Fix transient resets in cloud power statistics

[Wuty-zju]

Problem

Cloud-backed daily and monthly energy entities can transiently report resets or stale values:

17.3 -> 0 -> 17.5 kWh

Missing current-period data can also leave the previous period's final value visible after a day/month boundary. In addition, restored native values could be passed through value_ratio again for models that use scaled cloud statistics.

Root cause

• micloud_statistics_power_cost initialized missing daily/monthly values to 0.
• Parsed None or decreasing values reached self.props before entity-level validation.
• Restore extra data contains the already-normalized HA native value, while the normal cloud path expects an unscaled source value. Reusing the cloud conversion path can apply value_ratio twice.

Invariants

• Missing data is not a measured zero.
• Power statistics must be finite and non-negative.
• Daily energy is monotonic within one HA-local calendar day.
• Monthly energy is monotonic within one HA-local calendar month.
• At an observed local period boundary, the corresponding accumulated value is zero.
• After an offline boundary, stale restored data is discarded; the first current-period cloud sample may be non-zero.
• A restored HA native value must not be scaled again.

Solution

1. core/templates.py preserves missing current-day/current-month values as None, while retaining an explicit numeric zero.
2. core/device.py filters missing, invalid, and same-period decreasing power statistics before self.props.update() and dispatch. Other cloud statistics remain unchanged.
3. core/utils.py contains the shared power-cost attribute matching, period calculation, and finite non-negative numeric validation used by both layers.
4. sensor.py retains the entity boundary as a second guard, performs local day/month rollover, discards stale cross-period restores, and directly restores the already-normalized HA native value without reapplying value_ratio.
5. Suffixed attributes such as power_cost_today_2 are tracked independently.

The device and entity guards are intentionally separate: the device guard protects the shared property store, while the entity guard covers HA restore state and direct entity updates.

Scope

This PR does not change:

• Xiaomi cloud endpoints, request parameters, or polling intervals;
• device customizations, units, ratios, device class, or state class;
• unrelated cloud statistics or sensors;
• the integration version;
• Recorder data or historical database rows.

Historical raw states are not rewritten automatically. Permanent long-term statistic errors should be reviewed through Home Assistant's supported statistics tools.

Validation

Focused replay passed for:

• missing, empty, non-numeric, negative, boolean, NaN, and infinite values;
• same-day and same-month decreases;
• day/month rollover with zero and non-zero first samples;
• independent suffixed attributes;
• stale cross-period restore;
• scaled restore (1750 * 0.01 = 17.5, restored as 17.5, not 0.175).

Repository validation:

• Existing pytest suite on Home Assistant 2026.2.3: 170 passed.
• python -m compileall -q custom_components/xiaomi_miot: passed.
• git diff --check: passed.

Examples

Missing update:       17.5 -> ignored -> 17.6
Same-day decrease:    17.5 -> 16.8 ignored
Online day boundary:  14.0 -> 0.0 -> 1.2
Offline day boundary: 14.0 -> unavailable -> 1.2
Same-month decrease:  488.1 -> 470.0 ignored
Month boundary:       488.1 -> 0.0 -> first valid sample
Scaled restore:       17.5 -> restart -> 17.5

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[Copilot]

Pull request overview

This PR hardens cloud-backed daily/monthly energy (“power cost”) statistics updates to prevent transient missing/invalid cloud responses from being emitted as 0, which Home Assistant interprets as a reset for total_increasing sensors and can lead to double-counted long-term statistics.

Changes:

• Update the micloud_statistics_power_cost template to distinguish missing data (None) from a measured zero and to ignore invalid/negative record values.
• Add device-side validation and filtering for power statistics responses, including monotonicity enforcement within the current local day/month and independent handling of suffixed keys (_2, _3, …).
• Add focused tests covering missing/malformed responses, explicit zeros vs missing records, monotonicity rules, suffix handling, and DST fallback.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
tests/test_power_cost_statistics.py	Adds regression/unit tests for missing data handling, monotonicity, suffix independence, and failure modes.
custom_components/xiaomi_miot/core/templates.py	Changes the cloud power-cost template to output None for missing periods and ignore invalid record values.
custom_components/xiaomi_miot/core/device.py	Adds response validation, template error handling for power-cost stats, and an attribute filter to reject invalid/decreasing same-period updates.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.