Repository navigation
Releases
v1.24.0
🌟 What's new in v1.24.0?
Your numbers are now complete and fair, and every dashboard card got a make-over. This is a big release: please update the integration and re-paste your cards (see After updating). No breaking changes to entities or automations.
🔥 What it means for your heat pump
| You now get | Why it matters |
|---|---|
| Every minute of runtime counted — a heat pump that runs all day shows close to 24 hours | A Home Assistant restart during a run used to drop the part before the restart (up to 11 hours a day on a 24/7 system). Runs lost earlier are rebuilt once from the stored COP measurements (#417, #419) |
| A fair quality score — each penalty grows with how far a run is from good | A 2-minute backup-heater blip no longer costs as much as 30 minutes of backup heater. All your stored cycles are rescored once (#421) |
| Correct "short pause" detection and learned thresholds | The pause before a run included the run itself, so Short off almost never fired (#415) |
| No false short runs when the heat pump switches between heating and hot water | The compressor keeps running; it is now one run, not a stop and a start (#419) |
| Cleaner COP | The old 16.5 outlier is gone from the month COP; COP below datasheet no longer fires at start-up; heating-curve advice compares readings from the same moment (#415) |
| Correct ΔT with the backup heater on | ΔT is now measured before the backup heater, so its heat no longer inflates your ΔT (#423) |
| Heating-curve advice stays visible during hot water | The advice pauses while the tank heats, but your measured curve stays on the cards (#427) |
| ESPAltherma v2 aligned with its own name catalog | Values of exactly 0 or 1 are no longer lost; cooling is recognised without a hot-water tank and for "Auto Cool" (#413, #423) |
⭐ The new quality score
Every closed run starts at 100. Each criterion subtracts points in proportion to how far the run is from good, up to a maximum. The Last cycle sensor shows what each criterion cost (attribute quality_penalties).
| Criterion | Max points | How it is counted now | Example | Before v1.24.0 |
|---|---|---|---|---|
| ⏱️ Short run | 30 | 30 × shortfall to the good run length | good run 45 min: a 40 min run → 3.3, a 10 min run → 23.3 | 30 for any run under the threshold |
| 🌡️ Low ΔT | 20 | 20 × shortfall to the good ΔT | good ΔT 5 K: 4 K → 4, 2.5 K → 10 | 20 |
| ⏸️ Short pause | 20 | 20 × shortfall to the good pause | good pause 20 min: 15 min → 5, 2 min → 18 | 20 |
| 🔁 Many short runs today | 20 | 0 up to half the threshold, then rising to 20 at 1.5 × the threshold | threshold 50 %: 25 % → 0, 50 % → 10, 75 % → 20 | 20 |
| ⚡ Backup heater | 20 | 1 point per % of the run time (step 2 counts double) | 2 of 300 min → 0.7, 8 of 300 min → 2.7, 30 of 300 min → 10 | 10, however short |
| ❄️ Defrost | 10 | 1 point per % of the run time above 5 % (normal defrosting is free) | 6 of 300 min → 0, 30 of 300 min → 5 | not counted |
Not penalised: a run that continues after a heating ↔ hot-water switch, a run that is still going (Quality today waits until it ends), backup heater during a legionella cycle, and missing data. The thresholds are your own options (good run, good ΔT, good pause). Full explanation: Quality score.
Expect your scores to change: most long runs score higher, very short runs with backup heater lower. That is the new formula, not a fault.
🎨 Dashboard cards: update them too
All seven cards were checked against the data the integration really writes, in English, Dutch and French, and unified (#427):
- Your heat pump model in every header — title + model, family, power and refrigerant.
- Dark mode — follows the Home Assistant dark mode; every card stays readable.
- One set of options on every card —
lang_override(en / nl / fr),theme_override(auto / light / dark),show_help, in the card header. Before this release these options never reached the cards; now they work. - Your language everywhere — English, Dutch, French; numbers in your own format (5.27 or 5,27).
