How content moves through its lifecycle
The status ladder from draft to archived, plus versions, changesets, trash and audit entries — what each records, and which of them can be undone.
A page moves through Kywi in four overlapping ways: its status changes, its edits accumulate a history, batches of it release together, and some of what happens to it is recorded and some isn't. Each has its own surprise, and this page states each one first.
One ladder, and skipping it is not a special power
content status ladder
Content moves through draft, pending, approved, published and archived, and whether a piece of content has to sit in the queue on its way to published depends on the role moving it, not on a setting a site admin flips for someone. An author can only move draft to pending; an editor can already skip straight from any of draft, pending or approved to published in one action — that's the whole reason the queue is skippable at all, and it applies to every editor by default, not as an escalation. A site admin or super admin isn't on a per-status table at all and can move content between any two statuses, including backwards. One guard now enforces this ladder on every door that writes a status — creating, editing, bulk publish, submitting — except one: restoring an old version, below, skips it completely. See Bypassing the queue: who can transition directly, and for the queue itself, Reviewing and approving submissions and Publishing, scheduling and expiry.
Every change leaves a trail behind it
version / version snapshot
Every status transition, and any edit that carries a layout, writes a numbered snapshot of the page's fields and layout as they stood at that moment — version history accumulates on its own rather than something you maintain. Restoring one writes its fields, layout and status back onto the live page through the same validated path an ordinary save uses, and it is itself a forward-moving edit: restoring creates a new version recording the restore, so nothing is erased, only added to. Two things about a restore are easy to miss. It is the one status-writing action with no ladder check at all, so restoring an old archived snapshot onto a page that's currently published archives it on the spot — a move the ladder otherwise never allows an editor from published. And history is pruned as it grows: each new snapshot trims that page's history back to its newest fifty, with no protected slot for a version you specifically want to keep. See Version history and rollback.
Publishing more than one thing at once
changeset
A changeset batches a set of content changes — creates, updates, deletes — so they release together in one transaction: one item failing to apply rolls the whole publish back, not just that item. An item in the batch can carry a specific captured version, applied onto the live page before it publishes, or none, in which case publishing just flips the current row live as-is.
Deleting, and what gets recorded along the way
soft delete / trash
Trash merges two different soft-deleted sources — content and media — into one surface, and deleting either moves it there rather than removing it outright. Emptying trash has two different blast radii: emptying the trash of the site you're working in is one action, and emptying every site's trash on the install at once is a separate, more tightly gated action. Restoring or permanently deleting a single item, by contrast, always stays inside whichever site that item belongs to. See Trash and permanent deletion and Trash and recovering deleted work.
audit entry
An audit entry is not written for every change — only the workflow path (submitting, approving, rejecting) and revision approve/reject create one. A direct write — creating, updating, publishing or deleting a piece of content outside that workflow, including one made by an agent through its own tools — leaves no audit-log row at all. And where a row is written, it carries the same user id an interactive session's write would; there's no separate marker for a key-driven write versus a person typing at a keyboard, so an audit entry alone can't tell the two apart. Recovering from a change nothing logged means reading version history for the state before it and checking trash, not searching for an audit entry that was never written. See When the agent gets it wrong and Version history and rollback.