explaints v0.1.1

ExplainTS: self-explaining computed values

Motivation

Imagine the salary of your employees starts at 2000$ and every year it grows by 5% of this original amount. You have an API that allows your employees to check what their income will be after a certain number of years.

Your API could simply work like this:

> getExpectedIncome({years: 19})
  3900

But, if the rules to compute that value become complex, involving many different parameters, it could be better to return not only the value, but also the way it was computed:

> getExpectedIncome({years: 19})
  {
       value: 3900,
       reason: "expected salary",
       method: "sum",
       sources: [
           {
               value: 2000,
               reason: "base salary",
               method: "fixed",
           },
           {
               value: 1900,
               reason: "expected bonus",
               method: "mult",
               sources: [
                   {
                       value: 2000,
                       reason: "base salary",
                       method: "fixed",
                   },
                   {
                       value: 0.05,
                       reason: "yearly bonus (5%)",
                       method: "fixed",
                   },
                   {
                       value: 19,
                       reason: "years of work",
                       method: "input",
                   },
               ],
           },
       ],
   }

This brings several advantages:

• transparency: all values that affect the final result are clearly visible, allowing the client to understand how a result was obtain and to re-execute all the steps to guarantee correctness
• reactivity: parameters may be referenced in more computation trees; updating a parameter will automatically affect all values that are directly or indirectly computed from it
• efficiency: the tree of values can be created, once and for all, at startup; after that, it can be reused and patched as needed; changes to the parameters are applied to the computed values lazily, i.e. only when the values are requested again

Example

import * as ets from "explaints";

interface Parameters
{
    years_of_work: number;
}

const baseSalary = ets.makeComputed("base salary", 2000, "fixed");

const expectedSalary = ets.makeComputed("expected salary", "sum", 0, "sum", [
                           base,
                           ets.makeComputed("expected bonus", "mult", 1, "mult", [
                               base,
                               ets.makeComputed("yearly bonus (5%)", 0.05, "fixed"),
                               ets.makeComputed("years of work", (params: Parameters) => params.years_of_work, "input"),
                           ]),
                       ]);

function getExpectedIncome(params: Parameters)
{
    return expectedSalary.explain(params);
}