Dashboard

Route: /dashboard

Scenario: You’ve just pushed a release and want to know in 30 seconds whether anything broke, whether the overnight fix runs succeeded, and what your users reported since you last logged in.

Admin console · animated

Click to expand

The dashboard answers those questions without clicking into any sub-page. It’s your starting point every morning — a live snapshot of the entire Plan → Do → Check → Act pipeline.

What to look for first

When you land on the dashboard, scan these three things in order:

1. PDCA flow canvas (centre) — are reports stuck at one stage? A pile-up at “Plan” means the triage queue is backed up. A pile-up at “Do” means fix-worker runs are failing or waiting. Click the glowing stage to jump straight there.

2. KPI tiles (below the canvas) — if “Fixes failed” is non-zero, open the Fix orchestrator immediately. If “LLM cost” spiked overnight, check the cost chart for the culprit pipeline.

3. Triage queue (bottom-left) — un-triaged reports waiting for a human decision. This is your to-do list for the day.

PDCA flow canvas

The animated graph in the centre shows live report counts at each pipeline stage:

The stage with the most activity glows with an animated gradient edge. Click any stage card to jump to its page.

On mobile the canvas becomes a compact PDCA Cockpit — same counts, same links.

KPI tiles (14-day window)

Charts row (14 days)

Two bar charts sit side by side. Both include event annotation markers — thin vertical lines for deploys, cron runs, and BYOK key changes — so you can answer questions like “why did cost jump on Tuesday?” without digging through logs.

• Reports by day — inbound volume. A sudden drop may mean the SDK stopped sending (check SDK health).
• LLM cost by day — AI spend. A spike after a quiet period usually means a PDCA run or a fine-tuning job fired.

Triage queue

The most recent un-triaged reports, ordered by severity. Click any row to open the report detail and triage it. “View all reports →” goes to the full queue with all filters.

If the queue is empty, that’s a good sign — but if the “Open backlog” KPI is high while the queue looks empty, check the Processing queue for stuck items.

Insights row

• Top error-generating components — shows which parts of the codebase generate the most reports. If CheckoutButton is top of the list for three weeks running, it’s time to look at the Lessons page and see if a recurring pattern has been promoted.
• Integration health — a compact status badge for each BYOK provider. A red badge here means fix-worker and PDCA runs will fail until it’s fixed.
• Activity feed — recent system events: fix dispatched, PR opened, tier-up, PDCA run completed.

Common tasks

Morning check (2 minutes)

1. Glance at the PDCA canvas — no pile-up at a single stage? Good.
2. Check “Fixes failed” — zero? Move on. Non-zero? Open Fix orchestrator.
3. Scan the triage queue — triage the top 2–3 reports before standups.
4. Check integration health badge — all green? You’re done.

After a deploy

1. Open the Reports by day chart and look at today’s bar.
2. Check if a new deploy annotation appears. If report volume spikes right after the annotation, a regression likely shipped.
3. Click the spike bars to drill into the individual reports.

Investigating a cost spike

1. Find the spike on LLM cost by day.
2. Hover the event annotation on that date — was it a PDCA run? A fine-tuning job?
3. If unexplained, open Integration health → Recent LLM calls log.

Related pages

• Inbox — all open actions prioritised in one list
• Reports & triage — the full triage queue
• Fix orchestrator — fix-worker run detail
• Integration health — full LLM health and cron monitoring