- Fixes you will see:
- Simple card: ΔT of the last cycle now shows (it was always "—"); Q is the real quality score.
- Summary: no longer always Dutch; real defrost events per day.
- Pattern analysis: the tab you tap is the tab you see.
- Heating curve: clean points with a tooltip, a blinking live point, and the curve stays visible during hot water.
- LWT fine-tuning: keeps your tab and pop-up when it refreshes; blue = lower, green = optimal, red = raise.
- No longer needed: mushroom-template-card (stack-in-card, button-card and card-mod are enough).
| Card | YAML | Docs | Changelog |
|---|---|---|---|
| Simple card | simple-card.yaml | README | CHANGELOG |
| COP & energy | cop-energy.yaml | README | CHANGELOG |
| Summary | daily-summary.yaml | README | CHANGELOG |
| Daily summary (tabbed) | daily-summary-tabbed.yaml | README | CHANGELOG |
| Heating curve | heating-curve.yaml | README | CHANGELOG |
| LWT fine-tuning | card.yaml | README | CHANGELOG |
| Pattern analysis | pattern-analysis.yaml | README | CHANGELOG |
Settings shared by all cards (header, language, dark mode, variables): Shared card settings.
✅ After updating
- Update the integration: HACS → Daikin Cycle ML → Update, then restart Home Assistant. The first start runs a one-time clean-up: it rebuilds lost runs, rescores all cycles and recalculates the daily summaries. That start can take a little longer than usual.
- Re-paste every card you use (links in the table above): Edit dashboard → card → ⋮ → Edit in YAML, replace everything with the new YAML, Save. The old cards keep working but miss the fixes, the model header and dark mode.
- (Optional) Set your preferences in the card header's
variables:—lang_override,theme_override,show_help. Cards now follow your Home Assistant UI language first; setlang_override: nl(orfr) if you want the integration language instead. The COP & energy card now follows the HA dark mode by default (theme_override: lightkeeps the old look). - (Optional) Remove mushroom if no other card of yours uses it.
- Check: Download diagnostics →
coordinator.runtime_check.coverage_pctshould...
v1.23.3
🌟 What's new in v1.23.3?
More accurate COP and backup-heater numbers, and alerts that fire when they should. No breaking changes.
- Backup heater on its own power meter? New optional BUH power sensor (Configure → Device → Sensors). Its power is added to your heat pump meter, so COP no longer reads too high while the backup heater runs.
- No more false "Backup heater excessive" alerts. The alert now compares backup-heater minutes with compressor minutes. Before, one long run with a few minutes of backup heater counted as a full "backup heater day".
- Native ESPAltherma counters are kept when the integration switches to the ESPAltherma v2 entities (compressor runtime, defrosts, backup heater and starts no longer fall back to estimates).
- "COP vs datasheet" works again with the ESPAltherma v2 entities (it stayed empty) (#388).
- "Short run" alerts work with adaptive thresholds. If many of your runs are short, the learned threshold could drop to 2 minutes and the alert never fired. It now never goes below half of your own setting (#408).
- Cleaner COP history. Old impossible COP readings (stored before the COP checks existed) are cleaned up once, so week/month COP no longer shows values like 0.47 or 16.5 (#408).
- Your own COP sensor, checked. Diagnostics now compare the COP in use with the integration's measurement and ESPAltherma's estimate, so a template COP sensor that reads high is easy to spot (#408).
✅ After updating
- Update via HACS → Daikin Cycle ML → Update, then restart Home Assistant. The first start runs a one-time clean-up of old COP data and daily summaries (a few seconds).
- Backup heater on a separate meter? Configure → Device → Sensors → BUH power sensor → pick that meter.
- Using your own COP template sensor? Consider clearing COP sensor: the integration's own COP (your power meter + ESPAltherma heat, from one snapshot) is usually more accurate. Compare first in Download diagnostics →
cop_cross_check. - No card changes needed.
Found something? Open an issue and attach your diagnostics.
