[Tier 3] Park: re-evaluate on a signal, don't proactively pick up

[albinati]

Tracker for genuine future work that doesn't justify hands-on time right now. Each child issue has a specific re-evaluation trigger.

Issues + re-eval triggers

#	Story	Re-eval trigger
#195	V11-B quantile-based scenario perturbations	When ENERGY_STRATEGY_MODE flips to `savings_first` for peak-export (currently strict_savings — see memory `feedback_near_zero_grid_cost_policy`)
#197	V11-D occupancy inference	If you stop using the manual `PLAN_GUESTS_TODAY` flag and want auto-detect (#161 should land first)
#275	auto-set ENERGY_STRATEGY_MODE from OPTIMIZATION_PRESET	If you start switching presets often enough to want the coupling automated
#263	Split heat-pump deviation from base load in MPC tracker	If brief audit shows recurring `load_upside` re-plans that turn out to be HP-driven (then split the axis)
#336	PV calibration by solar elevation buckets	If observed structural sunrise/sunset PV bias survives the hour-of-day cal
#322	Conditional shutdown commit	Oct 2026 (deferred per memory `project_2026_05_11_climate_strip_incident` — below 10% saving bar)
#309	Dispatch starvation when Daikin quota tight	After 30 days of post-Phase-A heartbeat decouple data; tune reserve only if quota burns recur
#261	Quartz adapter → open.quartz.solar HTTPS	When Quartz publishes site-level HTTPS (current EB-HTTP workaround is stable)
#184	Daikin tank reheat anomaly detection	If a post-active-flip incident makes diagnostic value tangible

Why park, not close

Each is real future work, just not the highest-leverage thing on the board today. Closing them would lose the context that made them worth filing.

Definition of done

This epic auto-closes when all children either ship (move to Tier 1/2 first) or get explicitly re-deferred to a later date. Don't pick these up just because they're in scope; pick them when their re-eval trigger fires.