Reading the console

How to read the operator console when a number is missing, a job is a sample, or a machine is quiet.

Reading the console

The operator console is the signed-in control plane. It is there to tell you what the fleet has actually reported. A dash, a blank percent, or the words "no samples" are a result. They mean the console refused to invent a number.

Source: control-plane/web/lib/health/pulse.js builds the operator pulse. control-plane/web/lib/data.ts supplies the dashboard figures. The rules below match the tests named in planning/TODO.md cards hs-pulse-0001 and hs-pulse-0002, and the commit e32f2924 (fix: stop inventing demo MRR and storage totals).


1. Start on the Command Center

Open the operator home after you sign in.

Read the headline first. It is a count of things that need a person, computed from the rows the database returned. It is not a slogan.

Then read the Machines line. It names each state it saw:

  • answering
  • degraded
  • quiet
  • provisioning

Those counts add up to the number of machines in the list. A machine that is still provisioning is counted. It is not dropped so the other numbers look tidy.

If the line says a machine is quiet, that machine has not given a fresh heartbeat. Open that machine before you trust any app that is supposed to run there.

2. Treat a dash as "no sample"

Uptime, backup success, platform MRR, and image storage each need their own samples.

You seeIt meansIt does not mean
a dash, or "no heartbeat samples"No server in view has an uptime sample.The fleet is at 100% or at 0%.
"no backup reports"No server in view has a backup-success sample.Backups are failing, or passing.
a dash for platform MRRThis session has no billing sample.Revenue is zero, or revenue is fine.
a dash for image storageNo agent has reported image bytes.The bucket is empty.

A single server that does have heartbeat history may still show its own percent. That percent is labeled as heartbeat history. It is not copied onto the fleet.

Demo mode does not call a provider, a model, or a billing API. Numbers you see there came from the in-memory rows, or they are dashes.

3. Read a recovery card the same way

On Recovery, a running job with no agent evidence is badged Sample · unproven. The evidence panel still says there is no agent evidence yet. The progress percent is a dash.

That badge means the card is showing the shape of a job so you can learn the screen. It is not a restore you can cite.

A drill has a mode:

  • restore is an end-to-end restore drill.
  • check is an integrity check of the repository. It does not prove a restore.

History rows that predate the evidence fields read as legacy, proof unknown. Leave them as history. Do not promote them to a passing drill.

4. What to do with each kind of gap

  1. Quiet machine. Open the machine. Run hostssh doctor on it if you can reach it. A doctor failure names the missing piece (Docker, the control-plane pin, the license file).
  2. No backup sample. Run a capture, or open the last capture's receipt. Absence of a percent is not a license to skip the capture.
  3. Sample · unproven job. Ignore it for operations. Start a real job only when you mean to, on a destination you have already emptied. See Recovery commands.
  4. MRR or storage dash. Leave it. Do not type a number into a status update. Billing and image bytes show up when those subsystems have rows.

5. Customer console versus operator console

Customers land in the customer console. They see their tenant. They do not see other tenants' machines, invoices, or recovery jobs.

Operators land in the operator console (the pages under the admin section). A customer session that hits an operator URL is sent away. The admin pages are documented in God mode.

6. When the page and this guide disagree

This guide describes the control plane in git main at e32f292468c4104a1f9fe8bf8db2f22c159567f1.

On 2026-11-03 the running container's SOURCE_COMMIT was 584e9a8709264c14d3952f01bb41ff6a88472c14. That build is older than the commit this guide describes. Pushing git does not swap the container. An operator redeploy does. Until that redeploy, a live page can still be the older build.

How to check, and how the redeploy works, is in the developer doc Control plane deploy.