📚 Docs: Configuration — sensors · Alerts — buh_excessive · ML — adaptive thresholds
🛠️ Technical Details & Fixes
Added:
- BUH power sensor -
buh_power_sensor_entity(options Device → Sensors, EN/NL/FR); added topower_sensor_entityin_power_from_meter(W/kW normalised, invalid/idle ignored); compressor detection unchanged; attributesbuh_power_sensor,buh_meter_w. PR #410. - COP cross-check - diagnostics
coordinator.cop_cross_check(used / flow × ΔT ÷ power / ESPAltherma estimate / deviation %). PR #409.
Fixes:
- AUTO source tie - when v2 wins a 12/12 tie (rule unchanged) the monolithic sensor's extra attributes are merged in (individual wins per key);
source_resolution.supplemented_from_mono. PR #407. - #388 Datasheet normalisation - outdoor temperature via the alias layer (
Outdoor air temp.(R1T)/R1T-Outdoor air temp.);last_recordonly exists on the closing tick. PR #407. - buh_excessive - ratio = BUH minutes (
runtime_acc.<day>) / compressor minutes (daily_summary) over 7 d, as documented; cycle ratio kept as fallback (prod-1: 0.444 → 0.011). PR #407. - #408 Diagnostics follow-ups - effective short-run threshold ≥ half the configured value; migration v30 flags pre-gate COP samples (>
MAX_COP, heating/DHW <COP_FLOOR) and rebuilds/deletes theircop_hourlyhours;daily_summary.duration_min= within-day portion (≤duration_max), all days re-rolled once. PR #409.
Tests: 6,774 tests, 100% coverage; new suites reproduce prod-1 v1.23.2 diagnostics (AUTO tie, normalisation without last_record, BUH ratio), the learned-threshold collapse, the v30 migration on a real database, and the BUH power sensor.
Changelog: CHANGELOG.md | Compare v1.23.2...v1.23.3
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.23.2
🌟 What's new in v1.23.2?
One more fix for heat pumps that run through the night. No breaking changes.
- A run that is still going now shows up in the daily charts (#400). Before, a long run (for example started at 22:00 and still running at 08:00) only appeared in the per-day charts once the compressor stopped. Now its runtime shows on each day it covers while it runs, updated every 15 minutes. The Today sensors were already correct.
- Good to know: a run is never cut at midnight. It stays one run; the integration splits its runtime over the days it covers.
✅ After updating
- Update via HACS → Daikin Cycle ML → Update, then restart Home Assistant.
- No card changes needed (if you skipped v1.23.1: re-paste Daily Summary Tabbed and Pattern Analysis from
dashboard/cards/).
🙏 Thank you
- CuttingStorm (Tweakers forum) — for the run-across-midnight report that led to this check (#400)
Found something? Open an issue and attach your diagnostics.
📚 Docs: Database — daily summary semantics
🛠️ Technical Details & Fixes
Fixes:
- #400 Running cycle in the live rollup - cycle rows are written on close only, so the 15-min live rollup (R782) re-rolled today without the running cycle.
async_rollup_day(open_cycle=...)now adds the detector's open cycle (runtime + longest within-day portion; count / quality / BUH / defrost still on close), and the live rollup rolls every day the open cycle covers (DST-safe). PR #405.
Tests: 6,722 tests, 100% coverage; new suite runs a real detector 22:00 → 03:00 through the live rollup (charts at 01:00 and after close).
Changelog: CHANGELOG.md | Compare v1.23.1...v1.23.2
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.23.1
🌟 What's new in v1.23.1?
A fix release for the daily overview when your heat pump runs past midnight, fewer lost measurements, and better self-diagnosis. No breaking changes.
- Runs past midnight no longer mess up the daily overview (#400).
- The morning after a night run no longer shows quality 0 %. A day that only contains the tail of last night's run now shows "—".
- "Longest run" can no longer be longer than the day's total runtime: each day now shows its own part of a run that crosses midnight.
