Release notes 0.1.182

31 October 2026: Every portal page now leads with its conclusion and has one filled button.

Release notes · Updated 2026-10-31 · Describes release 0.1.182

0.1.182 — 31 October 2026

A new portal layout: one decision per page

• Every portal page now leads with its conclusion and has one filled button. The page says in one line what it is for, then what happened, then the one next step; detail sits in folded sections with one-line summaries, and engineer detail (routing evidence, raw payloads) under Diagnostics. Other actions sit in a More actions menu that never repeats the page's primary. Features and page addresses are unchanged (the API additions are listed below); buttons that only jumped to another part of the page are gone.
• The sidebar follows the operator's day: Home (Dashboard, Needs you, Finish setup while setup is unfinished), then Incidents (Incidents, Problems, Service dependencies), Compliance (the four finding lists and Audit & coverage) and Changes (Changes, Runbooks), then Insights, Agent quality and the settings groups folded at the bottom. Each group says what it is for. A card under the menu counts the approvals waiting and opens the queue. "Service correlation" is now Service dependencies (same address) and the posture list reads Configuration.
• The darker flat look is the workspace brand "scifi", retuned flat (opaque cards, no glow or glass, Inter for text) and checked for contrast. The default brand and theme do not change; an owner picks the brand per workspace.

Incident and finding page

• The record page answers in the operator's order: a header with priority, severity, status and five questions (what, where, who changed it, how we know, since when — "not recorded" when nothing answers one); What happened (the agent's conclusion, the likely cause from the ranked changes, the evidence); What you need to do, whose one button performs the step itself (release the fix, approve, roll back first, re-analyse, verify, grant budget) instead of jumping to a layer. The L1–L7 steps sit on one numbered step bar that opens each layer, the blast radius is drawn as a path, and the Timeline shows its last six rows with "Show all".
• Re-analyse is always under More actions when the viewer may run it and the record's AI budget is not used up, not only when it is the next step.

Change page, Changes list and Runbooks

• The change record opens on one decision: class, status and tier, a Request → Classify → Approve → Window → Run → Verify step bar read off the record, and one primary per state (Approve, "Approve — runs in the next window", Run standard change, Review execution, Verify now, Roll back, Close review, Download change request). The change request document stays the first section under the decision. An approver who has voted is not offered a second vote; a finished change skips Classify.
• The Changes list has count tiles, All / Requests / Seen outside tabs, Class and Approvals columns and a class filter. API: GET /changes and /changes/summary take change_class and route_id; the summary adds a by_class count; the window block carries next_window.
• Runbooks moved under Changes and show as cards: taken off pre-approval first, then certified, the rest folded, each with its steps by phase, who certified it and its runs linked to the changes they made. Re-certify opens the route's own four-eyes certify dialog.

Lists, dashboard and the other pages

• Incident and finding lists open with severity chips, a 14-day trend, and one folded "N grouping suggestions" strip for that list's own pillar. Merge says what it closes and that it cannot be undone; Dismiss is now Keep separate. The incidents table gains an AI one-line cause, a priority stripe and a Test tag. The finding lists group open findings into risk cards (vulnerabilities by package set and resource, the rest by control); KEV is not sent by any scanner the platform reads and the card says so. API: GET /findings/groups; GET /merge-proposals carries the pillar and takes ?pillar= / ?exclude_pillar=; a merge suggestion over the same records no longer stacks a second card.
• Dashboard: a KPI row (unresolved incidents, MTTA, MTTR, alert noise reduced, auto-handled), Pillar health, opened-versus-resolved and the monthly noise table, and one human-gate card with the queue. A workspace with nothing connected and nothing recorded gets the setup steps instead of a dashboard of zeros.
• Needs you groups items by kind, most urgent kind first. Problems are ranked cards, most recurring first (GET /problems?sort=count). Service dependencies opens on the dependency map; the alert list is its Alerts received tab. Audit log checks the chain on arrival and folds three or more identical rows into one. Compliance scores, Report, Usage & billing, Agents activity, Quick start, Estate map ("Harvest now" is now Refresh inventory now) and the settings pages follow the same layout. The agent roster from 0.1.181 (pillar agents, AI roles, what each does, running or idle) sits in the Agents activity page's Right now card and role rows.

Numbers and words

• Dates read one way across the portal ("Oct 31, 2026, 09:43 UTC"), costs show two decimals and "< $0.01" for a cost under a cent, and counts use the right singular or plural.

Next steps

• Next: Release notes 0.1.181
• All What's new pages
• Documentation home