Use uncertainties in plots and fits

doc/architectural_decisions/005-variance-addition.md

@@ -0,0 +1,17 @@
+# Why do we add 0.5 to variance?
+
+## Status
+
+Current
+
+## Context
+
+If we pass counts data to `scipp` as variance this is correct- but if one of the counts is 0 then its variance is 0, leading to a division by 0 error when calculating its weight for fitting `weight = 1 / doc["data"][self.yerr]`.
+
+## Decision
+
+Our solution was to add 0.5 (`VARIANCE_ADDITION`) to each count to calculate variance. The actual data should be unchanged, the +0.5 is only for uncertainty calculation.
+
+## Justification
+
+The above approach is both "smooth" and converges towards sqrt(N) in the limit with high counts, and should also mean that we never get an uncertainty of zero in the fitting side.