- Defrosts and backup heater show up in the Events tab (#400). The per-day event counts now use the same real numbers as the Today block, so they are no longer empty or 0 %.
- No more lost COP measurements (#353). If the database is briefly busy, measurements are kept and saved a moment later instead of being dropped.
- Lighter nightly maintenance (#353). The database is only compacted when there is actually space to reclaim.
- Better diagnostics (#353). If an update is ever slow, the diagnostics file now shows when it happened and which step caused it.
- All default thresholds in one place (#193). See
docs/cycle-thresholds.json.
✅ After updating
- Update via HACS → Daikin Cycle ML → Update, then restart Home Assistant.
- Re-paste two dashboard cards to get the Events fix: Daily Summary Tabbed and Pattern Analysis. Copy the latest YAML from
dashboard/cards/(HACS does not update cards you pasted). Other cards need no change.
🙏 Thank you
- CuttingStorm (Tweakers forum) — spotted the 0 % quality, "longest > total" and empty event counts after a run across midnight (#400)
Found something? Open an issue and attach your diagnostics.
📚 Docs: Database — daily summary and maintenance · Dashboard cards · Cycle thresholds
🛠️ Technical Details & Fixes
Fixes:
- #400 Daily summary across midnight -
daily_summary_recent.quality_avgisnullfor a row without cycles started that day;daily_summary.duration_maxis the longest within-day portion (always ≤total_duration_s);_apply_mergekeeps MAX ofbuh_used/defrost_used/dT_maxof absorbed runs; newevents_by_day(BUH minutes, defrost events fromruntime_acc.<day>) used by Daily Summary Tabbed Events and Pattern Analysis. PR #401. - #353 A3c slow tick - diagnostics
coordinator_timing.tick_duration_max_ts/_tick/_phases_ms; cop samples hittingdatabase is lockedare queued (max 240) and flushed on the next insert (database.pending_cop_samples); nightly VACUUM only when ≥ 20 % pages are free, explicitrun_maintenancevacuum: truealways runs (database.last_vacuum). PR #402. - #193 Canonical thresholds -
docs/cycle-thresholds.json(schema 1) with a drift test againstconst.py/engine/physics.py; Carnot factor 0.65. PR #403.
Tests: 6,703 tests, 100% coverage; new suites for cross-midnight daily summary (incl. Node-rendered cards), real SQLite write-lock reproduction, conditional VACUUM, slow-tick attribution and threshold drift.
Changelog: CHANGELOG.md | Compare v1.23.0...v1.23.1
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
1.23.0
🌟 What's new in v1.23.0?
Dashboards that work in every language, more honest COP numbers, corrected Daikin datasheets, and first support for bivalent (heat pump + boiler) systems. No breaking changes.
- Dashboards work in Dutch and French. Cards stayed empty when Home Assistant created your entities in Dutch or French (e.g.
cyclus_statusinstead ofcycle_state). Every card now finds its sensors in any language — even after you rename them. Reported in #347 and #369 — thank you @JoHuTi and @MBGek! - More realistic COP. Impossible readings from an external COP sensor (above 11 or infinite) are ignored, and short start-up dips below 1.0 no longer drag your averages down.
COP today, weekly/monthly COP, SPF and heating-curve advice now only use trustworthy samples (#388). - Corrected Daikin datasheets. 24 models now use official Daikin figures (with source), including EPRA, ERGA, ERLA, EDLA, EBLA, EBLQ and the EVLQ hybrids. Hybrids no longer get a false "COP below datasheet" alert (#392).
- Hybrid units recognised by their indoor code. EHYHBH / EHYHBX / CHYHBH / Rotex codes (e.g.
EHYHBX08AAV3) map to the right EVLQ model. - Bivalent systems (heat pump + external boiler). Enable
Bivalent Operationin ESPAltherma and the integration keeps boiler periods out of your heat-pump COP and stops false short-cycle alerts during boiler handovers (#288). Raised by @fransh47 in Discussion #176 — thank you!
