[segran2]

Summary

Improve NIBE F750 telemetry using parameter data observed on real F750 hardware through MyUplink Cloud API v2.

The change:

• corrects parameter 40013 documentation to BT7 hot-water top temperature
• exposes parameter 40940 as canonical hp_degree_minutes
• ignores NIBE's -32768 unavailable-value sentinel instead of emitting bogus telemetry
• removes embedded U+00AD soft hyphens from generated metric names
• adds an F750 regression test based on hardware-observed parameter data

No F750 electrical compressor-power mapping is added. Parameter 10012 was not present in the hardware capture, so the driver does not infer electrical power from unrelated parameters.

Driver evidence

• Driver id: myuplink
• Device make/model: NIBE F750
• Firmware tested: hardware-tested F750 via MyUplink Cloud API v2
• Protocol/register source: MyUplink Cloud API v2 /v2/devices/{deviceId}/points
• Hardware test performed: Yes — live F750 telemetry captured and used to build the regression fixture
• Known limits: No verified F750 electrical compressor-power parameter was observed; no hp_power_w mapping is claimed for F750

Observed parameters include:

• 40004 — BT1 outdoor temperature
• 40013 — BT7 hot-water top temperature
• 40033 — BT50 room temperature
• 40940 — current degree-minute value (DM)
• 41778 — current compressor frequency
• 43084 — internal additional-heat power
• -32768 — unavailable-value sentinel observed on multiple points

Safety

• The driver starts read-only, or this is a separately reviewed control change.
• Vendor signs are converted at the driver boundary.
• Cached telemetry becomes stale instead of being re-emitted as fresh.
• No credentials, serial numbers, private addresses or site data are included.

The driver remains read-only. This PR adds no control/write path.

Control evidence

Not applicable — read-only telemetry change.

• HIL evidence or required: not applicable
• Control changes: not applicable; no control path changes.

Checks

• Commits include Signed-off-by.
• make test-driver ID=myuplink
• make check

Full make check: 4810 passed, 923 skipped.

All 91 drivers passed the sandbox safety check. Manifest, index, devices and support-status generation are clean. FTW baseline and host API checks also passed.

Main integration — 27 October 2026

Merge commit 378737133c354901058b48695519097166922175 integrates main 57984984a9bf6939d5206c5a05138296d9c6d940 while preserving the original F750 commit and author history. No rebase or force-push was used.

• Resolve the three conflicts in CHANGELOG.md, manifests/myuplink.yaml and generated index.yaml. Keep main's changelog and add the F750 entry under Unreleased; regenerate hashes, sizes and catalog/support files with the repository tools.
• Main's fix(myuplink): retry a failed read soon and keep a rotated token (1.2.4) #193 already uses myuplink 1.2.4 for failed-read retry timing and rotated-token persistence. Preserve that implementation and tests, and advance this additive F750 telemetry candidate to 1.3.0 with make bump-driver ID=myuplink LEVEL=minor, avoiding two different drivers with version 1.2.4.
• Preserve BT7 documentation, canonical degree minutes, unavailable-value filtering and soft-hyphen normalization. Include the reviewer's requested changelog clarification: unavailable canonical temperatures are omitted, not emitted as 0 °C.
• The driver remains read-only. No new F750 compressor-power mapping or control path is introduced.

Validation on the integrated tree:

• make test-driver ID=myuplink: 39 passed, 16 skipped; this includes the F750 fixture and main's poll-timing/token-persistence tests.
• make check: 4,990 passed, 936 skipped; generated catalog/manifest checks pass.
• Signed-channel version check against the published drivers-beta manifest at main 5798498: passed; only myuplink is reported changed, with a higher version.
• Whitespace check against main passes.

The previous hardware-capture evidence remains the basis of the F750 fixture. No new live-device acceptance is claimed for this merged candidate, and nothing was installed or published to a signed channel. New-head CI is checked separately from the older green runs. The earlier 4,810-test count above belongs to the original branch tree.

Final-head CI

All five GitHub jobs pass on 378737133c354901058b48695519097166922175: drivers, signed-channel version validation, DCO, public-boundary and history-secret-scan. Validation run. GitHub reports this PR conflict-free. The PR remains unmerged.

[miravoss26]

Mira's read (pr-steward sweep). 2-line summary: MyUplink F750 telemetry. BT7 relabel, degree-minutes as a canonical metric, the -32768 sentinel is ignored, and soft hyphens are stripped from metric names. The driver stays read-only (control: false unchanged).

Checks: the size (18431) and sha256 pinned in index.yaml and manifests/myuplink.yaml match the driver file at this head. The new harness test covers the F750 payload, including the sentinel and soft-hyphen cases.

Findings:

• [low] decode_temp used to emit 0 for a missing value and now emits nothing. That's the intent, but it changes what dashboards see for an absent point. Worth one changelog line.

Security screen: no secrets, no new hosts or endpoints.

Safe to merge from my read. Not auto-merge eligible (repo not on the POLICY allowlist), so a human merges.

[miravoss26]

Review: myuplink 1.2.3 → 1.2.4, NIBE F750 telemetry fix from hardware-captured data.

Correctness:

• 40013 relabeled BT6 → BT7 is comment-only (register id unchanged) — matches the hardware capture in the new test fixture ("Hot water top BT7").
• Real bug fixed: the three canonical temp metrics used decode_temp(...) or 0, so an unavailable NIBE point (-32768 sentinel) was emitted as a false 0 °C reading — exactly the invented-number pattern no_invented_numbers.lua exists to catch. New valid_raw() filters the sentinel before decode, and the metrics are now omitted (not zeroed) when unavailable. Same fix applied to the generic "everything else" loop, so it isn't just the three canonical ones.
• New hp_degree_minutes (param 40940) is additive and follows the existing canonical-metric pattern.
• Soft-hyphen (U+00AD) strip in sanitize_metric_name is a real fix too — the test fixture shows it live in vendor parameter names ("com­pressor fre­quency").
• No F750 electrical compressor-power mapping added, as the description says — param 10012 untouched, correctly left out given it wasn't in the hardware capture.

Tests: new Lua harness test against a real hardware-captured response fixture, verifying both the emitted canonical metrics and that the three sentinel-valued points (hp_reference_air_speed_sensor, hp_current_temperature_system_1, hp_set_point_temp_system_1_heat) are correctly absent. Python wrapper runs it via make check's lua harness. Good coverage for exactly the bug being fixed.

Security: no secrets, no new network destinations, read-only (control: false unchanged) — matches the PR's own safety checklist.

All 5 CI checks green (signed channel, public-boundary, drivers, dco, history-secret-scan). One thing for a human: GitHub shows this PR as conflicting with main right now — needs a rebase, not a content concern. Safe to merge from my read once that's resolved.

[miravoss26]

Reviewed. This fixes a real correctness bug: previously decode_temp(...) or 0 emitted a bogus 0 °C whenever NIBE's -32768 unavailable-value sentinel showed up on a present parameter — now those are correctly skipped instead of faked. New valid_raw() filters the sentinel consistently for both the named temp params and the generic per-parameter loop (and PARAM_DEGREE_MINUTES is added to the canonical-dedup set, so it won't double-emit). The U+00AD soft-hyphen strip in sanitize_metric_name uses the right UTF-8 byte pair (194, 173).

New Lua harness test + pytest wrapper cover: sentinel-filtered absence (3 cases), the new hp_degree_minutes metric, and the existing temp/power paths. Read-only driver, no control path touched, matches the PR's own safety checklist. CI green across drivers/signed-channel/public-boundary/dco/secret-scan.

Nothing blocking. Safe to merge from my read.