05 — Planning — Subsystem Test Map¶

The planning subsystem (RunPlanManager) is the authoritative rule-evaluation point for a job start: it builds the effective start plan (queue + payload + preflight), evaluates blocker/modifier rules and their fan-out, computes the confirmation token for a reduced run, and produces the runtime path-block report when a rule fires mid-job. Covered by 89 tests across 4 files.

Source: custom_components/eufy_vacuum/planning/ Architecture reference: 18 — The Access Graph

Coverage map¶

Source module	Stmts	Cov	Test files	Layer	Mocking
`run_plan.py`	604	93%	`test_run_plan_start_plan.py`, `test_run_plan_manager.py`, `test_run_plan_helpers.py` (unit), `test_water_vocabulary_boundary.py` (unit)	int + unit	bare x1

What's tested¶

Start-plan gates (SP) — _build_effective_start_plan across the state machine: ready blank graph, partial graph → incomplete_access_graph, rules without graph → access_graph_required_for_rules, single-dock without grants → access_graph_required, a matching direct blocker (→ blocked rooms + confirmation), a matching modifier (→ modified rooms), and access-dependency cascade (a blocked parent blocks its child).
Modifier fan-out (SP) — a rule's fan_out_room_ids apply to a derived target, plus the per-target guard branches (non-numeric / unknown / self / not-selected / blocked targets dropped; no-entity / no-match / empty-changes early-continues).
Runtime path-block report (SP) — get_runtime_path_block_report (the #11 regression guard): reachability propagation marks a room reachable via an accessible parent; remaining rooms are classified directly- vs indirectly-blocked; idle job → None.
Water-usage estimation + helpers (unit) — estimate_job_water_usage, _settings_profile_display, water-rate/level math.
Canonical/provider vocabulary boundary (WV, unit, added 2026-08-07) — test_water_vocabulary_boundary.py (15 tests). Core owns the canonical water KEY SPACE (off/low/medium/high) that the rate table is keyed on; it may not assume a provider's vocabulary equals it. Both shipped brands hide the difference — Roborock's declared values already ARE the canonical keys, and Eufy's "Low" reaches "low" through a generic case-folding rule — so neither exercises the declared-alias path, and it could regress unnoticed. These tests use a brand whose water words resemble nothing canonical (SynthMist, SynthFlood), so aliases are the only possible route, with a mutation control (WV-2) that withdraws them and requires the same values to stop resolving.

How it's tested¶

RunPlanManager(manager) over the real manager fixture; _seed(...) lays down managed rooms and merges per-room overrides, _blocker(...) / _modifier(...) build rule dicts, and binary-sensor states drive rule matches.

Known gaps¶

run_plan.py (93%) leaves two fan-out per-target guards (the int()-except on a non-numeric target id and the not-selected continue), the module-level helper except-paths (_safe_int / _safe_float TypeError/ValueError arms), the by_time wash-cycle branch and other water-usage partial branches, and the defensive early-returns / disabled-rule and non-dict-room skips inside get_runtime_path_block_report (no-queue, no-remaining, structural-issue, and unchanged-signature returns) — all defensive or low-value branch tails.