feat: configurable incentive params, updated GloBird rates, options flow bugfix#1
Merged
Merged
Conversation
- Add aiohttp>=3.9.0 to manifest.json requirements (used by config_flow.py for Amber API validation via async_get_clientsession) - Add named incentive constants (INCENTIVE_ZEROHERO_CREDIT, INCENTIVE_SUPER_EXPORT, INCENTIVE_FREE_POWER, INCENTIVE_CRITICAL_PEAK_EXPORT, INCENTIVE_CRITICAL_PEAK_IMPORT, INCENTIVE_PEAK_SOLAR_FEEDIN, INCENTIVE_PROMPT_PAYMENT) to const.py - Add INCENTIVE_PARAMS dict mapping each incentive to its calculation parameters Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace raw string literals in incentive lists with INCENTIVE_* constants - Move incentive constant definitions before GLOBIRD_PLAN_DEFAULTS to avoid forward references - All 4 plans (zerohero, four4free, boost, glosave) now use named refs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Restore 4 deleted test files from git history (tariff_engine, amber_calculator, coordinator, helpers) - Remove SSH password from CLAUDE.md - Fix FOUR4FREE tariff_type: TARIFF_FLAT_STEPPED → TARIFF_TOU to match actual two-rate TOU import structure - Add FOUR4FREE branch to OptionsFlow async_step_incentives so users editing options retain their incentive toggles - Fix INCENTIVE_PARAMS Super Export window: 10am-2pm → 6pm-8pm to match tariff_engine.py implementation - Fix INCENTIVE_PARAMS critical_peak_import: align description and value (5c/kWh credit, not 10c surcharge) - Remove aiohttp from manifest.json requirements (ships with HA) - Fix MetricsWonSensor TypeError: add None guard for globird_import - Remove redundant inline dt_util import in ProviderDailyCostSensor - Add import_cost_today_c and export_earnings_today_c properties to TariffEngine, update coordinator to use public API Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…low fix - Fix OptionsFlowWithReload + add_update_listener conflict that crashed options editing on HA 2026.3+ (ValueError) - Make Super Export cap, window, and rate user-configurable (was hardcoded to 10kWh / 6-8pm / 15c — now defaults to 15kWh / 6-9pm / 15c per April 2026 fact sheets, backward-compatible for existing installs) - Make ZEROHERO credit window user-configurable (was hardcoded 6-8pm, now defaults to 6-9pm with auto-scaling threshold) - Update all plan default rates to April 2026 fact sheets: ZEROHERO 115.50c/day, FOUR4FREE 103.40c/day, BOOST 110.00c/day, GLOSAVE 88.00c/day (pre-fill only, users set their own rates) - Add Peak Solar Feed-in to ZEROHERO defaults (2c/kWh 4-11pm) - ZEROHERO export tariff changed to flat 0.00 (incentives replace TOU) - Fix pre-existing broken tests (supply_charge_today_c, stale date) - Bump version to 1.2.0 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (5)
📝 WalkthroughSummary by CodeRabbitRelease Notes – v1.2.0
WalkthroughThis PR refactors the PriceHawk integration to support configurable incentive parameters (time windows, rates, caps) for ZEROHERO and Super Export trackers. It removes automatic config-update handling, introduces new configuration UI fields, adds comprehensive test coverage across tariff engine and calculator components, updates translations and GitHub templates, and bumps the integration version to 1.2.0. Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~50 minutes Possibly related issues
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Remove docs/superpowers from tracking, update .gitignore to exclude dev-only files (PDFs, PROGRESS.md, project-overview.html, etc.). Rewrite README with badges, structured sections, full sensor/plan/incentive docs. Add GitHub issue templates for bugs, features, and rate updates. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This was referenced Apr 6, 2026
6 tasks
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…multi-plan layout Full rewrite of custom_components/pricehawk/www/dashboard.html (2447 -> 940 LOC) replacing the Amber-vs-current-plan two-comparator view with the multi-plan ranked layout from plan section 5.1. Visual language ported from assets/dashboard-v3-apple.html (dark default, Outfit + IBM Plex Mono, noise + ambient bg). Per-provider colour tokens (--amber-primary, --globird-primary) replaced with semantic ones (--accent-positive, --accent-negative, --accent-neutral) per the Phase 3.0 pivot away from provider-specific branding. Scaffold layout per plan section 5.1: - NAV bar (brand + connection status + clock + theme toggle) - HERO row: current cost card + savings-vs-best-alt card - PERIOD TABS: [Today][Week][Month*][3 Month][Year], active swaps data - RANKED ALTERNATIVES table: #/plan/peak/supply/saving, click -> drill - DRILL-IN CARD: per-plan stats + "Pin as Named Comparator" button - DATA HEALTH FOOTER: backfill state / days loaded / last ranking / count Entity reads are NOT wired yet — sample data renders so the scaffold is visually verifiable before commit 3.5/2 binds real sensor values. WebSocket connection logic copied verbatim from the previous dashboard: - WS URL derived from location.protocol (AEGIS rule: never hardcode ws://) - Token from URL params, postMessage, parent.hassConnection, or localStorage hassTokens (AEGIS rule: never hardcode the token) CSP connect-src extended to include ws(s)://*.local:* so the dashboard works on Ryan's HA Green at homeassistant.local (plan section 5.3 surprise #1). Existing localhost + Nabu Casa entries preserved. Active period tab persists to localStorage so re-opens land on the user's last view rather than defaulting to month every time. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…ll entity reads
Hooks the scaffold from commit 3.5/1 up to the Phase 3.2 / 3.3 / 3.4
sensors. After WebSocket auth completes, fires a get_states + subscribes
to state_changed events for the 16 tracked entities.
Tracked entities (per plan section 5.2 + Phase 3.3 / 3.4 worker notes):
- 5 x sensor.pricehawk_current_cost_{today,week,month,3month,year}
- 5 x sensor.pricehawk_best_alt_cost_{...} (NOT _best_alternative_cost_)
- 5 x sensor.pricehawk_savings_{...}
- 5 x sensor.pricehawk_named_cost_{...} (NOT _named_comparator_cost_)
- sensor.pricehawk_ranked_alternatives
- sensor.pricehawk_backfill_status
Hero row binding:
- Current cost card reads sensor.pricehawk_current_cost_<window>.
- Savings card reads sensor.pricehawk_savings_<window> and colours green
/ red / muted around the +/- $0.005 deadband.
- Best-alt name pulled from ranked_alternatives.attributes.alternatives[0]
(sensor is sorted ascending by cheap-rank score per summarize_for_sensor).
- Projected annual extrapolates active-window savings * 365/window_days.
Period tabs swap activeWindow and re-call renderHero() — all rollup
bindings re-evaluate against the new window's entity ID. Active class
mirrors localStorage so the tab UI stays in sync on cold loads.
Ranked alts table render:
- Pulls ranked_alternatives.attributes.alternatives[].
- Renders rank-pill (#1 gold), plan name + brand, peak rate, supply,
saving. Saving column only fills for the #1 plan (the cheapest); #2..N
show "—" because we don't have per-alt cost rollups — only the best-alt
rollup. Avoids fabricating numbers that don't match the sensor.
- Click row → drill-in card slides up below + plan ID persists in
selectedPlanId so re-renders after state_changed events preserve
selection.
- Empty state ("Waiting for the daily ranking job…") covers first-install
before the first ranking run completes.
Drill-in card:
- Stats grid: peak rate, daily supply, customer type, plan ID,
cheap-rank score (when present).
- "Pin as Named Comparator" deep-links to the integration's Configure
page (/config/integrations/integration/pricehawk). Per plan section 5.3
surprise #2 + plan section 9 REVISIT 4: HA doesn't support per-step
deep-linking; the deep-link is the locked UX for this phase.
Data Health footer renders backfill state with state-coloured value
(green=complete / amber=running / red=failed / muted=idle), days_loaded,
ranked_alternatives.last_run as relative + absolute time, and the
alternatives count.
Empty-state UI for first-run users (plan section 5.3 surprise #3): when
backfill_status.days_loaded < 7, hero rollup values are replaced with an
"Accruing… [n/365]" pill instead of showing $0.00 — surfaces clearly
that we don't have enough history yet rather than implying zero spend.
XSS hardening: all attribute-sourced strings (plan_id, display_name,
brand, customer_type) pass through escapeHtml() before innerHTML
insertion. Catches any future CDR registry payloads that include
HTML-ish characters in brand names.
30s setInterval re-renders the ranked + footer cards so the relative
timestamps ("ran 27s ago / 3h ago") tick forward without waiting for
the next state_changed event.
TDZ fix: the period-tab boot block previously called setActiveWindow()
before the entity state store consts were declared, which tripped a
ReferenceError on attrs in strict mode. Boot now defers the first full
render to the explicit boot block at the script bottom.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…multi-plan layout Full rewrite of custom_components/pricehawk/www/dashboard.html (2447 -> 940 LOC) replacing the Amber-vs-current-plan two-comparator view with the multi-plan ranked layout from plan section 5.1. Visual language ported from assets/dashboard-v3-apple.html (dark default, Outfit + IBM Plex Mono, noise + ambient bg). Per-provider colour tokens (--amber-primary, --globird-primary) replaced with semantic ones (--accent-positive, --accent-negative, --accent-neutral) per the Phase 3.0 pivot away from provider-specific branding. Scaffold layout per plan section 5.1: - NAV bar (brand + connection status + clock + theme toggle) - HERO row: current cost card + savings-vs-best-alt card - PERIOD TABS: [Today][Week][Month*][3 Month][Year], active swaps data - RANKED ALTERNATIVES table: #/plan/peak/supply/saving, click -> drill - DRILL-IN CARD: per-plan stats + "Pin as Named Comparator" button - DATA HEALTH FOOTER: backfill state / days loaded / last ranking / count Entity reads are NOT wired yet — sample data renders so the scaffold is visually verifiable before commit 3.5/2 binds real sensor values. WebSocket connection logic copied verbatim from the previous dashboard: - WS URL derived from location.protocol (AEGIS rule: never hardcode ws://) - Token from URL params, postMessage, parent.hassConnection, or localStorage hassTokens (AEGIS rule: never hardcode the token) CSP connect-src extended to include ws(s)://*.local:* so the dashboard works on Ryan's HA Green at homeassistant.local (plan section 5.3 surprise #1). Existing localhost + Nabu Casa entries preserved. Active period tab persists to localStorage so re-opens land on the user's last view rather than defaulting to month every time. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…ll entity reads
Hooks the scaffold from commit 3.5/1 up to the Phase 3.2 / 3.3 / 3.4
sensors. After WebSocket auth completes, fires a get_states + subscribes
to state_changed events for the 16 tracked entities.
