v0.4.0-beta.2

Pre-release for community validation (#128 round 2). The beta.1 solar
field mapping is now capture-validated: on the tester's real 2026-07-01
capture the parser reproduces the AGL app's figures exactly (sold to grid
8.019 kWh / $1.363 vs the app's 8.02 / $1.36; consumption 6.072 / $2.254 vs
6.07 / $2.25). The mismatch reported against beta.1 was a window artifact
— a cumulative-since-backfill sensor compared against the app's
billing-period tile — not a data bug. Beta.2 adds bill-period sensors that
match the app tile directly.

Added

• Bill-period solar sensors (#128): Solar sold this period (kWh) and
  Solar feed-in credit this period (AUD) — computed from the generation
  statistics since the current bill period start, so they line up with the
  AGL app's "Sold To Grid" tile. They read unknown until the generation
  series has backfilled to the trailing rewindow (a mid-backfill partial
  could never match the app). Note: on quarterly billing the 30-day backfill
  window is shorter than the bill period, so these sensors cover the stored
  history only.
• Solar feed-in rate sensor (#128): AGL reports the feed-in tariff under
  gstExclusiveRates (FiT is GST-free) — the parser previously only read
  gstInclusiveRates, so the rate never surfaced. New Solar feed-in rate
  sensor (AUD/kWh) registers on solar contracts when the plan carries a
  feed-in rate.

Fixed

• Generation series now backfills independently of consumption (#128).
  Previously the fetch window was resolved from the consumption series'
  resume point alone, so an install that upgraded into solar support only
  ever received the trailing 7-day rewindow of export history — the older
  ~23 days of the 30-day backfill never arrived. Each series now resolves its
  own chunked fetch range from its own resume point; disjoint ranges skip the
  other series' days (no extra load on AGL's BFF).
• Zero-export days no longer stall the solar backfill. A successfully
  fetched day whose feed-in intervals are all zero (cloudy day, or a solar
  system newer than the 30-day backfill floor) now writes a single
  zero-delta marker row so the generation resume point advances; previously
  a full chunk of such days was refetched forever and the period sensors
  never unlocked. Found by Codex review on the beta.2 PR.
• tests/fixtures/solar_hourly_response.json replaced with the tester's real
  full-day capture (48 slots, 11 non-zero export slots, mixed
  normal/peak feedIn types); reconciliation totals are locked in as
  regression tests.

---

---

What to test (solar users)

This pre-release is gated on community validation in #128.

After upgrading, expect the two new period sensors to read unknown for up to ~4 daily poll cycles — they stay silent until the generation history fully covers the billing period, by design.

1. Solar sold this period / Solar feed-in credit this period vs the AGL app's billing-period "Sold To Grid" kWh and $ — same window, should agree within rounding. (Quarterly billing: periods older than the 30-day backfill window are covered only partially.)
2. Per-day: pick a sunny day ≥3 days back; the Energy dashboard "Return to grid" bar should match the app's daily solar view.
3. Solar feed-in rate sensor should equal your plan's feed-in tariff (c/kWh ÷ 100).
4. History should keep extending daily, including across cloudy zero-export days, and should eventually reach days before the current billing period started.

Do not compare the cumulative Solar generation sensor against the app's billing-period tile — different windows by design.

Found a discrepancy? Report on #128 with a same-day triple: the day, the app's figures, and the sensor/dashboard figures.