🚧 Implement continous simulation

README.md

@@ -394,13 +394,17 @@ requirement ID for each item in parentheses.
       (FR-9.2.5). *(Plain `serve` without a scenario already works —
       see Completed above.)*
 
-### Phase 3 — Long-duration (continuous) simulation (spec §8)
-The soak-testing feature requested 2026-10-03 — specified in §8,
-not yet implemented. Datasets are **optional**: signal logic
-(plugins, scheduled steps, degradation) runs in parallel with CSV
+### Phase 3 — Long-duration (continuous) simulation (spec §8) — IMPLEMENTED 2026-10-04
+The soak-testing feature specified in §8 is implemented: continuous
+mode with the fixed tick pipeline, optional streaming CSV datasets,
+degradation models, invariants/rate limits with bounded episode
+reporting, snapshots, and the gym-style `PlantEnv`. 30 new tests
+(`tests/test_continuous.py`); the shipped example
+`scenarios/pump_soak.example.yaml` demonstrates the full DSL.
+Datasets are **optional**: signal logic runs in parallel with CSV
 streaming, or entirely without it.
 
-- [ ] Plant environment `scenario/plantenv.py` (FR-8.2.14–16, requested
+- [x] Plant environment `scenario/plantenv.py` (FR-8.2.14–16, requested
       2026-10-04): gym-style `PlantEnv` — deterministic engine-owned
       tick clock (never wall-clock); per tick:
       `observe()` returns the WAGO input image in engineering units
@@ -411,44 +415,44 @@ streaming, or entirely without it.
       become thin adapters; external (non-Python) decision sources
       step the env over a socket; one decision source per output
       alias per tick (ownership conflicts are errors).
-- [ ] `scenario/datasets.py` — streaming CSV playback for designated
+- [x] `scenario/datasets.py` — streaming CSV playback for designated
       inputs (OPTIONAL driver): header-driven time column,
       `interpolation: linear|hold`, `repeat` with seamless seams,
       `time_factor` acceleration, `until` limits, bounded memory, ERROR
       verdict with file/line on malformed data (FR-8.2.1–8.2.11).
-- [ ] Parallel logic pipeline — fixed per-tick order: datasets → edge
+- [x] Parallel logic pipeline — fixed per-tick order: datasets → edge
       detection → plant plugins → degradation → invariants/rate
       limits → scheduled steps → snapshots; plugins may read
       dataset-driven and logic-driven aliases in the same tick
       (FR-8.2.12).
-- [ ] Logic-only continuous runs — no `datasets:` needed: demand
+- [x] Logic-only continuous runs — no `datasets:` needed: demand
       profiles generated inside plugins, top-level `until:` bounds the
       run; degradation/invariants/rate limits/snapshots apply
       identically (FR-8.2.13).
-- [ ] Continuous-mode run-end + validation rules: run ends when all
+- [x] Continuous-mode run-end + validation rules: run ends when all
       drivers exhausted; scenario with no datasets, no scheduled steps
       and no `until:` rejected (FR-8.2.1).
-- [ ] `scenario/degradation.py` — wear counters (runtime hours, start
+- [x] `scenario/degradation.py` — wear counters (runtime hours, start
       cycles, load-hours) and degradation models (`linear`,
       `exponential`, `staircase`, plugin shapes) computing plant
       parameters such as `pump_capacity` (FR-8.3.1–8.3.2).
-- [ ] Degradation hooks in plant plugins: `engine.parameter()` accessor
+- [x] Degradation hooks in plant plugins: `engine.parameter()` accessor
       used in loop physics, e.g. actual flow = nominal × capacity
       (FR-8.3.3).
-- [ ] `maintenance_at` thresholds: diagnosis injection via register
+- [x] `maintenance_at` thresholds: diagnosis injection via register
       `0x1050`, maintenance-alarm assertions, `maintenance_reset`
       action restoring capacity and clearing diagnoses (FR-8.3.4–8.3.5,
       FR-8.3.8).
-- [ ] `scenario/invariants.py` — per-tick `invariants` with bounded
+- [x] `scenario/invariants.py` — per-tick `invariants` with bounded
       episode aggregation and `rate_limits` over sliding one-hour
       windows (FR-8.4.1–8.4.2).
-- [ ] `snapshot_every` periodic dumps of aliases, wear counters, and
+- [x] `snapshot_every` periodic dumps of aliases, wear counters, and
       degradation parameters (FR-8.4.3).
-- [ ] Continuous-mode verdict rules and bounded reporting for unattended
+- [x] Continuous-mode verdict rules and bounded reporting for unattended
       multi-day runs (FR-8.4.4–8.4.5).
-- [ ] Scheduled sequence steps via `at: <duration>` interleaved with
+- [x] Scheduled sequence steps via `at: <duration>` interleaved with
       datasets on the sim-time axis (FR-8.4.6).
-- [ ] Acceptance tests per spec §8.8 (streaming 100k-row CSV,
+- [x] Acceptance tests per spec §8.8 (streaming 100k-row CSV,
       interpolation checks, repeat determinism, time_factor math,
       alias-ownership validation, degradation/recovery round-trip,
       episode recording, full 28-day soak, logic-only run, mixed