Refoss EM16P in Home Assistant: setup traps, meter agreement and recorder load

Method

1. Installed the monitor on a 120/240 V split-phase service, with the two mains on one channel per leg, and added it to Home Assistant through the Refoss RPC custom integration.
2. Watched the readings in Home Assistant during installation and after each wiring change or restart, and compared them with the monitor's own web page.
3. Compared daily energy for October 2, 3 and 4, 2026 with the utility meter's readings from an Emporia Utility Connect, using daily totals from Home Assistant's long-term statistics. The house total was the two mains added together; branch channels were never added to it.
4. Compared the two devices' own records for November 1 to 7, 2026: the monitor's minute history exported from its local web page, and the Emporia's hourly and daily export, matched by complete hour and complete day.
5. Counted the state rows Home Assistant's recorder wrote in 24 hours, before and after excluding most of the monitor's voltage sensors from the recorder.

Setup

Monitor

Refoss EM16P (sold as the EM16 Pro), firmware 3.1.11. The two mains and one branch channel are wired so far. The integration created one voltage sensor per channel, 18 in all.

Integration

Refoss RPC (Official Maintenance), a custom integration installed through HACS. Its manifest declares local push.

Comparison meter

Emporia Utility Connect, which reads the utility's smart meter and reports through Emporia's service. Home Assistant reads it through the Emporia Vue integration.

Evidence

Installation observations from my notes of September 25 to 27, 2026. October comparison from a workbook I built on October 5, 2026 from Home Assistant statistics. November comparison from both devices' own exports, downloaded November 7, 2026. Recorder counts from audits on October 25 and 27, 2026 and a recheck on November 3, 2026. Only the November exports and the web page total were checked again for this note.

Results

Adding a voltage reference while running

A bank whose reference wire was connected while the monitor was on stayed at about 0.7 V for 25 minutes. A reboot from Home Assistant brought it to about 125 V. The monitor appears to detect its voltage references only when it starts.

Readings right after a restart

For a minute or two after a restart the mains added up to about 1.7 times the comparison meter, with mixed signs. They then settled to about 1.01 times. I treat the first couple of minutes after a restart as unusable.

A current clamp on backwards

Power and power factor read negative, and energy went to the return counters, which Home Assistant's Energy dashboard would count as export. The fix is a circuit factor setting on the monitor's own web page; the integration does not expose it.

Total on the monitor's web page

The overview added the mains and a branch channel together, so the branch was counted twice: 1.65 kW against 1.17 kW. Choosing only the mains fixed it until the page was reloaded. The choice did not stay, checked September 26, 2026. On November 7, 2026 the overview's energy for the day was still the two mains plus the branch, while the page's own total line counted the mains only. In Home Assistant I add the two mains only.

Two Wi-Fi profiles

The monitor keeps two Wi-Fi profiles on one network adapter, each with its own optional fixed address, and moves between access points. With a different fixed address on each profile, its address changed whenever it moved. Giving both profiles the same address fixed it on September 27, 2026.

Daily agreement, October 2 to 4, 2026

From Home Assistant's statistics, the Refoss mains read 2.0, 0.7 and 1.0 percent above the utility meter's readings on the three days, and 1.2 percent above them over all three days.

Daily agreement, November 2 to 6, 2026

From the devices' own exports, the Refoss mains read between 0.9 and 1.0 percent above the utility meter's readings on each of the five complete days, and 0.9 percent above them over all five.

Hourly agreement, November 1 to 7, 2026

This uses every complete hour in the full export, from early on November 1 to the evening of November 7: 159 hours. Over those hours the Refoss mains read 0.9 percent above the utility meter's readings, and 8 in 10 hours were between 0.4 and 1.5 percent high.

Local minute history

Asked for 40 days, the monitor's export held about six and a half days of minute records, November 1 to 7, 2026. None of those minutes was missing.

Recorder writes from the voltage sensors

The 18 voltage sensors wrote 95,082 of 439,678 state rows in 24 hours, about 22 percent, in the October 25, 2026 audit. The October 27 recheck found 100,560 of 459,348, also about 22 percent.

After excluding 16 of the 18 voltage sensors

I kept the voltage sensors for the two mains and excluded the rest. On November 3, 2026 the excluded sensors wrote no rows in 24 hours, and total state rows were 373,700, about 19 percent below the October 27 count. That 24-hour window included a Home Assistant restart.

Failures

• My notes also hold a single installation-day reading against the comparison meter, but its recorded total does not match the channel readings written beside it. I could not verify it, so it is left out.
• The utility meter's readings reach me through Emporia's service, which adds its own timing. My notes from October 17, 2026 record the utility's own daily figures about 1 percent below Emporia's. I have not checked that again, so the Refoss could be further from the billing meter than these ratios show.
• The two Emporia sources differ slightly: its own export and Home Assistant's statistics disagree by a few tenths of a percent on some October days. Each comparison uses one source throughout.
• Three October days and five November days are short comparisons. I did not measure drift over months, and the monitor keeps only about a week of minute history locally.
• The before and after recorder counts come from different 24-hour windows, and one included a restart. They show the size of the change, not a controlled measurement.
• The reference-at-startup behavior, the restart settling and the web page total were each seen during installation and not repeated as controlled tests.
• I did not test the monitor or the integration without internet access or the vendor's cloud.

Removed for privacy

• Network addresses, the adapter's hardware address and Wi-Fi network names
• Device names and the serial numbers of both meters
• Circuit names and what each branch channel powers
• Panel and breaker layout
• Daily household energy totals, which are given only as ratios
• Home Assistant entity names

Only ratios are given for the daily and hourly comparisons, which is enough to judge agreement but not to recompute the totals. The exports themselves are not published. Without circuit names, the branch observations cannot be tied to particular loads.

Related review