✅ After updating
- Re-paste your dashboard cards. HACS updates the integration, not the cards you pasted. Copy the latest YAML from
dashboard/cards/into each card. - Check your model (Settings → Devices & Services → Daikin Cycle ML → Reconfigure). If you picked one of these, switch to your real model — they do not exist in Daikin's range: EPRA04/06/11/14/16/18EAV3, ERLA04/06/08/12DAV3, EPGA10DAV3. Your data is kept.
- Bivalent or hybrid? In ESPAltherma uncomment
Bivalent Operation,Boiler Operation DemandandBoiler DHW Demand— see Configuration §4.8.
🙏 Thank you
A big thank you to everyone who reported problems and shared diagnostics — this release exists because of you:
- @JoHuTi — empty dashboards on a Dutch install (#347)
- @MBGek — dashboards on a hybrid EVLQ08CV3, with diagnostics that also exposed the hybrid COP reference (#369)
- @fransh47 — bivalent heat pump + gas boiler setup (Discussion #176 → #288)
Found something? Open an issue and attach your diagnostics.
📚 Docs: Troubleshooting — empty cards (§4.6) and COP guards (§5.1) · Model compatibility · Dashboard cards
🛠️ Technical Details & Fixes
Fixes:
- #347 / #369 Dashboards empty on non-English installs - all 7 cards resolve entities via English id → entity-registry
translation_key→ NL/FR id; Jinja headers viaintegration_entities(). English output byte-identical. Pattern Analysis Dutch aliases fixed (cop_sww_*,cop_degradatie_week). PR #390. - #388 COP chain impossible values -
cop_sensor_entitygated on finite and ≤MAX_COP; classifier-flagged samples (is_transition_artifact=1) excluded fromcop_today, hourly rollup, SPF, heating-curve advice and hybrid means;cop_normalized_a7w35raw-attribute fallbacks. PRs #389, #391. - #369 Heating/DHW COP < 1.0 stored flagged via
physics.below_cop_floor(Carnot classifier contract unchanged). PR #394. - #392 Datasheet audit - Daikin product-table values with
ref+verifiedfor 24 models; ERGA/EBLA second point relabelled A7/W55 → A7/W45; EVLQ A7/W35 nominal 5.04 / 4.45 (was 5.85 / 5.75);erga08evh7datasheet; 11 non-catalogue models flaggedverified: false(kept). PRs #393, #394. - #288 Bivalent support (phase 1) - boiler signals trusted when the source publishes
Bivalent Operation; ERGA EVH aliases. PR #395. - ESPAltherma alignment -
MODEL_ATTRIBUTES.jsonat betab613d10; daily model-compat drift job fixed (missing label). PR #393.
Follow-ups: #397 remaining datasheet verification (EPGA, EDLQ, EBLQ011–016, EPSK, ERRQ) · #398 bivalent phase 2 (wizard, bivalent_active, boiler energy, alerts).
Tests: 6,649 tests, 100% coverage; new suites for dashboard entity resolution (Node-executed card JS, real HA template engine), datasheet audit, config-flow E2E for all 60 models, COP gate/floor and bivalent signals.
Changelog: CHANGELOG.md | Compare v1.22.10...v1.23.0
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.22.10 - Bug fixes + diagnostics hardening
🌟 What's new in v1.22.10?
A focused patch release fixing 4 bugs since v1.22.9. No breaking changes.
- Reliable source health.
source_staleno longer fires false alarms on stable weather. Tracks publish timestamp, not just value changes. - Ready for HA 2027. Diagnostics no longer uses deprecated device-registry APIs removed in HA 2027.9 / 2027.10.
- Cleaner Recorder history.
sensor.alerts_todayattributes exceeded HA Recorder's 16 KB limit on busy days. Now size-capped in 3 tiers. - Visible notify health.
notify_healthdiagnostic is now emitted. See per-service fail streaks without reading logs.
