Experiment record

SwitchBot Plug Mini OpenAPI: stale values explained by layered reporting thresholds

Independent logging and a manufacturer development-team response show that long unchanged OpenAPI values can be normal for Plug Mini because reporting is triggered by load-band-specific change thresholds.

Short conclusion

For the investigated SwitchBot Plug Mini (JP), polling OpenAPI v1.1 every one or five minutes did not guarantee that the returned electrical status values were freshly reported by the device. SwitchBot support, after escalation to its development team, stated that Plug Mini uses a layered reporting mechanism based on load-dependent change thresholds. At the approximately 110 W load involved in this case, a 30 W change was required before a new report was generated.

This means that a collector’s retrieval timestamp must not automatically be treated as the underlying device-state update timestamp.

Objective

Determine why voltage, electricCurrent, weight, and electricityOfDay could remain exactly unchanged for long periods even while the OpenAPI endpoint was polled repeatedly, and distinguish a client-side defect, a device/cloud fault, and intentional reporting behavior.

Observed facts

Manufacturer verification

SwitchBot support reported the following result from its development-team investigation:

Load bandChange required for report
≤ 10 W> 4 W
≤ 100 W> 10 W
≤ 300 W> 30 W
≤ 500 W> 50 W
≤ 700 W> 70 W
≤ 1000 W> 100 W
≤ 1500 W> 100 W

The support response stated that the investigated load was about 110 W, so the 30 W threshold applied. It also stated that, for the investigated device, the 100–300 W threshold could be changed from 30 W to 10 W through a backend command.

With that manufacturer response, the earlier change-of-value explanation moved from an unconfirmed hypothesis to VERIFIED (supported by additional direct evidence) for the existence and threshold behavior of this reporting mechanism in the investigated case.

How to read the evidence labels

This site uses evidence labels so that direct observations, interpretations, and unresolved questions are not presented as if they had the same level of support.

LabelMeaning on this site
VERIFIEDSupported by additional confirmation such as a manufacturer response, specification, or independent verification
OBSERVEDDirectly seen or measured during the experiment or in retained logs
INFERREDThe interpretation that best fits the observations, but is not directly confirmed
HYPOTHESISA plausible explanation that still requires testing or confirmation
DISPROVENAn explanation that investigation did not support under the tested conditions

The open questions below are not a separate claim-evidence state; they simply identify points for which this investigation did not reach a conclusion.

Evidence summary for this experiment

VERIFIED

OBSERVED

INFERRED

weight is most consistent with instantaneous active power in watts. The retained experiment notes report a calculation of weight / (voltage × current) over 223 samples, with a mean of about 0.893 and median of about 0.945, physically plausible for an AC load. No retained manufacturer specification directly confirming the field meaning was available, so this remains an inference rather than VERIFIED.

Open questions

Practical implication for collectors

A one-minute API polling interval means only that the collector asks for status once per minute. It does not prove that the Plug Mini generated a new measurement report once per minute. Long-running collectors should detect repeated identical snapshots and represent freshness separately from retrieval time.

Reproduction resources

For a separate purchasing/reproduction aid, see the SwitchBot Plug Mini OpenAPI reproduction BOM.

The BOM is a separate purchasing and reproduction aid. Product availability, successor listings, and affiliate links shown there are not evidence for the findings, analysis, or conclusions of this Experiment.

About this record and privacy

This article was organized from the experiment records created on 2026-08-12 together with manufacturer information obtained through 2026-08-19. Device identifiers, MAC addresses, purchase/order data, email addresses, credentials, and household-identifying details are not published.

The original raw logger files were not retained, so Raw Data is not published for this Experiment. Sample counts, averages, medians, and similar numerical summaries in this article come from values retained in the experiment notes.

Structured data

A machine-readable structured record for this Experiment is publicly available as JSON.

View experiment.json