Loading
Connecting to the barometer…
Trend — what is changing
The board's other pages answer "what is broken now". These answer "is this getting worse, and what breaks next" — the questions a status page cannot.
Pipeline reliability · 14 days
One column per day, newest right. Read from the Data Transfer API, not from
JOBS_BY_PROJECT — a scheduled query that fails before its job materialises leaves no
row there, so a heatmap built on job history renders green straight through a real outage.
Campaign write surface
Everything else here is about reading. This is whether we can write — and it fails silently: a brand with a dead Blinkit credential just stops working, in a collection nobody reads.
The chain
Each data line, left to right, as it actually flows. A red square tells you which stage broke, which is what tells you who to call — a stale served table with a green source is a warehouse problem; the same stale table with a red source is not ours at all.
| Data line | Source | Ingest | Derive | Serve | Source system | MCP tool it feeds |
|---|
Open problems, by owner
Grouped by who can fix it, not by where it surfaced. Anything marked not paging is real but deliberately below the alert threshold.
Service uptime
Pinged every 5 minutes. A 401 is the healthy answer for an OAuth-gated MCP transport — it proves the process, the route and the auth layer are all alive. Each target declares the codes that mean healthy for it, so the board does not page on a working server that simply refuses anonymous callers.
| Service | Status | Latency | Observed | Endpoint |
|---|
Pipelines
Every scheduled query on the project, read live from the Data Transfer API — so a pipeline added tomorrow appears here with no edit. Schedules are stored in UTC; IST is derived, because the 5h30 offset is exactly the arithmetic that gets done wrong at 2am. The cron jobs below run on the pipeline-runner box, which this collector cannot reach, so their status is inferred from the table each one produces and every row says so.
| Pipeline | Schedule (UTC) | IST | Last run | State | Next run |
|---|
Data freshness by tenant
Source and served side by side for every live tenant, because that pairing is what makes a row diagnosable: a stale source with a stale served table is the collector's problem; a fresh source with a stale served table is ours. Numbers in each cell are the newest date held, with its age in days. Registry-driven — a newly onboarded brand appears with no edit.
| Tenant | Brands | SOH feed | Sales feed | Ads feed | Inventory served | Actuals served | Ads served | PO served |
|---|
MCP tool usage · last 24h
Acknowledged — known broken, deliberately not paging
Each entry has an expiry date. Past it, the problem pages again. An acknowledgement is a promise to come back, not a mute.