🛠️ Technical Details & Fixes
Fixes:
- #379 source_stale false positive - uses
State.last_reported. - #381 HA 2027 deprecations -
async_entries_for_config_entry. - #382 alerts_today 16 KB Recorder - 3-branch size cap.
- #384 notify_health missing - now emitted.
Changelog: CHANGELOG.md | Compare v1.22.9...v1.22.10
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.22.9 - Notification Polish & Stability
🌟 What's new in v1.22.9?
This release focuses on stability, cleaner logs, and a smoother dashboard experience.
- Smarter Notifications: Missing notify services (e.g.,
notify.telegram) no longer flood your Home Assistant logs with tracebacks. They are now silently handled, deduped, and surfaced via a clean HA repair if they persist. - LWT Fine-Tuning Card: The dashboard cards, including the LWT Fine-Tuning card, continue to deliver precise heating curve advice. We've resolved background flakiness that could occasionally cause dashboard updates to stall.
- General Polish: Cleaned up underlying test frameworks and scripts to ensure future updates deploy even faster.
🛠️ Technical Details & Fixes
Fixes:
- #367 - ServiceNotFound in notify dispatch bypasses dedup state — Fixes log spam and adds
notify_healthdiagnostic property. - #366 - Thread leak in live_rollup teardown — Fixes flaky test coverage causing CI blocks.
- Chore - Fixed
resolve_issue.pyscript to prevent false release claims and typo generation (vunreleased).
Changelog: View the full CHANGELOG.md for a complete list of changes.
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.22.8 — Accurate compressor runtime & smarter cold-snap advice
v1.22.8 — Accurate compressor runtime & smarter cold-snap advice
⭐ If this project helps you, consider giving it a star.
What you'll notice
Your compressor runtime is finally accurate
sensor.daikin_cycle_ml_runtime_compressor_today was underreporting. It only counted completed cycles that started today — missing overnight cycle tails and the currently-running cycle.
On our reference system, the sensor showed 3.4 h when ESPAltherma reported 10.3 h.
Now: the sensor matches the firmware counter within 0.6 %. If you have a fixed-scale card for this sensor, bump the maximum to 12 h or higher.
sensor.daikin_cycle_ml_duty_cycle_today (derived from the above) also corrects: from 33 % → 99.8 % on a continuous-compressor day, with its band attribute correctly flipping to saturated.
No more "lower your setpoint" advice during cold snaps
During a cold snap, if the backup heater (BUH) is running, the system used to still advise "lower the LWT" — advice that would push BUH usage higher and reduce efficiency. A 1-hour cached analysis wasn't being invalidated when BUH kicked in.
Now: heating_curve_advice returns state=keep with reason=buh_active while BUH is running, and flips back the moment BUH stops. A new suppressed_by attribute shows exactly why.
Smoother Home Assistant
~5 % of coordinator ticks were blocking the HA Core event loop for 7–25 s, briefly starving other integrations. Root cause: a COP-analysis query running every 30 s → 5 760 DB queries/day.
Now: that query is throttled to once per minute — 50 % fewer queries, fewer slow ticks.
Sample rate is now honest
Diagnostics showed samples_last_24h: 198 vs samples_expected_24h: 2880 (looked like 6.9 %). But the system had only been up for 103 minutes — the real rate was 95.7 %.
Now: diagnostics show samples_window: since_start with sample_rate_pct: 95.7 when uptime is under a day, and samples_window: 24h once it exceeds a day.
Under the hood
Fixes you won't see but should know about:
- Native-first counters. Whenever ESPAltherma reports a runtime counter (compressor, BUH, defrost, starts), that value now wins over our computed fallback. Every counter exposes
source: native/source: computedso you always know where the number came from. _read_floathardened. Firmware values oftrue,NaN, orinfare rejected cleanly instead of silently poisoning calculations.- BUH attribute leak.
buh_step1_s/buh_step2_s/buh_total_sno longer appear onruntime_compressor_today(they belong onruntime_buh_today).
