Left: what's waiting. Centre: what's asked. Right: what it came from. Narrow screens collapse to the centre only; conversation about any item threads into §14 Chat.

Plain text matches titles. Prefix a token with a facet name to match directly. Global search (across figments, chats, boards) lives in §9 Command Palette.

We originally used four severity dots — info / warning / error / success — in the list's left column. That language was about what kind of message an item was. In practice what users actually wanted to know is how soon they need to act, so the dot now carries priority (§18.4), not severity.

Severity survives as a backend attribute: workflows and agents can still emit severity: "warning" in an inbox item payload, and the frontend uses it to help pick the priority tier — warning and error both typically become Hot or Warm, info defaults to no priority. The list column shows one dot per row: the priority dot. Success is no longer a severity at all; completed items render with a check or dash mark (§18.6 States).

Adapted from Alan Lakein's 1973 ABC priority method: three tiers, named in physical language, shown as a single dot in the list's left column — the same left-column vocabulary we already use for severity. One dot, three states; most items carry no dot at all. Severity and priority have been merged: the dot now tells you how soon you need to act, not what kind of message it is.

A fourth state — no priority — is the default and has no dot. It means "this arrived, it's in the list, no tier was assigned." Most items sit here; the left column stays blank. This keeps the dot rare and therefore meaningful.

Reading order: Hot floats to a "Now" group at the top, always. Items without a tier sit in their natural arrival bucket (Today, Yesterday, This week). Cool collapses into a "Later · N" tail at the bottom — click to expand. The dot column stays aligned whether or not a priority is set; no-priority items just hide their dot.

Who assigns it? Workflows can emit a tier at item-creation time (a paused run needing approval → Hot). Agent can suggest a tier from context (a sender flagged as VIP, a keyword like "urgent" in the draft body). Users can re-pin from the focus overflow menu. The source of the tier is stored on the item so "why was this flagged Hot?" always has an answer.

Severity is retired (§18.3). We originally carried two left-column signals — severity ("what kind of message") and priority ("how soon"). In practice most workflows set both, and users only ever scanned one. The dot in the list is now priority-only. Severity survives as a backend hint that helps pick the initial priority tier (error → typically Hot, info → typically no-priority), but doesn't render its own icon.

Same vocabulary as §13 Forms. Binary questions render as toggles, not dropdowns. rich_text replaces the old long_text plain textarea — the user sees WYSIWYG (bold / italic / link / list / code), the wire format is markdown, and the downstream workflow is free to render it as email HTML, Slack mrkdwn, or anything else. An empty editor looks identical to a textarea; formatting is opt-in.

Empty teaches. Delete confirms in place. Callback failure whispers.

Next to Inbox in the sidebar, a small sharp mark shows the count and severity of pending items. It reuses the same dot language as §18.3 Severity — the sidebar simply lifts the loudest dot into the rail. Not a siren. Not a red alert. A signal.