opensense
opensense

First 30 days

Operator runbook for the first month

A week-by-week runbook for the first 30 days of a new OpenSense site. Targets a single café or small kitchen, three or four sensors, one operator. Adapt cadences for larger operations.

Week 0 — install day

  • Choose sensor placements with food on the shelves. The warmest stored item is the sensor location; "the geometric centre of the fridge" is wrong if there is hot air dumping at the top.
  • Pair every sensor and watch one reading land before walking away. Do not assume.
  • Photograph each sensor location and upload to the device page. In 3 months you (or a different operator) will need to know which probe is "front-fridge".
  • Connect Telegram. Send the test alert. Confirm it lands.
  • Connect email recipients. Send the test alert. Confirm it lands.
  • Set the site's timezone. Send a test report; verify the day boundaries.

Week 1 — observation

You will get false alarms in the first week. That is normal and correct — the system is showing you that defaults do not match your operation.

Common false-alarm causes in week 1, in order of frequency:

  1. Delivery door propped open. Set a 15-minute grace period; if deliveries are routine, widen to 30 minutes.
  2. Defrost cycle. Configure a time-of-day window on the rule.
  3. Probe in the wrong place. Move the probe; do not soften the alarm.
  4. Compressor short-cycle on an old fridge. Add a second probe at the opposite end of the cabinet to confirm the swing is real.
  5. Captive-portal WiFi. Move sensors to a dedicated SSID if possible.

After every false alarm, leave a note in the acknowledgement field explaining what you changed. This is what the auditor wants to see in month 6.

Week 2 — refine

By week 2 the false-alarm rate should be 0–1 per device per week. If it is higher, investigate before suppressing. A 5-alarms-per-week rate is the system telling you the fridge is not holding temperature; suppressing the alarms with a wider operating range masks a real maintenance issue.

Tasks:

  • Review every channel's last-7-day chart. Are the band edges tight? Loose? Adjust per channel.
  • If you have a freezer that hits −18 °C ± 1 °C consistently, tighten the operating range to capture compressor decline early.
  • Set up a monthly report subscription (auto-render on the 1st at 02:00 site-local, email to all recipients).
  • Confirm at least one acknowledgement note is in the system from a real event. Practice the workflow.

Week 3 — backfill and audit-prep

  • Print the first 14-day report on paper. Sign and date the cover page. Drop it in the same folder as the paper logs you are migrating from (if any).
  • If you migrated from a paper logbook, file the last hand-written page next to the cover sheet. Inspector understanding: "we moved on this date."
  • If you have not had a real alarm yet, force one. Set the operating range to room temperature for 60 seconds, save, revert. The audit trail records the test as alarm_opened → alarm_cleared within minutes; tag it "test event" in the note. Auditors prefer to see one practice event over a fully clean month-1.

Week 4 — first monthly report

  • Render the month-1 PDF. Read it like an auditor would.
  • Look for the "OFFLINE — no data" stripes. Investigate any gap > 1 h.
  • Look at the excursion log. Each entry should have an acknowledgement note; if any do not, retroactively add a note explaining what you understand now.
  • Calibration record: enter the date of any sensor calibration events you have performed (none expected in month 1; the field starts populating in month 6 if you run periodic checks).

Week 5+ — steady state

After the first month:

  • The system runs itself; you read the monthly report.
  • A real alarm fires ≈ every 6–12 weeks per device for normal cold-chain operation. If it is more frequent, the underlying equipment is failing.
  • Calibrate the sensors annually: take a portable calibrated probe, hold it next to each installed sensor for 5 minutes, record the delta. Apply offsets in the channel config if drift > 0.5 °C.
  • Rotate device tokens annually. The dashboard flags devices with 365-day-old tokens; do not let the flag carry into month 14.

What you should not do

  • Do not soften alarms because they are annoying. They are signalling a real-world problem; fix the world.
  • Do not delete a device because it has had alarms. The audit trail will show the deletion; auditors are suspicious.
  • Do not share login credentials with multiple operators. Wait for the Team tier or rotate the magic-link email per operator.