Scales and measuring

A scale is a degree, not a probability. It describes ordered situations and the model returns a position along them. It is kept distinct from predicates on purpose: “how much does this disrupt work” is a different question from “is this disruptive”.

Defining one

import { scale } from "jevlish";
import type { Ticket } from "./support.js";

// A degree, not a probability. Levels are ordered situations; the model
// returns a position along them. Two to ten levels.
export const disruption = scale<Ticket>("how much the reported problem disrupts the customer's work")
  .from("Work continues normally; the problem affects appearance only")
  .through("The task remains possible through a workaround")
  .to("The task cannot be completed")
  .named("disruption");

disruption.levels.length; // 3
disruption.top; // 2: scores run from 0 to top

.from(lowest) starts the levels, .through(level) adds intermediate ones (repeatable), and .to(highest) finishes. A scale needs at least two levels and supports at most ten. Each level is a description of a situation, not a number; the numbers are positions, 0 through scale.top.

A scale compiles to a Jev Score question with the levels as its criteria.

Measuring one subject

given(x).measure(scale) resolves to a Judgment<Measurement>.

import { given } from "jevlish";
import { ticket } from "./support.js";
import { disruption } from "./vocabulary.js";

const measurement = await given(ticket)
  .seenAs((t) => ({ subject: t.subject, body: t.body }))
  .measure(disruption);

if (measurement.status === "decided") {
  const { score, normalized, level, levelDescription, confidence } = measurement.value;
  score; // expected position, 0 .. disruption.top
  normalized; // score / top
  level; // the most likely level's index
  levelDescription; // that level's text
  confidence; // what the policy compared against policy.score.minConfidence
}

A decided measurement carries:

• score: the expected position on the scale, a real number from 0 to top.
• normalized: score / top, from 0 to 1.
• level and levelDescription: the most likely level and its text.
• confidence and probabilities: what the policy compared against policy.score.minConfidence, and the distribution behind it.

The judgment is uncertain when confidence misses the threshold.

Ranking with a scale

from(items).rankedBy(scale, order) measures each accepted item and sorts by score. See Filtering and ranking.

Inside ask

A scale can be one of the questions in given(x).ask({ ... }), alongside conditions and Choices; it produces a Judgment<Measurement> under its key. See Asking several questions.