Evidence gate and weights
Weights are provisional researcher assumptions (no AHP yet). Changing a weight changes scoring only; the evidence gate stays at the baseline weights.
Ranking
Explanation layer
Every explanation is audited: numbers must come from the engine output, unsupported claims are rejected, and any model output that fails falls back to the deterministic text. The language layer never calculates or changes a score.
Climate analysis (exposure inputs, not scored)
BIM integration and envelope code screen (ENS RETV)
Load a project JSON produced by python engine/bim_ifc.py model.ifc (IFC walls, layers, windows, true north), or use the built-in sample model. RETV is a whole-envelope check, not a wall-material certification.