Versioning

Problem Statement

• The model covers years.
• Users might use a single model for years.
• They may leave a model untouched for months before returning to it.
• How cab we evolve the model safely?

Options

• always update to latest model on load v ongoing support of old models?
• hide updates from users v make explicit?
• single version for model v versions for different elements?
  • aka model structure & data vs formula implementations
• store version inside v outside model?

Formula implementations

• special problems here
• changes may be subtle because of floating point arithmetic
• how accommodate third party implementations with potential forever support?

Mechanisms for Solution

• semantic versioning (obviously)
• store migration tools for each major version
• handle formulas that error - good practice anyway
  • use algorithm for cost calculations