"Carve hands you a draft, not a decision. Nothing it produced is real until you say Accept."
The Gated Review State
When Carve produces a scaffold, it does not write it to the plan. It loads it into the plan editor in a gated review state: autosave is suspended, and what you are looking at is explicitly a proposal, not the live plan document. You can edit it freely — fix a date, delete a bad guess, merge two days — and none of it touches the real plan yet. Only an explicit Accept persists the proposal, and accepting first snapshots the previous plan into a revisions folder. This is the same propose-then-accept loop from the consultation track, now made concrete and button-shaped.
Editing the Proposal Is Partial Acceptance
The gated state quietly solves the "partial acceptance" requirement of the import contract. You do not need a checkbox next to every field asking "accept this one?" — you just edit the proposal directly before accepting it. Keep the good days, rewrite the questionable ones, delete what Carve got wrong, and then Accept the result. Editing-before-accepting is the cheap, natural form of partial acceptance: the proposal is a draft you shape, and Accept is the single moment it becomes canonical. And because accept snapshots the prior plan first, even accepting is reversible — the previous version is preserved as a revision.
Re-Carving Without Losing Your Work
What if you already have a substantial plan and want to carve more into it? A naive importer would overwrite your careful edits with fresh AI output. Carve instead sends the current document along and merges, explicitly preserving the edits you already made. This matters because by the time you re-carve, the plan is no longer raw stone — it is stone you have already worked. An import tool that clobbers your prior work every time you feed it more input is one you stop trusting fast. Carve treats your accepted, edited plan as something to build on, not something to replace.