Volume integrator changes:
- Hard physical floor at 0 added to _updatePredictedVolume. Without
it, a basin seeded below dryRunSafetyVol (calibration / startup
/ low seed) under continued net-outflow drifted volume arbitrarily
negative; the level output looked clamped only because
_calcLevelFromVolume floors at 0, masking the underlying drift.
- New cumulative diagnostic: underflowVolume.predicted.atequipment
(m³) + getOutput().predictedUnderflowVolume. Non-zero indicates a
flow-balance error (over-reported outflow / missing inflow).
- The transition-only dryRunSafetyVol clamp is preserved so
startup-from-empty doesn't snap to 2.1 m³ on tick 1.
Spill flow refactor (taxonomic + bug fix):
- Synthetic spill moved from flow.predicted.out.<child='overflow'>
to its own position flow.predicted.overflow.<default>. The spill
is a derived quantity, not a physical sub-source sharing a position
with pumps — .child() was the wrong knob.
- Removes the spillPrev self-subtraction in the integrator (no longer
needed: outflowTotal at ['out','downstream'] cleanly excludes spill).
- Closes a latent fall-through bug exposed during this work:
.child('overflow').getCurrentValue() returned the value of any
available sibling child when overflow itself didn't yet exist.
Hardened separately in generalFunctions@a516c2b.
- _selectBestNetFlow folds the overflow position into the outflow
side so the predicted net-flow balance still reads ~0 while pinned.
Tests: 70/70 pass. 4 new subtests cover the 0-floor, accumulated
underflow tracking, getOutput surface, and refill-from-empty.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
d8490aa949