Do I need to do anything?
No. Auto-updates via HACS. After upgrade:
runtime_compressor_todaycorrects within ~30 s.- Fixed-scale cards for that sensor → bump
maxto 12 h or higher. - Historical
daily_summaryrecords are unaffected — only today's live values change.
Known open issue
The tick p95 residual (25 s max) isn't fully eliminated — the throttle removes ~50 % of the load, but some slow ticks remain. Tracking on #353. If you're on slow bind-mount storage and notice lag, please attach iostat -x 5 output to that issue.
🔧 Technical Notes
Related PRs
| PR | Title |
|---|---|
| #361 | A3a sample-rate label + A3b cop_today throttle |
| #362 | A3 tick metrics decision record |
| #363 | A4 BUH excessive troubleshooting |
| #364 | Release v1.22.8 |
| #365 | CHANGELOG sync |
Issue: #353 — A1/A2/A3/A4 tracker
Design record: docs/design/a3-tick-metrics.md
Files Changed
| Area | File | Change |
|---|---|---|
| Engine | engine/cop_analyzer.py |
A2 BUH suppress |
| Engine | engine/runtime_source.py |
native-first resolvers |
| Coordinator | coordinator.py |
A3a _record_tick_health, A3b _refresh_cop_today |
| Sensor | sensor.py |
_resolve_attrs, source attr |
| Storage | db.py, store.py |
cycles_intersecting |
| Diagnostics | diagnostics.py |
samples_window, sample_rate_pct |
| Tests | tests/test_v1228_*.py |
new (A1/A2/A3) |
| Docs | docs/design/a3-tick-metrics.md, README.md |
new |
CI & Quality
- 6 093 tests, 100 % coverage
- Ruff: 0 errors · Mypy: clean
- CI matrix: 9/9 green (ruff, mypy, coverage, build, HACS, Hassfest, CodeQL, guard)
Residual (A3c)
25 s max-tick origin not fully eliminated. Needs prod-1 iostat -x 5 + grep daikin_cycle_ml /home-assistant.log. Tracked on #353.
☕ Support
v1.22.7 — BUH counter fix: wall-time + native-first
v1.22.7 — BUH counter fix: wall-time + native-first
🐛 What Was Wrong
runtime_buh_today was 2× too high when both BUH Step1 and Step2 were active. A prod-1 user reported 90 min of backup-heater runtime on a day the firmware counter showed 42.6 min.
Root cause: each tick added dt_s (30s) to both _buh_step1_s and _buh_step2_s, then the sensor summed them — 60s per 30s wall-time.
Related: ESPAltherma v2 exposes precise firmware counters (Stats: Backup heater today, Stats: Defrosts today, Stats: Compressor starts today) that were unused. We now prefer them with computed fallback.
✅ What Changed
Wall-time counter (core fix)
- New
_buh_total_s— max 1×dt_sper tick regardless of step1/step2 runtime_buh_todayreads_buh_total_sinstead ofstep1_s + step2_sstep1_s/step2_skept as diagnostic detail in attrs- Restore migration: pre-v1.22.7 payloads (no
buh_total_s) backfill viamax(step1, step2)
Native-first for ESPAltherma v2
New engine/runtime_source.py with strict resolvers:
| Counter | Priority |
|---|---|
| BUH today | Stats: Backup heater today (min) × 60 → computed |
| Defrosts today | Stats: Defrosts today → computed |
| Compressor starts | Stats: Compressor starts today → computed |
| Compressor runtime | computed primary (firmware resets on reboot) |
Each sensor exposes a source attribute (native / computed).
Hardening
- Strict
isinstance(attrs, dict)— reject MagicMock (float(Mock) == 1.0false positive) - Safe
isinstance(snap, dict)guards in sensor accessors
CI infrastructure
- Pre-push hook skips full pytest suite on branch-deletion pushes — from 500s+ to <2s.