Tracked entities (per plan section 5.2 + Phase 3.3 / 3.4 worker notes):
- 5 x sensor.pricehawk_current_cost_{today,week,month,3month,year}
- 5 x sensor.pricehawk_best_alt_cost_{...} (NOT _best_alternative_cost_)
- 5 x sensor.pricehawk_savings_{...}
- 5 x sensor.pricehawk_named_cost_{...} (NOT _named_comparator_cost_)
- sensor.pricehawk_ranked_alternatives
- sensor.pricehawk_backfill_status
Hero row binding:
- Current cost card reads sensor.pricehawk_current_cost_<window>.
- Savings card reads sensor.pricehawk_savings_<window> and colours green
/ red / muted around the +/- $0.005 deadband.
- Best-alt name pulled from ranked_alternatives.attributes.alternatives[0]
(sensor is sorted ascending by cheap-rank score per summarize_for_sensor).
- Projected annual extrapolates active-window savings * 365/window_days.
Period tabs swap activeWindow and re-call renderHero() — all rollup
bindings re-evaluate against the new window's entity ID. Active class
mirrors localStorage so the tab UI stays in sync on cold loads.
Ranked alts table render:
- Pulls ranked_alternatives.attributes.alternatives[].
- Renders rank-pill (#1 gold), plan name + brand, peak rate, supply,
saving. Saving column only fills for the #1 plan (the cheapest); #2..N
show "—" because we don't have per-alt cost rollups — only the best-alt
rollup. Avoids fabricating numbers that don't match the sensor.
- Click row → drill-in card slides up below + plan ID persists in
selectedPlanId so re-renders after state_changed events preserve
selection.
- Empty state ("Waiting for the daily ranking job…") covers first-install
before the first ranking run completes.
Drill-in card:
- Stats grid: peak rate, daily supply, customer type, plan ID,
cheap-rank score (when present).
- "Pin as Named Comparator" deep-links to the integration's Configure
page (/config/integrations/integration/pricehawk). Per plan section 5.3
surprise #2 + plan section 9 REVISIT 4: HA doesn't support per-step
deep-linking; the deep-link is the locked UX for this phase.
Data Health footer renders backfill state with state-coloured value
(green=complete / amber=running / red=failed / muted=idle), days_loaded,
ranked_alternatives.last_run as relative + absolute time, and the
alternatives count.
Empty-state UI for first-run users (plan section 5.3 surprise #3): when
backfill_status.days_loaded < 7, hero rollup values are replaced with an
"Accruing… [n/365]" pill instead of showing $0.00 — surfaces clearly
that we don't have enough history yet rather than implying zero spend.
XSS hardening: all attribute-sourced strings (plan_id, display_name,
brand, customer_type) pass through escapeHtml() before innerHTML
insertion. Catches any future CDR registry payloads that include
HTML-ish characters in brand names.
30s setInterval re-renders the ranked + footer cards so the relative
timestamps ("ran 27s ago / 3h ago") tick forward without waiting for
the next state_changed event.
TDZ fix: the period-tab boot block previously called setActiveWindow()
before the entity state store consts were declared, which tripped a
ReferenceError on attrs in strict mode. Boot now defers the first full
render to the explicit boot block at the script bottom.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…multi-plan layout Full rewrite of custom_components/pricehawk/www/dashboard.html (2447 -> 940 LOC) replacing the Amber-vs-current-plan two-comparator view with the multi-plan ranked layout from plan section 5.1. Visual language ported from assets/dashboard-v3-apple.html (dark default, Outfit + IBM Plex Mono, noise + ambient bg). Per-provider colour tokens (--amber-primary, --globird-primary) replaced with semantic ones (--accent-positive, --accent-negative, --accent-neutral) per the Phase 3.0 pivot away from provider-specific branding. Scaffold layout per plan section 5.1: - NAV bar (brand + connection status + clock + theme toggle) - HERO row: current cost card + savings-vs-best-alt card - PERIOD TABS: [Today][Week][Month*][3 Month][Year], active swaps data - RANKED ALTERNATIVES table: #/plan/peak/supply/saving, click -> drill - DRILL-IN CARD: per-plan stats + "Pin as Named Comparator" button - DATA HEALTH FOOTER: backfill state / days loaded / last ranking / count Entity reads are NOT wired yet — sample data renders so the scaffold is visually verifiable before commit 3.5/2 binds real sensor values. WebSocket connection logic copied verbatim from the previous dashboard: - WS URL derived from location.protocol (AEGIS rule: never hardcode ws://) - Token from URL params, postMessage, parent.hassConnection, or localStorage hassTokens (AEGIS rule: never hardcode the token) CSP connect-src extended to include ws(s)://*.local:* so the dashboard works on Ryan's HA Green at homeassistant.local (plan section 5.3 surprise #1). Existing localhost + Nabu Casa entries preserved. Active period tab persists to localStorage so re-opens land on the user's last view rather than defaulting to month every time. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 17, 2026
…ll entity reads
Hooks the scaffold from commit 3.5/1 up to the Phase 3.2 / 3.3 / 3.4
sensors. After WebSocket auth completes, fires a get_states + subscribes
to state_changed events for the 16 tracked entities.
Tracked entities (per plan section 5.2 + Phase 3.3 / 3.4 worker notes):
- 5 x sensor.pricehawk_current_cost_{today,week,month,3month,year}
- 5 x sensor.pricehawk_best_alt_cost_{...} (NOT _best_alternative_cost_)
- 5 x sensor.pricehawk_savings_{...}
- 5 x sensor.pricehawk_named_cost_{...} (NOT _named_comparator_cost_)
- sensor.pricehawk_ranked_alternatives
- sensor.pricehawk_backfill_status
Hero row binding:
- Current cost card reads sensor.pricehawk_current_cost_<window>.
- Savings card reads sensor.pricehawk_savings_<window> and colours green
/ red / muted around the +/- $0.005 deadband.
- Best-alt name pulled from ranked_alternatives.attributes.alternatives[0]
(sensor is sorted ascending by cheap-rank score per summarize_for_sensor).
- Projected annual extrapolates active-window savings * 365/window_days.
Period tabs swap activeWindow and re-call renderHero() — all rollup
bindings re-evaluate against the new window's entity ID. Active class
mirrors localStorage so the tab UI stays in sync on cold loads.
Ranked alts table render:
- Pulls ranked_alternatives.attributes.alternatives[].
- Renders rank-pill (#1 gold), plan name + brand, peak rate, supply,
saving. Saving column only fills for the #1 plan (the cheapest); #2..N
show "—" because we don't have per-alt cost rollups — only the best-alt
rollup. Avoids fabricating numbers that don't match the sensor.
- Click row → drill-in card slides up below + plan ID persists in
selectedPlanId so re-renders after state_changed events preserve
selection.
- Empty state ("Waiting for the daily ranking job…") covers first-install
before the first ranking run completes.
Drill-in card:
- Stats grid: peak rate, daily supply, customer type, plan ID,
cheap-rank score (when present).
- "Pin as Named Comparator" deep-links to the integration's Configure
page (/config/integrations/integration/pricehawk). Per plan section 5.3
surprise #2 + plan section 9 REVISIT 4: HA doesn't support per-step
deep-linking; the deep-link is the locked UX for this phase.
Data Health footer renders backfill state with state-coloured value
(green=complete / amber=running / red=failed / muted=idle), days_loaded,
ranked_alternatives.last_run as relative + absolute time, and the
alternatives count.
Empty-state UI for first-run users (plan section 5.3 surprise #3): when
backfill_status.days_loaded < 7, hero rollup values are replaced with an
"Accruing… [n/365]" pill instead of showing $0.00 — surfaces clearly
that we don't have enough history yet rather than implying zero spend.
XSS hardening: all attribute-sourced strings (plan_id, display_name,
brand, customer_type) pass through escapeHtml() before innerHTML
insertion. Catches any future CDR registry payloads that include
HTML-ish characters in brand names.
30s setInterval re-renders the ranked + footer cards so the relative
timestamps ("ran 27s ago / 3h ago") tick forward without waiting for
the next state_changed event.
TDZ fix: the period-tab boot block previously called setActiveWindow()
before the entity state store consts were declared, which tripped a
ReferenceError on attrs in strict mode. Boot now defers the first full
render to the explicit boot block at the script bottom.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
…multi-plan layout Full rewrite of custom_components/pricehawk/www/dashboard.html (2447 -> 940 LOC) replacing the Amber-vs-current-plan two-comparator view with the multi-plan ranked layout from plan section 5.1. Visual language ported from assets/dashboard-v3-apple.html (dark default, Outfit + IBM Plex Mono, noise + ambient bg). Per-provider colour tokens (--amber-primary, --globird-primary) replaced with semantic ones (--accent-positive, --accent-negative, --accent-neutral) per the Phase 3.0 pivot away from provider-specific branding. Scaffold layout per plan section 5.1: - NAV bar (brand + connection status + clock + theme toggle) - HERO row: current cost card + savings-vs-best-alt card - PERIOD TABS: [Today][Week][Month*][3 Month][Year], active swaps data - RANKED ALTERNATIVES table: #/plan/peak/supply/saving, click -> drill - DRILL-IN CARD: per-plan stats + "Pin as Named Comparator" button - DATA HEALTH FOOTER: backfill state / days loaded / last ranking / count Entity reads are NOT wired yet — sample data renders so the scaffold is visually verifiable before commit 3.5/2 binds real sensor values. WebSocket connection logic copied verbatim from the previous dashboard: - WS URL derived from location.protocol (AEGIS rule: never hardcode ws://) - Token from URL params, postMessage, parent.hassConnection, or localStorage hassTokens (AEGIS rule: never hardcode the token) CSP connect-src extended to include ws(s)://*.local:* so the dashboard works on Ryan's HA Green at homeassistant.local (plan section 5.3 surprise #1). Existing localhost + Nabu Casa entries preserved. Active period tab persists to localStorage so re-opens land on the user's last view rather than defaulting to month every time. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
…ll entity reads
Hooks the scaffold from commit 3.5/1 up to the Phase 3.2 / 3.3 / 3.4
sensors. After WebSocket auth completes, fires a get_states + subscribes
to state_changed events for the 16 tracked entities.
Tracked entities (per plan section 5.2 + Phase 3.3 / 3.4 worker notes):
- 5 x sensor.pricehawk_current_cost_{today,week,month,3month,year}
- 5 x sensor.pricehawk_best_alt_cost_{...} (NOT _best_alternative_cost_)
- 5 x sensor.pricehawk_savings_{...}
- 5 x sensor.pricehawk_named_cost_{...} (NOT _named_comparator_cost_)
- sensor.pricehawk_ranked_alternatives
- sensor.pricehawk_backfill_status
Hero row binding:
- Current cost card reads sensor.pricehawk_current_cost_<window>.
- Savings card reads sensor.pricehawk_savings_<window> and colours green
/ red / muted around the +/- $0.005 deadband.
- Best-alt name pulled from ranked_alternatives.attributes.alternatives[0]
(sensor is sorted ascending by cheap-rank score per summarize_for_sensor).
- Projected annual extrapolates active-window savings * 365/window_days.
Period tabs swap activeWindow and re-call renderHero() — all rollup
bindings re-evaluate against the new window's entity ID. Active class
mirrors localStorage so the tab UI stays in sync on cold loads.
Ranked alts table render:
- Pulls ranked_alternatives.attributes.alternatives[].
- Renders rank-pill (#1 gold), plan name + brand, peak rate, supply,
saving. Saving column only fills for the #1 plan (the cheapest); #2..N
show "—" because we don't have per-alt cost rollups — only the best-alt
rollup. Avoids fabricating numbers that don't match the sensor.
- Click row → drill-in card slides up below + plan ID persists in
selectedPlanId so re-renders after state_changed events preserve
selection.
- Empty state ("Waiting for the daily ranking job…") covers first-install
before the first ranking run completes.
Drill-in card:
- Stats grid: peak rate, daily supply, customer type, plan ID,
cheap-rank score (when present).
- "Pin as Named Comparator" deep-links to the integration's Configure
page (/config/integrations/integration/pricehawk). Per plan section 5.3
surprise #2 + plan section 9 REVISIT 4: HA doesn't support per-step
deep-linking; the deep-link is the locked UX for this phase.
Data Health footer renders backfill state with state-coloured value
(green=complete / amber=running / red=failed / muted=idle), days_loaded,
ranked_alternatives.last_run as relative + absolute time, and the
alternatives count.
Empty-state UI for first-run users (plan section 5.3 surprise #3): when
backfill_status.days_loaded < 7, hero rollup values are replaced with an
"Accruing… [n/365]" pill instead of showing $0.00 — surfaces clearly
that we don't have enough history yet rather than implying zero spend.
XSS hardening: all attribute-sourced strings (plan_id, display_name,
brand, customer_type) pass through escapeHtml() before innerHTML
insertion. Catches any future CDR registry payloads that include
HTML-ish characters in brand names.
30s setInterval re-renders the ranked + footer cards so the relative
timestamps ("ran 27s ago / 3h ago") tick forward without waiting for
the next state_changed event.
TDZ fix: the period-tab boot block previously called setActiveWindow()
before the entity state store consts were declared, which tripped a
ReferenceError on attrs in strict mode. Boot now defers the first full
render to the explicit boot block at the script bottom.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
* feat(coordinator): Phase 3.4 commit 1/2 — named comparator wiring Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(sensor): Phase 3.4 commit 2/2 — NamedComparatorRollupSensor × 5 Surfaces the Phase 3.4 commit 1/2 wiring as user-visible HA sensors: - `sensor.pricehawk_named_cost_today` - `sensor.pricehawk_named_cost_week` - `sensor.pricehawk_named_cost_month` - `sensor.pricehawk_named_cost_3month` - `sensor.pricehawk_named_cost_year` Subclasses the Phase 3.3 `PeriodRollupSensor` base; overrides `native_value` + `extra_state_attributes` to read the `"named"` key from `daily_cost_history` rather than extend the base's `_ROLLUP_KIND` dispatch enum (which Phase 3.3 just shipped — localise the new behaviour here instead of rewriting the base contract for one extra kind). Registration is CONDITIONAL on `"named" in coordinator._providers`, so users who haven't pinned a plan don't see 5 permanently- unavailable entities clutter their HA UI. Reads `_providers` directly (not `data["providers"]`) so the registration check fires on first setup before the coordinator has populated its data dict. CHANGELOG.md gains a Phase 3.4 block under [Unreleased] documenting the OptionsFlow step, the new sensors, lock/ranking interaction (none), and the persistence-through-rank-churn behaviour (the pin survives the plan dropping out of cheap-rank top-K because it lives in options, not in derived ranking state). strings.json + translations/en.json gain entity blocks for the 5 new sensors. Descriptions call out the tick-by-tick cadence difference vs the other ranked alternatives (which only refresh at daily rollover) so users know what they bought by pinning. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(named): persist provider across restart + skip unique_id collision Two CodeRabbit findings on the Phase 3.4 named comparator. 1. The named provider added to `self._providers["named"]` was constructed in `__init__` and `_apply_options_update` but never persisted or restored. Every other CdrPlanProvider in the coordinator (current plan, amber, flow_power, localvolts) round- trips through `_async_persist_state` / `async_restore_state`, so on HA restart their per-day accumulators (import_cost_today, kwh) survive — but `named` would reset to zero while the others kept state, so today's rollup deltas would lie until midnight rollover. Added named.to_dict() to the persist block and a from_dict() call in restore, using `today = dt_util.now().date()` (already in scope above) to satisfy the AEGIS rule that from_dict MUST receive an explicit HA-timezone date (no `date.today()` fallback). 2. `GenericProviderCostSensor(provider_id="named")` produces unique_id `{entry}_named_cost_today`. `NamedComparatorRollupSensor` for the "today" window produces the same unique_id (since its _ROLLUP_KIND is "named" and the base class composes `{kind}_cost_{window}`). HA's entity registry drops the second registration silently, breaking whichever sensor the dashboard depends on first. Added a `provider_id == "named"` skip in the providers loop in `async_setup_entry` — the named comparator is exposed via its dedicated 5-window rollup family instead. Also skips the matching GenericProviderRateSensor pair so we don't register orphan rate sensors for the "named" key. (Only one such loop exists; the Phase 3.3 rollup loop and Phase 3.4 named-rollup loop don't have the same collision since their _ROLLUP_KIND differs.) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
) * docs(planning): lock Phase 3.2 → 3.5 implementation plan 11-commit execution plan covering universal HA-history backfill (3.2), period rollup sensors (3.3), named comparator drill-in (3.4), and dashboard rewrite (3.5). Locks architectural decisions up front (daily_cost_history as single source of truth, named comparator as just another CdrPlanProvider, rollup as computed-on-demand) so the executing model doesn't re-derive them at each commit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(cdr): Phase 3.2 commit 1/4 — pure-logic HA history replay module Adds ``cdr/history_replay.py`` with three pure functions that establish the public API surface Phase 3.2 commit 2 wires the coordinator to. No HA imports — the module is unit-testable outside the integration runtime, matching the ``cdr/ranking_job.py`` pattern. Public API: - ``states_to_half_hour_slots`` — converts raw (ts, power_w, unit) tuples to evaluator-shaped slot dicts aligned to 30-min boundaries. Handles kW→W unit conversion, gap-protection clamping (6 min cap matches ``cdr/streaming.py``), and string-vs-float power values (HA's recorder serialises some sensor states as strings). - ``replay_day_through_plan`` — wraps ``evaluate()`` with the standard exception-swallow pattern (mirrors ``deep_rank``). Returns None on evaluator exception OR zero slot count. - ``fan_out_replay`` — generator yielding (date_str, {plan_key: aud}) per day. Streaming output keeps peak RAM at ~one day × N plans instead of all-days × N plans of full CostBreakdown objects. 25 tests covering boundary alignment, unit conversion, gap protection, sign handling, plan-failure isolation, date-order preservation, and opt-in entry_options pass-through. Pattern follows ``tests/test_coordinator_ranking.py``: stdlib only, no pytest-asyncio. Foundation for Phase 3.2 commit 2 (backfill.py rewrite) and Phase 3.3 (rollup sensors read the daily_cost_history rows this module helps populate). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(backfill): Phase 3.2 commit 2/4 — rewrite backfill.py as HA-side adapter Rewrites ``backfill.py`` from a 369-LOC Amber-API-coupled module into a thin coordination layer over the pure-logic ``cdr.history_replay`` fan-out. New public API: backfill_daily_cost_history(hass, grid_sensor_entity, plans, *, days_back=30, entry_options=None, existing_history=None) Internals pull recorder history day-by-day (NOT one big query — a 30-day single ``state_changes_during_period`` on a 1Hz grid sensor returns 100K+ State objects), convert to evaluator slots, fan out across N plans via the streaming generator, and merge per-date rows into the coordinator's ``daily_cost_history``. Final list is capped at 180 entries (matches live coordinator slice). Day-by-day queries are deliberate and NOT parallelised. HA's recorder uses a single executor pool so concurrent queries serialise anyway and just bloat task count. Per-query memory is bounded; the SQLite index on ``last_changed`` means 30 small queries are not meaningfully slower than 1 big one. This is commented inline so CR doesn't suggest parallelisation. Legacy ``fetch_amber_price_history`` retained — still used by ``coordinator._replay_amber_today_from_api`` to seed the Amber accumulator on a fresh install. The Phase 3.2 backfill itself no longer fetches Amber prices: Amber's role narrowed to a *truth overlay* written once daily by the live coordinator rollover. Test rewrite: 14 legacy Amber-API tests deleted (those exercised ``backfill_from_history``, ``_build_amber_price_index``, ``_find_amber_rate``, ``_parse_history_states`` — all removed). 14 new tests cover ``_local_date_string`` (AEST-safe formatting), ``_states_to_tuples`` (State + dict shapes), ``_merge_into_history`` (insert/merge/cap), and ``backfill_daily_cost_history`` end-to-end with the recorder mocked at the import boundary. ``tests/conftest.py`` extended with ``homeassistant.components``, ``homeassistant.components.recorder``, and ``homeassistant.components.recorder.history`` mocks so the lazy recorder import inside the backfill resolves under the test harness. ``__init__.py``'s ``handle_backfill`` service updated to call the new API with the current CDR plan; commit 4 will shrink it further to a one-line delegate through the coordinator (once the wrapper + status sensor land). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(coordinator): Phase 3.2 commit 3/4 — async_run_backfill + auto-kickoff Wires the new universal HA-history backfill into the coordinator and schedules its first run automatically after the initial ranking job completes. Coordinator additions: - ``build_backfill_plan_set`` — module-level pure helper (mirrors the ``cdr.ranking_job`` pattern) composing the {plan_key: plan_body} dict from current plan + top-K ranked alternatives. Lives outside the class so it's unit-testable without HA's app context (the coordinator's ``DataUpdateCoordinator[T]`` base gets mocked away by ``tests/conftest.py``). - ``_build_backfill_plan_set`` — thin instance wrapper. - ``async_run_backfill`` — coordinator-side run wrapper modelled on ``async_run_ranking_job``. Status-tracked via ``_backfill_status`` state machine (``idle | running | complete | failed``) plus ``_backfill_last_run_at``, ``_backfill_days_loaded``, ``_backfill_plans_replayed``, ``_backfill_error`` attributes that the Phase 3.2 commit 4 status sensor will surface. - Reuses ``_ranking_lock`` to serialise against the ranking job — both mutate ``_daily_cost_history``. REVISIT: split if contention observed in prod (cost of being wrong is brief serialisation of two rare operations). - Local import of ``backfill_daily_cost_history`` inside ``async_run_backfill`` (``# noqa: PLC0415``) so the HA recorder isn't loaded at module-import time. Matches the existing pattern at ``_replay_amber_today_from_api``. ``__init__.py`` kickoff: - After ``async_run_ranking_job`` task is scheduled, schedule a second task that AWAITS the ranking lock (so the first ranking run finishes and the alternatives list is populated) THEN runs ``async_run_backfill(days_back=30)``. Without the wait, the first backfill would replay history through only the current plan, missing the ``alt_*`` columns. Tests: 7 new in ``test_coordinator_helpers.py`` covering the pure helper — current-plan composition, alt_* prefix keying, plan-cache fallback to alt body, malformed-input skipping, empty/non-dict ``cdr_plan`` graceful return. Pattern matches the existing ``_extract_peak_rate_c_inc_gst`` test block in the same file. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(sensor): Phase 3.2 commit 4/4 — BackfillStatusSensor + service one-liner Final Phase 3.2 commit. Adds the status sensor users will read from their dashboards / automations and collapses the ``pricehawk.backfill_history`` service handler to a one-line delegate through ``coordinator.async_run_backfill``. New entity: - ``sensor.pricehawk_backfill_status`` — state machine read-through. State: ``idle | running | complete | failed``. Attributes: ``last_run`` (ISO timestamp), ``days_loaded``, ``plans_replayed``, ``error``. All sourced from the coordinator's status attributes written by ``async_run_backfill``. Service handler: - ``handle_backfill`` shrunk from a 60-LOC inline pipeline to defensive ``days`` coercion + a single ``await coordinator.async_run_backfill(days_back=...)`` call. Status tracking, recorder pulls, plan composition, and persistence all happen inside the coordinator method now; failures surface on the sensor rather than getting lost to log lines. Service description: updated ``services.yaml`` to reflect the new flow (replay-through-CDR-plan, no Amber API, status sensor pointer). Unused ``CONF_GRID_POWER_SENSOR`` import removed from ``__init__.py`` (the coordinator now owns the lookup). Tests: 4 new BackfillStatusSensor smoke tests in ``test_review_improvements.py`` exercising the property-read contract (state defaults to idle, running propagates, datetime → ISO, error attribute surfaces on failed runs). The sensor class itself can't be imported under the conftest mock tree (CoordinatorEntity + SensorEntity multiple inheritance from MagicMocks triggers a metaclass conflict), so the tests mirror the EXACT property bodies inline. Integration test on Ryan's HA will catch any drift. CHANGELOG.md updated under ``[Unreleased]`` documenting the full Phase 3.2 surface area: new module, rewritten backfill, status sensor, service signature change, recorder mocks in conftest. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(backfill): type-annotate target dict union for pyright `target = by_date.get(date_str)` infers as `dict[str, Any] | None` but the subsequent `target[plan_key] = aud` assignment narrows incorrectly without an explicit annotation. Annotate the union so pyright accepts both branches (None → fresh dict, dict → mutate). No behaviour change; pytest 760/760 still passes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(backfill): docstring accuracy on plan-set composition + recorder mock Two docstring corrections flagged by Sourcery — no behaviour changes. 1. `build_backfill_plan_set` previously claimed it returned `{}` when the current plan was missing. The implementation actually returns alternatives even without a current plan (alts-only backfill is intentional so rollup sensors can still surface comparative data). Updated the docstring to describe the real contract: callers must treat the absence of the current-plan entry as a "no-signal" condition for the active plan at that time. 2. `_patch_recorder` claimed a 3-tuple return `(get_instance_mock, history_call_mock, dt_util_now_mock)` but the helper actually returns a 2-tuple `(get_instance, history_mock)` (the dt_util mock was removed earlier when the backfill stopped depending on it). Fixed the docstring and added a precise `tuple[MagicMock, MagicMock]` return annotation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(backfill): days-loaded delta + test type hints + defang plan-doc CodeRabbit round-2 fixes for PR #73: - coordinator.async_run_backfill: ``_backfill_days_loaded`` now reports the delta (new days added by THIS run), not the total merged-history length. Capture ``prev_len`` before reassignment, compute ``new_days = max(0, len(result) - prev_len)``, clamp negatives to zero. Return value and log line updated to match. Docstring spells out the delta semantics so future callers don't misread the API. - tests/test_review_improvements.py::TestBackfillStatusSensor: add ``-> None`` return type hints to the four public test methods so mypy strict-mode is satisfied. Underscored helpers (_coord, _native_value, _attrs) left as-is. - .planning/PHASE-3.2-to-3.5-PLAN.md: section 1.5 was documenting ``error_message`` but the implementation uses ``error`` (matches sensor.py:571 + tests). Sync the plan to the code. Also defang the literal insecure-WebSocket scheme in section 8.4 (``ws-//`` with a parenthetical note) so the dashboard-protocol-safety recipe and generic CI security scans don't flag the plan doc itself. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(backfill): defang plan-doc + double-check status before lock CodeRabbit round-3: - Replace literal ws://localhost:* token in PHASE-3.2-to-3.5 plan doc with descriptive prose so secret/security scanners stop tripping. - Short-circuit async_run_backfill on _backfill_status BEFORE acquiring _ranking_lock so a concurrent caller does not block. Keep the re-check inside the lock to close the read/acquire race. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(backfill): repo-relative plan path + arithmetic + failure-path metadata reset CodeRabbit round-4 fixes on PR #73: - Plan doc: replace literal absolute `/Users/.../pricehawk/` path with `<REPO_ROOT>` placeholder so the planning markdown isn't tied to one user's machine layout. - Plan doc: fix arithmetic typo in commit total. 4+3+2+3 = 12, not 11. - Coordinator `async_run_backfill`: reset stale success metadata (`_backfill_days_loaded`, `_backfill_plans_replayed`) in the failure path so the status sensor doesn't surface misleading counts from a prior successful run after a failure. Set `_backfill_last_run_at` to the time of THIS (failed) run, matching success-path semantics. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(plan): MD040 fence langs + accurate kW description + named-provider lookup Round-5 CR doc nits on the planning artifact: - Add `text`/`bash` language tags to 3 fenced code blocks (dependency graph, dashboard ASCII layout, pre-push checks) for MD040 compliance. - Fix test description: kW→W is multiply-by-1000, not "doubles". - Plan snippet's named-provider check should read `coordinator._providers` (the instance dict) not `coordinator.data["providers"]`. Matches the shipped code in Phase 3.4's sensor.py registration. Plan doc only; no shipped code changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(cdr): Phase 3.3 commit 1/3 — pure-logic rollup module Add `cdr/rollup.py` with four pure functions over `daily_cost_history`: `filter_window` (rolling window selection), `sum_window` (per-key sum with sparse-row tolerance), `best_alternative_for_window` (lexicographic tie-break for determinism across ticks), and `savings` (current minus best, sign-preserving). No HA imports. Floats throughout. Returns `(None, 0)` rather than `(0.0, 0)` for missing-data states so the sensor displays `unknown` rather than misleading `$0.00`. 27 stdlib-only tests covering empty history, malformed dates, sparse alt presence, string-coerced numerics, explicit-zero days, ties, and prefix-based alt key scanning. Wires the source of truth that Phase 3.3 commit 2 will bind 15 rollup sensors to. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(sensor): Phase 3.3 commit 2/3 — PeriodRollupSensor + 15 registrations Add ``PeriodRollupSensor`` base + three subclasses (``CurrentCostRollupSensor``, ``BestAlternativeRollupSensor``, ``SavingsRollupSensor``) reading from ``daily_cost_history`` via ``cdr.rollup``. Three kinds × five windows = 15 new entities (``sensor.pricehawk_{current,best_alt,savings}_cost_{today,week,month,3month,year}``). Per plan §3.1 "Surprise risk — last_reset": only the ``today`` window sets ``last_reset`` to midnight. Rolling week/month/3month/year leave it unset — HA's TOTAL state-class tolerates this for monotonic-with- occasional-corrections series, and setting an artificial midnight reset would falsely re-attribute the prior day's value as today's spend. Per plan §8.6: no ``RollupStrategy`` interface; the three kinds are dispatched by inline ``if self._ROLLUP_KIND`` in the base class. Floats throughout (no Decimal). Lazy ``from .cdr.rollup`` import inside ``native_value`` (``# noqa: PLC0415`` annotated) to keep the import-time footprint identical to pre-3.3. 3 sensor smoke tests added to ``tests/test_review_improvements.py``, following the same property-body mirror pattern used by Phase 3.2's ``BackfillStatusSensor`` tests (CoordinatorEntity + SensorEntity metaclass conflict prevents direct construction under the conftest mock tree). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(strings): Phase 3.3 commit 3/3 — entity translations + CHANGELOG Add the ``entity.sensor.*`` block to both ``strings.json`` and ``translations/en.json`` for all 15 Phase 3.3 rollup sensors (3 kinds × 5 windows). Names + descriptions disambiguate the new ``savings_cost_today`` rollup from the existing real-time ``saving_today`` sensor (different math, both valid). CHANGELOG entry under [Unreleased] documents: - The new pure-logic module + 15 sensors. - ``last_reset`` semantics (today-only midnight reset). - The distinction from the legacy ``saving_today`` sensor. - The "sparse data → ``None`` rather than ``$0.00``" contract. No new tests in this commit — the rollup logic is covered by 27 tests in ``test_rollup.py`` (commit 1/3) and the sensor dispatch by 3 smoke tests in ``test_review_improvements.py`` (commit 2/3). Translations are static JSON. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(rollup): defensive type guards on filter_window + sensor provider deref Two CodeRabbit findings — both add defensive guards without changing the happy-path behaviour. 1. `cdr.rollup.filter_window` iterates `history` and calls `.get("date")` on each row unconditionally. The typed signature says `list[dict[str, Any]]`, but restored state from `.storage` or 3rd- party callers can slip in scalars (a corrupted entry, a future schema change). A non-dict row raises AttributeError and crashes every sensor read that hits this codepath. Skip non-dict rows up front; the rest of the parse/window logic is unchanged. 2. `PeriodRollupSensor.native_value` reads `self.coordinator._current_plan_provider.id` for the "current" and "savings" branches without guarding. The coordinator's __init__ raises ConfigEntryNotReady when `cdr_plan` is missing, so the attribute *should* always exist — but restart races, partial restore, or mocked coordinators in tests can briefly land here without it. Added an upfront guard that returns None (sensor shows `unknown`) instead of raising AttributeError. The "best_alt" branch is unchanged — it doesn't deref the provider. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(rollup): drop unsupported strings.json description + provider-guard tests CodeRabbit round-2 fixes for PR #74: - strings.json + translations/en.json: removed the ``description`` key from all 15 rollup ``entity.sensor`` entries. HA's entity schema accepts ``name`` only — ``description`` was silently ignored and triggered translation-schema warnings on startup. The descriptive copy lives in PLAN.md and CHANGELOG.md where it belongs. - tests/test_review_improvements.py::TestPeriodRollupSensorSmoke: the Phase 3.3 defensive guard added in sensor.py:657-660 (returns ``None`` when ``_current_plan_provider`` is missing or has no ``id``) had no test coverage. Mirror the guard in the ``_native_value`` helper and add two tests: - test_current_rollup_returns_none_when_provider_missing - test_savings_rollup_returns_none_when_provider_missing Both exercise the ``today`` and ``week`` windows so we cover the guard regardless of which window the user lands on first after a restart race or partial restore. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(rollup): empty-plan-id guard + negative-value sum test CodeRabbit round-3: - best_alternative_for_window: skip rows where the alt key is the bare prefix (``"alt_"`` with no plan id suffix) so a malformed history row cannot be ranked as the cheapest plan. - Add test_sum_window_handles_negative_values to cover FIT credits and refund rows summing alongside positive costs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Phase 3.4 — Named comparator drill-in (#75) * feat(coordinator): Phase 3.4 commit 1/2 — named comparator wiring Adds an OptionsFlow step that pins ONE CDR plan from the current ranked alternatives as a "named comparator", and wires the coordinator to construct a second CdrPlanProvider for that plan under the literal `"named"` key in `_providers`. The named provider then participates in the existing 30s tick loop (no new tick path) and contributes a `"named"` column to `daily_cost_history` at daily rollover, which the Phase 3.4 commit 2 rollup sensors will read. - `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN. We persist the FULL PlanDetailV2 body (not the summarised form) because the evaluator needs tariffPeriod data the summary omits. - `config_flow.py`: new `named_comparator` menu entry + step. The step reads ranked_alternatives + the per-day _ranking_plan_cache directly from the coordinator. Aborts with `no_ranked_alternatives` if either is empty (covers the post-install + post-midnight-cache-reset edge cases per plan §4.2 #1 + #3). Decision tree extracted to `plan_named_comparator_step()` so it's unit-testable without HA's app context (the OptionsFlow class itself becomes a MagicMock under the conftest mock tree). - `coordinator.py`: new `build_named_comparator_provider()` module-level helper (same testability rationale as build_backfill_plan_set). Called from both __init__ AND rebuild_engine so a fresh pin lands on the next OptionsFlowWithReload cycle without an HA restart. - `strings.json` + `translations/en.json`: new step title/description + menu label + two abort reasons. Keep distinct from the existing `comparators` step (which toggles live-API providers, not CDR pins). - 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree coverage including the full-body persistence guard, dedupe, default fallback when prior pin evicted from cache, plan_not_in_cache branch), 4 in test_coordinator_helpers (lifecycle of the named provider helper across all option shapes). Lock interaction with ranking lock: none. The named comparator joins the existing tick loop unchanged. The OptionsFlow step is a READ from _ranking_plan_cache, which is only written under the ranking lock — reading without the lock is safe because (a) the worst case is a brief torn read that resolves to the abort path, and (b) the lock is held for the duration of the ranking pipeline run, not just dict mutation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(sensor): Phase 3.4 commit 2/2 — NamedComparatorRollupSensor × 5 Surfaces the Phase 3.4 commit 1/2 wiring as user-visible HA sensors: - `sensor.pricehawk_named_cost_today` - `sensor.pricehawk_named_cost_week` - `sensor.pricehawk_named_cost_month` - `sensor.pricehawk_named_cost_3month` - `sensor.pricehawk_named_cost_year` Subclasses the Phase 3.3 `PeriodRollupSensor` base; overrides `native_value` + `extra_state_attributes` to read the `"named"` key from `daily_cost_history` rather than extend the base's `_ROLLUP_KIND` dispatch enum (which Phase 3.3 just shipped — localise the new behaviour here instead of rewriting the base contract for one extra kind). Registration is CONDITIONAL on `"named" in coordinator._providers`, so users who haven't pinned a plan don't see 5 permanently- unavailable entities clutter their HA UI. Reads `_providers` directly (not `data["providers"]`) so the registration check fires on first setup before the coordinator has populated its data dict. CHANGELOG.md gains a Phase 3.4 block under [Unreleased] documenting the OptionsFlow step, the new sensors, lock/ranking interaction (none), and the persistence-through-rank-churn behaviour (the pin survives the plan dropping out of cheap-rank top-K because it lives in options, not in derived ranking state). strings.json + translations/en.json gain entity blocks for the 5 new sensors. Descriptions call out the tick-by-tick cadence difference vs the other ranked alternatives (which only refresh at daily rollover) so users know what they bought by pinning. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(named): persist provider across restart + skip unique_id collision Two CodeRabbit findings on the Phase 3.4 named comparator. 1. The named provider added to `self._providers["named"]` was constructed in `__init__` and `_apply_options_update` but never persisted or restored. Every other CdrPlanProvider in the coordinator (current plan, amber, flow_power, localvolts) round- trips through `_async_persist_state` / `async_restore_state`, so on HA restart their per-day accumulators (import_cost_today, kwh) survive — but `named` would reset to zero while the others kept state, so today's rollup deltas would lie until midnight rollover. Added named.to_dict() to the persist block and a from_dict() call in restore, using `today = dt_util.now().date()` (already in scope above) to satisfy the AEGIS rule that from_dict MUST receive an explicit HA-timezone date (no `date.today()` fallback). 2. `GenericProviderCostSensor(provider_id="named")` produces unique_id `{entry}_named_cost_today`. `NamedComparatorRollupSensor` for the "today" window produces the same unique_id (since its _ROLLUP_KIND is "named" and the base class composes `{kind}_cost_{window}`). HA's entity registry drops the second registration silently, breaking whichever sensor the dashboard depends on first. Added a `provider_id == "named"` skip in the providers loop in `async_setup_entry` — the named comparator is exposed via its dedicated 5-window rollup family instead. Also skips the matching GenericProviderRateSensor pair so we don't register orphan rate sensors for the "named" key. (Only one such loop exists; the Phase 3.3 rollup loop and Phase 3.4 named-rollup loop don't have the same collision since their _ROLLUP_KIND differs.) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
* docs(planning): lock Phase 3.2 → 3.5 implementation plan
11-commit execution plan covering universal HA-history backfill (3.2),
period rollup sensors (3.3), named comparator drill-in (3.4), and
dashboard rewrite (3.5). Locks architectural decisions up front
(daily_cost_history as single source of truth, named comparator as
just another CdrPlanProvider, rollup as computed-on-demand) so the
executing model doesn't re-derive them at each commit.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(cdr): Phase 3.2 commit 1/4 — pure-logic HA history replay module
Adds ``cdr/history_replay.py`` with three pure functions that establish
the public API surface Phase 3.2 commit 2 wires the coordinator to.
No HA imports — the module is unit-testable outside the integration
runtime, matching the ``cdr/ranking_job.py`` pattern.
Public API:
- ``states_to_half_hour_slots`` — converts raw (ts, power_w, unit)
tuples to evaluator-shaped slot dicts aligned to 30-min boundaries.
Handles kW→W unit conversion, gap-protection clamping (6 min cap
matches ``cdr/streaming.py``), and string-vs-float power values
(HA's recorder serialises some sensor states as strings).
- ``replay_day_through_plan`` — wraps ``evaluate()`` with the
standard exception-swallow pattern (mirrors ``deep_rank``).
Returns None on evaluator exception OR zero slot count.
- ``fan_out_replay`` — generator yielding (date_str, {plan_key: aud})
per day. Streaming output keeps peak RAM at ~one day × N plans
instead of all-days × N plans of full CostBreakdown objects.
25 tests covering boundary alignment, unit conversion, gap protection,
sign handling, plan-failure isolation, date-order preservation, and
opt-in entry_options pass-through. Pattern follows
``tests/test_coordinator_ranking.py``: stdlib only, no pytest-asyncio.
Foundation for Phase 3.2 commit 2 (backfill.py rewrite) and Phase 3.3
(rollup sensors read the daily_cost_history rows this module helps
populate).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(backfill): Phase 3.2 commit 2/4 — rewrite backfill.py as HA-side adapter
Rewrites ``backfill.py`` from a 369-LOC Amber-API-coupled module into
a thin coordination layer over the pure-logic ``cdr.history_replay``
fan-out. New public API:
backfill_daily_cost_history(hass, grid_sensor_entity, plans,
*, days_back=30, entry_options=None, existing_history=None)
Internals pull recorder history day-by-day (NOT one big query — a
30-day single ``state_changes_during_period`` on a 1Hz grid sensor
returns 100K+ State objects), convert to evaluator slots, fan out
across N plans via the streaming generator, and merge per-date rows
into the coordinator's ``daily_cost_history``. Final list is capped
at 180 entries (matches live coordinator slice).
Day-by-day queries are deliberate and NOT parallelised. HA's
recorder uses a single executor pool so concurrent queries serialise
anyway and just bloat task count. Per-query memory is bounded; the
SQLite index on ``last_changed`` means 30 small queries are not
meaningfully slower than 1 big one. This is commented inline so CR
doesn't suggest parallelisation.
Legacy ``fetch_amber_price_history`` retained — still used by
``coordinator._replay_amber_today_from_api`` to seed the Amber
accumulator on a fresh install. The Phase 3.2 backfill itself no
longer fetches Amber prices: Amber's role narrowed to a *truth
overlay* written once daily by the live coordinator rollover.
Test rewrite: 14 legacy Amber-API tests deleted (those exercised
``backfill_from_history``, ``_build_amber_price_index``,
``_find_amber_rate``, ``_parse_history_states`` — all removed).
14 new tests cover ``_local_date_string`` (AEST-safe formatting),
``_states_to_tuples`` (State + dict shapes), ``_merge_into_history``
(insert/merge/cap), and ``backfill_daily_cost_history`` end-to-end
with the recorder mocked at the import boundary.
``tests/conftest.py`` extended with ``homeassistant.components``,
``homeassistant.components.recorder``, and
``homeassistant.components.recorder.history`` mocks so the lazy
recorder import inside the backfill resolves under the test harness.
``__init__.py``'s ``handle_backfill`` service updated to call the
new API with the current CDR plan; commit 4 will shrink it further
to a one-line delegate through the coordinator (once the wrapper +
status sensor land).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(coordinator): Phase 3.2 commit 3/4 — async_run_backfill + auto-kickoff
Wires the new universal HA-history backfill into the coordinator and
schedules its first run automatically after the initial ranking job
completes.
Coordinator additions:
- ``build_backfill_plan_set`` — module-level pure helper (mirrors
the ``cdr.ranking_job`` pattern) composing the {plan_key:
plan_body} dict from current plan + top-K ranked alternatives.
Lives outside the class so it's unit-testable without HA's app
context (the coordinator's ``DataUpdateCoordinator[T]`` base
gets mocked away by ``tests/conftest.py``).
- ``_build_backfill_plan_set`` — thin instance wrapper.
- ``async_run_backfill`` — coordinator-side run wrapper modelled on
``async_run_ranking_job``. Status-tracked via ``_backfill_status``
state machine (``idle | running | complete | failed``) plus
``_backfill_last_run_at``, ``_backfill_days_loaded``,
``_backfill_plans_replayed``, ``_backfill_error`` attributes
that the Phase 3.2 commit 4 status sensor will surface.
- Reuses ``_ranking_lock`` to serialise against the ranking job —
both mutate ``_daily_cost_history``. REVISIT: split if contention
observed in prod (cost of being wrong is brief serialisation of
two rare operations).
- Local import of ``backfill_daily_cost_history`` inside
``async_run_backfill`` (``# noqa: PLC0415``) so the HA recorder
isn't loaded at module-import time. Matches the existing pattern
at ``_replay_amber_today_from_api``.
``__init__.py`` kickoff:
- After ``async_run_ranking_job`` task is scheduled, schedule a
second task that AWAITS the ranking lock (so the first ranking
run finishes and the alternatives list is populated) THEN runs
``async_run_backfill(days_back=30)``. Without the wait, the
first backfill would replay history through only the current
plan, missing the ``alt_*`` columns.
Tests: 7 new in ``test_coordinator_helpers.py`` covering the pure
helper — current-plan composition, alt_* prefix keying, plan-cache
fallback to alt body, malformed-input skipping, empty/non-dict
``cdr_plan`` graceful return. Pattern matches the existing
``_extract_peak_rate_c_inc_gst`` test block in the same file.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(sensor): Phase 3.2 commit 4/4 — BackfillStatusSensor + service one-liner
Final Phase 3.2 commit. Adds the status sensor users will read from
their dashboards / automations and collapses the
``pricehawk.backfill_history`` service handler to a one-line delegate
through ``coordinator.async_run_backfill``.
New entity:
- ``sensor.pricehawk_backfill_status`` — state machine read-through.
State: ``idle | running | complete | failed``. Attributes:
``last_run`` (ISO timestamp), ``days_loaded``, ``plans_replayed``,
``error``. All sourced from the coordinator's status attributes
written by ``async_run_backfill``.
Service handler:
- ``handle_backfill`` shrunk from a 60-LOC inline pipeline to
defensive ``days`` coercion + a single
``await coordinator.async_run_backfill(days_back=...)`` call.
Status tracking, recorder pulls, plan composition, and
persistence all happen inside the coordinator method now;
failures surface on the sensor rather than getting lost to log
lines.
Service description: updated ``services.yaml`` to reflect the new
flow (replay-through-CDR-plan, no Amber API, status sensor pointer).
Unused ``CONF_GRID_POWER_SENSOR`` import removed from
``__init__.py`` (the coordinator now owns the lookup).
Tests: 4 new BackfillStatusSensor smoke tests in
``test_review_improvements.py`` exercising the property-read contract
(state defaults to idle, running propagates, datetime → ISO, error
attribute surfaces on failed runs). The sensor class itself can't be
imported under the conftest mock tree (CoordinatorEntity +
SensorEntity multiple inheritance from MagicMocks triggers a
metaclass conflict), so the tests mirror the EXACT property bodies
inline. Integration test on Ryan's HA will catch any drift.
CHANGELOG.md updated under ``[Unreleased]`` documenting the full
Phase 3.2 surface area: new module, rewritten backfill, status
sensor, service signature change, recorder mocks in conftest.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(backfill): type-annotate target dict union for pyright
`target = by_date.get(date_str)` infers as `dict[str, Any] | None`
but the subsequent `target[plan_key] = aud` assignment narrows
incorrectly without an explicit annotation. Annotate the union so
pyright accepts both branches (None → fresh dict, dict → mutate).
No behaviour change; pytest 760/760 still passes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(backfill): docstring accuracy on plan-set composition + recorder mock
Two docstring corrections flagged by Sourcery — no behaviour changes.
1. `build_backfill_plan_set` previously claimed it returned `{}` when
the current plan was missing. The implementation actually returns
alternatives even without a current plan (alts-only backfill is
intentional so rollup sensors can still surface comparative data).
Updated the docstring to describe the real contract: callers must
treat the absence of the current-plan entry as a "no-signal"
condition for the active plan at that time.
2. `_patch_recorder` claimed a 3-tuple return
`(get_instance_mock, history_call_mock, dt_util_now_mock)` but the
helper actually returns a 2-tuple `(get_instance, history_mock)`
(the dt_util mock was removed earlier when the backfill stopped
depending on it). Fixed the docstring and added a precise
`tuple[MagicMock, MagicMock]` return annotation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(backfill): days-loaded delta + test type hints + defang plan-doc
CodeRabbit round-2 fixes for PR #73:
- coordinator.async_run_backfill: ``_backfill_days_loaded`` now reports
the delta (new days added by THIS run), not the total merged-history
length. Capture ``prev_len`` before reassignment, compute
``new_days = max(0, len(result) - prev_len)``, clamp negatives to
zero. Return value and log line updated to match. Docstring spells
out the delta semantics so future callers don't misread the API.
- tests/test_review_improvements.py::TestBackfillStatusSensor: add
``-> None`` return type hints to the four public test methods so
mypy strict-mode is satisfied. Underscored helpers (_coord,
_native_value, _attrs) left as-is.
- .planning/PHASE-3.2-to-3.5-PLAN.md: section 1.5 was documenting
``error_message`` but the implementation uses ``error`` (matches
sensor.py:571 + tests). Sync the plan to the code. Also defang the
literal insecure-WebSocket scheme in section 8.4 (``ws-//`` with a
parenthetical note) so the dashboard-protocol-safety recipe and
generic CI security scans don't flag the plan doc itself.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(backfill): defang plan-doc + double-check status before lock
CodeRabbit round-3:
- Replace literal ws://localhost:* token in PHASE-3.2-to-3.5 plan doc
with descriptive prose so secret/security scanners stop tripping.
- Short-circuit async_run_backfill on _backfill_status BEFORE acquiring
_ranking_lock so a concurrent caller does not block. Keep the
re-check inside the lock to close the read/acquire race.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(backfill): repo-relative plan path + arithmetic + failure-path metadata reset
CodeRabbit round-4 fixes on PR #73:
- Plan doc: replace literal absolute `/Users/.../pricehawk/` path with
`<REPO_ROOT>` placeholder so the planning markdown isn't tied to one
user's machine layout.
- Plan doc: fix arithmetic typo in commit total. 4+3+2+3 = 12, not 11.
- Coordinator `async_run_backfill`: reset stale success metadata
(`_backfill_days_loaded`, `_backfill_plans_replayed`) in the failure
path so the status sensor doesn't surface misleading counts from a
prior successful run after a failure. Set `_backfill_last_run_at` to
the time of THIS (failed) run, matching success-path semantics.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(plan): MD040 fence langs + accurate kW description + named-provider lookup
Round-5 CR doc nits on the planning artifact:
- Add `text`/`bash` language tags to 3 fenced code blocks (dependency
graph, dashboard ASCII layout, pre-push checks) for MD040 compliance.
- Fix test description: kW→W is multiply-by-1000, not "doubles".
- Plan snippet's named-provider check should read `coordinator._providers`
(the instance dict) not `coordinator.data["providers"]`. Matches the
shipped code in Phase 3.4's sensor.py registration.
Plan doc only; no shipped code changes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(cdr): Phase 3.3 commit 1/3 — pure-logic rollup module
Add `cdr/rollup.py` with four pure functions over `daily_cost_history`:
`filter_window` (rolling window selection), `sum_window` (per-key sum
with sparse-row tolerance), `best_alternative_for_window` (lexicographic
tie-break for determinism across ticks), and `savings` (current minus
best, sign-preserving).
No HA imports. Floats throughout. Returns `(None, 0)` rather than
`(0.0, 0)` for missing-data states so the sensor displays `unknown`
rather than misleading `$0.00`. 27 stdlib-only tests covering empty
history, malformed dates, sparse alt presence, string-coerced numerics,
explicit-zero days, ties, and prefix-based alt key scanning.
Wires the source of truth that Phase 3.3 commit 2 will bind 15 rollup
sensors to.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(sensor): Phase 3.3 commit 2/3 — PeriodRollupSensor + 15 registrations
Add ``PeriodRollupSensor`` base + three subclasses
(``CurrentCostRollupSensor``, ``BestAlternativeRollupSensor``,
``SavingsRollupSensor``) reading from ``daily_cost_history`` via
``cdr.rollup``. Three kinds × five windows = 15 new entities
(``sensor.pricehawk_{current,best_alt,savings}_cost_{today,week,month,3month,year}``).
Per plan §3.1 "Surprise risk — last_reset": only the ``today`` window
sets ``last_reset`` to midnight. Rolling week/month/3month/year leave
it unset — HA's TOTAL state-class tolerates this for monotonic-with-
occasional-corrections series, and setting an artificial midnight
reset would falsely re-attribute the prior day's value as today's spend.
Per plan §8.6: no ``RollupStrategy`` interface; the three kinds are
dispatched by inline ``if self._ROLLUP_KIND`` in the base class.
Floats throughout (no Decimal). Lazy ``from .cdr.rollup`` import
inside ``native_value`` (``# noqa: PLC0415`` annotated) to keep the
import-time footprint identical to pre-3.3.
3 sensor smoke tests added to ``tests/test_review_improvements.py``,
following the same property-body mirror pattern used by Phase 3.2's
``BackfillStatusSensor`` tests (CoordinatorEntity + SensorEntity
metaclass conflict prevents direct construction under the conftest
mock tree).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(strings): Phase 3.3 commit 3/3 — entity translations + CHANGELOG
Add the ``entity.sensor.*`` block to both ``strings.json`` and
``translations/en.json`` for all 15 Phase 3.3 rollup sensors (3 kinds
× 5 windows). Names + descriptions disambiguate the new
``savings_cost_today`` rollup from the existing real-time
``saving_today`` sensor (different math, both valid).
CHANGELOG entry under [Unreleased] documents:
- The new pure-logic module + 15 sensors.
- ``last_reset`` semantics (today-only midnight reset).
- The distinction from the legacy ``saving_today`` sensor.
- The "sparse data → ``None`` rather than ``$0.00``" contract.
No new tests in this commit — the rollup logic is covered by 27 tests
in ``test_rollup.py`` (commit 1/3) and the sensor dispatch by 3 smoke
tests in ``test_review_improvements.py`` (commit 2/3). Translations
are static JSON.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(rollup): defensive type guards on filter_window + sensor provider deref
Two CodeRabbit findings — both add defensive guards without changing
the happy-path behaviour.
1. `cdr.rollup.filter_window` iterates `history` and calls `.get("date")`
on each row unconditionally. The typed signature says
`list[dict[str, Any]]`, but restored state from `.storage` or 3rd-
party callers can slip in scalars (a corrupted entry, a future
schema change). A non-dict row raises AttributeError and crashes
every sensor read that hits this codepath. Skip non-dict rows up
front; the rest of the parse/window logic is unchanged.
2. `PeriodRollupSensor.native_value` reads
`self.coordinator._current_plan_provider.id` for the "current" and
"savings" branches without guarding. The coordinator's __init__
raises ConfigEntryNotReady when `cdr_plan` is missing, so the
attribute *should* always exist — but restart races, partial
restore, or mocked coordinators in tests can briefly land here
without it. Added an upfront guard that returns None (sensor
shows `unknown`) instead of raising AttributeError. The "best_alt"
branch is unchanged — it doesn't deref the provider.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(rollup): drop unsupported strings.json description + provider-guard tests
CodeRabbit round-2 fixes for PR #74:
- strings.json + translations/en.json: removed the ``description`` key
from all 15 rollup ``entity.sensor`` entries. HA's entity schema
accepts ``name`` only — ``description`` was silently ignored and
triggered translation-schema warnings on startup. The descriptive
copy lives in PLAN.md and CHANGELOG.md where it belongs.
- tests/test_review_improvements.py::TestPeriodRollupSensorSmoke: the
Phase 3.3 defensive guard added in sensor.py:657-660 (returns
``None`` when ``_current_plan_provider`` is missing or has no
``id``) had no test coverage. Mirror the guard in the
``_native_value`` helper and add two tests:
- test_current_rollup_returns_none_when_provider_missing
- test_savings_rollup_returns_none_when_provider_missing
Both exercise the ``today`` and ``week`` windows so we cover the
guard regardless of which window the user lands on first after a
restart race or partial restore.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(rollup): empty-plan-id guard + negative-value sum test
CodeRabbit round-3:
- best_alternative_for_window: skip rows where the alt key is the
bare prefix (``"alt_"`` with no plan id suffix) so a malformed
history row cannot be ranked as the cheapest plan.
- Add test_sum_window_handles_negative_values to cover FIT credits
and refund rows summing alongside positive costs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(coordinator): Phase 3.4 commit 1/2 — named comparator wiring
Adds an OptionsFlow step that pins ONE CDR plan from the current ranked
alternatives as a "named comparator", and wires the coordinator to
construct a second CdrPlanProvider for that plan under the literal
`"named"` key in `_providers`. The named provider then participates in
the existing 30s tick loop (no new tick path) and contributes a
`"named"` column to `daily_cost_history` at daily rollover, which the
Phase 3.4 commit 2 rollup sensors will read.
- `const.py`: add CONF_NAMED_COMPARATOR_PLAN_ID + CONF_NAMED_COMPARATOR_PLAN.
We persist the FULL PlanDetailV2 body (not the summarised form) because
the evaluator needs tariffPeriod data the summary omits.
- `config_flow.py`: new `named_comparator` menu entry + step. The step
reads ranked_alternatives + the per-day _ranking_plan_cache directly
from the coordinator. Aborts with `no_ranked_alternatives` if either
is empty (covers the post-install + post-midnight-cache-reset edge
cases per plan §4.2 #1 + #3). Decision tree extracted to
`plan_named_comparator_step()` so it's unit-testable without HA's
app context (the OptionsFlow class itself becomes a MagicMock under
the conftest mock tree).
- `coordinator.py`: new `build_named_comparator_provider()` module-level
helper (same testability rationale as build_backfill_plan_set). Called
from both __init__ AND rebuild_engine so a fresh pin lands on the
next OptionsFlowWithReload cycle without an HA restart.
- `strings.json` + `translations/en.json`: new step title/description +
menu label + two abort reasons. Keep distinct from the existing
`comparators` step (which toggles live-API providers, not CDR pins).
- 22 new tests: 10 in test_config_flow_phase_3 (full decision-tree
coverage including the full-body persistence guard, dedupe, default
fallback when prior pin evicted from cache, plan_not_in_cache
branch), 4 in test_coordinator_helpers (lifecycle of the named
provider helper across all option shapes).
Lock interaction with ranking lock: none. The named comparator joins
the existing tick loop unchanged. The OptionsFlow step is a READ from
_ranking_plan_cache, which is only written under the ranking lock —
reading without the lock is safe because (a) the worst case is a brief
torn read that resolves to the abort path, and (b) the lock is held
for the duration of the ranking pipeline run, not just dict mutation.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(sensor): Phase 3.4 commit 2/2 — NamedComparatorRollupSensor × 5
Surfaces the Phase 3.4 commit 1/2 wiring as user-visible HA sensors:
- `sensor.pricehawk_named_cost_today`
- `sensor.pricehawk_named_cost_week`
- `sensor.pricehawk_named_cost_month`
- `sensor.pricehawk_named_cost_3month`
- `sensor.pricehawk_named_cost_year`
Subclasses the Phase 3.3 `PeriodRollupSensor` base; overrides
`native_value` + `extra_state_attributes` to read the `"named"` key
from `daily_cost_history` rather than extend the base's
`_ROLLUP_KIND` dispatch enum (which Phase 3.3 just shipped — localise
the new behaviour here instead of rewriting the base contract for
one extra kind).
Registration is CONDITIONAL on `"named" in coordinator._providers`,
so users who haven't pinned a plan don't see 5 permanently-
unavailable entities clutter their HA UI. Reads `_providers` directly
(not `data["providers"]`) so the registration check fires on first
setup before the coordinator has populated its data dict.
CHANGELOG.md gains a Phase 3.4 block under [Unreleased] documenting
the OptionsFlow step, the new sensors, lock/ranking interaction (none),
and the persistence-through-rank-churn behaviour (the pin survives the
plan dropping out of cheap-rank top-K because it lives in options,
not in derived ranking state).
strings.json + translations/en.json gain entity blocks for the 5 new
sensors. Descriptions call out the tick-by-tick cadence difference vs
the other ranked alternatives (which only refresh at daily rollover)
so users know what they bought by pinning.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(named): persist provider across restart + skip unique_id collision
Two CodeRabbit findings on the Phase 3.4 named comparator.
1. The named provider added to `self._providers["named"]` was
constructed in `__init__` and `_apply_options_update` but never
persisted or restored. Every other CdrPlanProvider in the
coordinator (current plan, amber, flow_power, localvolts) round-
trips through `_async_persist_state` / `async_restore_state`, so
on HA restart their per-day accumulators (import_cost_today, kwh)
survive — but `named` would reset to zero while the others kept
state, so today's rollup deltas would lie until midnight rollover.
Added named.to_dict() to the persist block and a from_dict() call
in restore, using `today = dt_util.now().date()` (already in scope
above) to satisfy the AEGIS rule that from_dict MUST receive an
explicit HA-timezone date (no `date.today()` fallback).
2. `GenericProviderCostSensor(provider_id="named")` produces
unique_id `{entry}_named_cost_today`. `NamedComparatorRollupSensor`
for the "today" window produces the same unique_id (since its
_ROLLUP_KIND is "named" and the base class composes
`{kind}_cost_{window}`). HA's entity registry drops the second
registration silently, breaking whichever sensor the dashboard
depends on first. Added a `provider_id == "named"` skip in the
providers loop in `async_setup_entry` — the named comparator is
exposed via its dedicated 5-window rollup family instead. Also
skips the matching GenericProviderRateSensor pair so we don't
register orphan rate sensors for the "named" key. (Only one such
loop exists; the Phase 3.3 rollup loop and Phase 3.4 named-rollup
loop don't have the same collision since their _ROLLUP_KIND
differs.)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dashboard): Phase 3.5 commit 1/3 — strip Amber chrome, scaffold multi-plan layout
Full rewrite of custom_components/pricehawk/www/dashboard.html (2447 -> 940
LOC) replacing the Amber-vs-current-plan two-comparator view with the
multi-plan ranked layout from plan section 5.1.
Visual language ported from assets/dashboard-v3-apple.html (dark default,
Outfit + IBM Plex Mono, noise + ambient bg). Per-provider colour tokens
(--amber-primary, --globird-primary) replaced with semantic ones
(--accent-positive, --accent-negative, --accent-neutral) per the Phase 3.0
pivot away from provider-specific branding.
Scaffold layout per plan section 5.1:
- NAV bar (brand + connection status + clock + theme toggle)
- HERO row: current cost card + savings-vs-best-alt card
- PERIOD TABS: [Today][Week][Month*][3 Month][Year], active swaps data
- RANKED ALTERNATIVES table: #/plan/peak/supply/saving, click -> drill
- DRILL-IN CARD: per-plan stats + "Pin as Named Comparator" button
- DATA HEALTH FOOTER: backfill state / days loaded / last ranking / count
Entity reads are NOT wired yet — sample data renders so the scaffold is
visually verifiable before commit 3.5/2 binds real sensor values.
WebSocket connection logic copied verbatim from the previous dashboard:
- WS URL derived from location.protocol (AEGIS rule: never hardcode ws://)
- Token from URL params, postMessage, parent.hassConnection, or
localStorage hassTokens (AEGIS rule: never hardcode the token)
CSP connect-src extended to include ws(s)://*.local:* so the dashboard
works on Ryan's HA Green at homeassistant.local (plan section 5.3
surprise #1). Existing localhost + Nabu Casa entries preserved.
Active period tab persists to localStorage so re-opens land on the user's
last view rather than defaulting to month every time.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dashboard): Phase 3.5 commit 2/3 — wire rollup + ranked + backfill entity reads
Hooks the scaffold from commit 3.5/1 up to the Phase 3.2 / 3.3 / 3.4
sensors. After WebSocket auth completes, fires a get_states + subscribes
to state_changed events for the 16 tracked entities.
Tracked entities (per plan section 5.2 + Phase 3.3 / 3.4 worker notes):
- 5 x sensor.pricehawk_current_cost_{today,week,month,3month,year}
- 5 x sensor.pricehawk_best_alt_cost_{...} (NOT _best_alternative_cost_)
- 5 x sensor.pricehawk_savings_{...}
- 5 x sensor.pricehawk_named_cost_{...} (NOT _named_comparator_cost_)
- sensor.pricehawk_ranked_alternatives
- sensor.pricehawk_backfill_status
Hero row binding:
- Current cost card reads sensor.pricehawk_current_cost_<window>.
- Savings card reads sensor.pricehawk_savings_<window> and colours green
/ red / muted around the +/- $0.005 deadband.
- Best-alt name pulled from ranked_alternatives.attributes.alternatives[0]
(sensor is sorted ascending by cheap-rank score per summarize_for_sensor).
- Projected annual extrapolates active-window savings * 365/window_days.
Period tabs swap activeWindow and re-call renderHero() — all rollup
bindings re-evaluate against the new window's entity ID. Active class
mirrors localStorage so the tab UI stays in sync on cold loads.
Ranked alts table render:
- Pulls ranked_alternatives.attributes.alternatives[].
- Renders rank-pill (#1 gold), plan name + brand, peak rate, supply,
saving. Saving column only fills for the #1 plan (the cheapest); #2..N
show "—" because we don't have per-alt cost rollups — only the best-alt
rollup. Avoids fabricating numbers that don't match the sensor.
- Click row → drill-in card slides up below + plan ID persists in
selectedPlanId so re-renders after state_changed events preserve
selection.
- Empty state ("Waiting for the daily ranking job…") covers first-install
before the first ranking run completes.
Drill-in card:
- Stats grid: peak rate, daily supply, customer type, plan ID,
cheap-rank score (when present).
- "Pin as Named Comparator" deep-links to the integration's Configure
page (/config/integrations/integration/pricehawk). Per plan section 5.3
surprise #2 + plan section 9 REVISIT 4: HA doesn't support per-step
deep-linking; the deep-link is the locked UX for this phase.
Data Health footer renders backfill state with state-coloured value
(green=complete / amber=running / red=failed / muted=idle), days_loaded,
ranked_alternatives.last_run as relative + absolute time, and the
alternatives count.
Empty-state UI for first-run users (plan section 5.3 surprise #3): when
backfill_status.days_loaded < 7, hero rollup values are replaced with an
"Accruing… [n/365]" pill instead of showing $0.00 — surfaces clearly
that we don't have enough history yet rather than implying zero spend.
XSS hardening: all attribute-sourced strings (plan_id, display_name,
brand, customer_type) pass through escapeHtml() before innerHTML
insertion. Catches any future CDR registry payloads that include
HTML-ish characters in brand names.
30s setInterval re-renders the ranked + footer cards so the relative
timestamps ("ran 27s ago / 3h ago") tick forward without waiting for
the next state_changed event.
TDZ fix: the period-tab boot block previously called setActiveWindow()
before the entity state store consts were declared, which tripped a
ReferenceError on attrs in strict mode. Boot now defers the first full
render to the explicit boot block at the script bottom.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dashboard): Phase 3.5 commit 3/3 — design-spec divergence + CHANGELOG entry
Wraps Phase 3.5 with the two non-code deliverables called out in plan
section 5.2 commit 3.5/3.
assets/DESIGN.claude.md:
- New "PriceHawk Dashboard (divergence from this spec)" section at the
end of the file. Explains WHY PriceHawk doesn't follow the Claude
marketing-site spec (different surface context, different information
density, different brand) and WHAT it does inherit (typographic
rationale, card-as-surface model, accent-discipline rule).
- Documents the PriceHawk token map (--bg-base, --accent-positive
etc) for cross-reference.
- Keeps the rest of the Claude marketing-site spec intact — no edits
outside the new appended section.
CHANGELOG.md:
- Phase 3.5 block at the top of [Unreleased] above the existing 3.4
entry. Documents the dashboard rewrite (entity bindings, period-tab
swap, ranked alts render, drill-in, footer, empty-state), the CSP
connect-src extension for *.local deployments, the deleted
per-provider colour tokens, the deleted Amber-specific cards, and
the manual-UAT-only test strategy (per plan section 6.3 table).
dashboard_config.py: NO behavioural change. Plan section 5.2 commit
3.5/3 calls for a "verify cache-busting still works" check; verified
that `?v=<version>.<epoch>` is appended in setup_panel_iframe and is
independent of dashboard.html contents — the rewrite doesn't affect
it. No source edit needed.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(docs): add css language to design-spec fenced code block
CodeRabbit / markdownlint MD040: the fenced code block listing the
PriceHawk CSS custom properties (--bg-base, --bg-surface,
--accent-positive et al) opened with a bare ``` instead of ```css.
Tag the fence as ``css`` so the markdown renderer applies CSS syntax
highlighting and so MD040 stops flagging the block. Content is
unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Artic0din
added a commit
that referenced
this pull request
May 18, 2026
Two entity-id composition bugs: - HA auto-inserts underscore between digit + letter, so '3 Month' window resolves to entity slug '3_month' not '3month'. Dashboard was composing 'sensor.pricehawk_current_cost_3month' which doesn't exist; all 3 Month tab cards stuck on accruing. - Phase 3.5 worker's deviation note claimed entity ids were 'best_alt_cost_*' and 'named_cost_*'; actual auto-derived ids use the full friendly-name slugs 'best_alternative_cost_*' and 'named_comparator_cost_*'. Dashboard subscribed to non-existent entities, all best-alt + named-comparator cards blank. Fix: WINDOW_SLUG map handles digit-letter underscore; ENTITY composition uses correct friendly-name slugs. Also: ranked-alts savings column now estimates per-alt savings for #2..N by scaling best-alt rollup cost by cheap-rank score ratio. #1 still gets exact rollup figure. Was previously '—' for #2..N. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
OptionsFlowWithReload+add_update_listenerconflict crashes options editing on HA 2026.3+ withValueError: Config entry update listeners should not be used with OptionsFlowWithReloadtariff_engine.pyContext
GloBird released new fact sheets on 01-04-2026 and 02-04-2026 with rate changes across all plans. The ZEROHERO plan had significant structural changes:
A user reported the options flow crash on Discord (Speegs, 4/4/2026) with HA logs showing the ValueError.
Test plan
🤖 Generated with Claude Code