, or not yet checked
An entry about the code is cited when it points at a file and lines an agent actually opened. You can open the file from the entry. An entry about the code with no citation shows as not yet checked; it may be right, but nothing should be built on it until it is. When an agent commits such an entry, Naetive asks it to cite what it opened, and never to cite something it did not.
Decisions, constraints and milestones are different: they are backed by their record (the commit, the job, the document they came from), not by a line of code, so they carry no citation and no warning.
The page's line "N of M facts about the code cite a file" is this count.
Kept honest over time
Two things keep an old entry from misleading:
- Re-checks. An entry that has had nothing attached for three months says so ("not checked in 3 months"). Agents re-check these as they work in that topic; the mark clears when new evidence lands.
- Drift. When the code an entry cites changes, the entry is marked for re-verification rather than declared wrong: a changed file is a prompt to look, not proof of error. The next agent in that topic checks it against the code.
What you decide
Agents investigate and record. You provide direction and judgement. When two readings of the code are both plausible, or an agent disputes an entry, it reaches you as a small decision in Needs you with three choices: let an agent settle it, keep it, or drop it. That is the one thing an agent should not decide for you, surfaced on purpose.
See also
- How understanding gets recorded
- Understanding: what the project knows