Not changed
buh_excessivealert — 7-day ratio, not today's counter- COP / heat / electricity today — already use
power_sensor+cop_sensor - Defrost duration — no native field on ESPAltherma
- Compressor runtime — computed-primary per user preference
🧪 Testing & quality
- 5,960 tests pass, 100% coverage
- 20 new tests in
test_v1227_buh_walltime.py - Suite: 504s (was 528s, −4.5% via
--dist worksteal) - Ruff clean on changed files
- No regression — v1.22.6 FR / DAV3 / LWT card preserved
📊 Impact
| User | Before | After |
|---|---|---|
| Prod-1 (v2 ESPAltherma) | 5436 (90.6 min) |
2556 (42.6 min), source: native |
| v2 with Stats: | computed only | firmware-exact, source: native |
| v1 (no Stats:) | double-counted | wall-time computed, source: computed |
⚙️ Upgrade
No action required. Auto-updates via Marketplace (HA 2026.11+) or HACS.
After upgrade, runtime_buh_today corrects from next coordinator tick (~30s). Historical daily_summary per-cycle records unaffected.
🔗 Links
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs:
v1.22.6 — French translations, 3 new Altherma models, and a huge thank you
v1.22.6 — French translations, 3 new Altherma models, and a huge thank you
🇫🇷 Full French support
Every part of the integration is now available in French:
- Complete UI — 873 translation keys, parity with English (config_flow, options, repairs, services)
- Notification reports — status reports, stooklijn advice, COP-low alerts
- Alert texts — titles, labels, advice texts, severity labels, KMeans advice
- 6 dashboard cards — Simple, Heating Curve, COP & Energie, Pattern Analysis, Daily Summary Tabbed, LWT Fine-Tuning
- Auto-locale — picks French from
hass.language=frornotification_language=fr
🆕 3 Daikin Altherma 3 H HT models
- EPRA14DAV3 — Altherma 3 H HT split, single-phase
- EPRA16DAV3 — Altherma 3 H HT split, single-phase
- EPRA18DAV3 — Altherma 3 H HT split, single-phase
All three ship with verified EN14511 datasheets from Daikin (ETVH/ETVX/ETBH/ETSH16 indoor units).
🐛 What Changed
- Leaving water temp. before BUH (R1T) — now consumed as
thermal_power.physics.dt_from_attrsprefers compressor-side supply temp over R2T, improving ΔT accuracy on ESPAltherma v2 sources. - notification_engine — accepts
'fr'in language validation (options.notification_language). - Language detection — added
frbranch to LWT card, Daily Summary Tabbed, and alert engine. - Tests — 4 test files updated for +1 consumed attribute (21→22 operational, 25→26 consumed).
- New guard-test —
build_rich_alertinvalid-language fallback (coverage for thelanguage='xx'path).
🧪 Testing & quality
- ~5,926 tests pass, 100% coverage (branch + statement)
- Ruff clean on all changed files
- Full key parity — 873/873 French keys match English
- No regression — v1.22.5 fixes preserved (options flow, source_stale, dhw_installed)
🙏 A huge thank you to @azsono03
This release is 90% the work of @azsono03.
He:
- Translated the entire integration into French — 873 keys, 6 dashboard cards, all notifications
- Contributed 3 new Daikin Altherma 3 H HT models with verified EN14511 datasheets
- Fixed a subtle R1T-before-BUH consumption bug that affected ESPAltherma v2 users
- Did all this with remarkable attention to French typography (correct −, –, ×) and consistency with the existing Dutch translation
If French support is useful to you, please thank him. His fork: github.com/azsono03/daikin_cycle_ml.
⚙️ Upgrade
No action required.
- Home Assistant 2026.11+: auto-updates via Marketplace
- Home Assistant 2026.10 or older: auto-updates via HACS as before
If you switch notification_language to fr, the UI will follow. No restart needed.
🔗 Links
☕ Support
Daikin Cycle ML is free, local-only, and open-source. If it has helped you optimize your heat pump or save on energy costs, consider supporting the project:

