Repository navigation
Conversation
Pair fresh per-observation pilot evidence with Core feedback for srcfl/ftw#1874. Signed-off-by: Segran <goran@andersson-stehag.se>
miravoss26
left a comment
There was a problem hiding this comment.
Easee cloud driver 1.5.0: exposes the charger's confirmed pilot current (pilot_a, pilot_observed_at, pilot_confirmed_at_ms) and dynamic_limit_a as separate fields, only when the REST observation was confirmed recently and isn't in the future.
- Correctness, fine:
confirmed_currentrejects NaN, negatives, >80 A, missing source time, stale confirmation (REST_FRESH_MS) and future source times. Whenhost.unix_msis missing it skips the future check and emitspilot_confirmed_at_ms = 0, which the FTW side (#1905) treats as not fresh. That's the safe direction. - Low:
observation_confirmed_msis stamped innote_observationfor any wanted id, including values REST re-confirms unchanged. That's intended (the test name in #1905 says so), but it means "confirmed" = "REST returned it again", not "the charger re-sent it". Worth a comment in the code so nobody reads it as a fresh measurement. - Low:
spec/host-api-profile.jsongains four fields, so anything validating against the profile needs to pick them up. CI is green, so presumably done. - Security screen: no secrets, no new hosts or dependencies, no authz changes.
Safe to merge from my read. It pairs with srcfl/ftw#1905, which consumes these fields. Merge order doesn't matter, since old drivers keep the old contract.
Hardware validation on the exact PR commitTested
This run covered three-phase startup, pilot reporting and load-balancer attribution under a reduced pilot. Single-phase attribution, release of load balancing during a charge and communication-loss recovery were outside this run. |
miravoss26
left a comment
There was a problem hiding this comment.
easee_cloud 1.5.0: exposes the effective pilot current (pilot_a, pilot_observed_at, pilot_confirmed_at_ms) separately from the dynamic setting (dynamic_limit_a), and only counts a pilot as confirmed when REST or SignalR actually delivered it.
- Behavior change to check:
answering()now goes throughconfirmed_current, which also requires the observation to be withinREST_FRESH_MS, in range (0-80 A) and not future-dated. A stale-but-settled dynamic readback that was accepted before can now come back nil. Looks intended, but worth confirming it can't flip the existing limit attribution on a quiet charger. - The future-time check multiplies
value_atby 1000, so it assumessource_time_keyreturns epoch seconds. I couldn't confirm that from the diff, and the harness test only covers the 2099 case. host.unix_msis guarded, and the missing-pilot path emits 0 for the confirmation time. Good.- Security screen: no secrets, no new hosts or dependencies. Test data is synthetic. CI green.
Needs a human glance at the first point (freshness gate on the existing attribution path); otherwise fine from my read.
An Easee current-setting readback can remain at 15 A while the actual pilot and circuit allocation allow 13 A. Treating the setting as the pilot makes a real charging limit look unexplained. This adds independently confirmed pilot evidence for srcfl/ftw#1874.
Change
pilot_a,pilot_observed_at,pilot_confirmed_at_msanddynamic_limit_aseparately.Dependency
Stacked on #194 at
34347fb92b08474b52e6240b28e78498110ba580, targetingeasee-allocated-limit. This extends that work without rewriting or replacing it. Paired Core extension: srcfl/ftw#1905 (stacked on srcfl/ftw#1537). After driver integration, FTW still needs a separate pin/snapshot update to the merged driver commit; this PR does not publish a signed artifact.Validation
make test-driver ID=easee_cloud: 44 passed, 15 skipped.make check: 4,992 passed, 936 skipped; generated metadata and whitespace checks passed.On 27 October 2026 the paired logic ran locally on FTW v0.157.2-beta.1+local.2 with an Easee cloud charger and Audi e-tron. The live driver was the same logic with unpublished 1.4.1 metadata; this PR's 1.5.0 is a version/catalog change, not a second hardware-tested binary.
load_balancer_limit/info, and the actual EV sheet displayed “Limited by the charger’s load balancing” without a red warning.This is one-site evidence. Charger model/firmware identifiers were not captured, and no claim is made about the exact internal allocation algorithm or which household load triggered it. Communication-loss/recovery and release-of-load-balancing transitions were not hardware-tested here; freshness/order regressions are software evidence. Keep draft pending dependency integration, CI and remaining device identification/review. No site configuration, credentials, raw household history or device identifiers are committed.