Task Shift
Atlas
Open the explorer↖ Sahilverse home

Method · version 1 · 26 Oct 2026

A model small enough to argue with

The atlas uses a deliberately simple, deterministic rule set. Simplicity is the feature: you can see exactly why a task lands where it does, and change your mind.

Three ratings per task (0–3)

Ratings were assigned editorially by the author for this atlas, informed by the sources on the Read Further page. They are not derived from any dataset and have not been validated.

Four rules, applied in order

  1. R ≥ 2 → protect
  2. C = 3 and S ≤ 1 → automate
  3. C ≥ 2 and S ≥ 2 → rethink
  4. otherwise → augment

Order matters: a task needing presence is protected even if parts are codifiable. That is a value judgment, stated openly.

Uncertainty

A rating of 1 or 2 sits near a threshold where reasonable people might disagree. Each such rating adds one step: 0–1 boundary ratings = low, 2 = medium, 3 = contested.

Worked examples

TaskC R SRuleBin
Record daily cash, UPI and khata entries3 0 12automate
Decide who gets goods on udhaar1 3 31protect
Screen CVs against a shortlist2 1 33rethink
Decide whether a result is causal1 0 34augment

The last example shows a weakness: a high-stakes, hard-to-codify judgment lands in “augment” by default. Many readers will want to call it “protect”. The tool records that disagreement rather than hiding it.

What the model cannot tell you

Privacy

All sorting, ticks and skills are saved only in your browser's local storage. Exports are generated on your device. The site makes no network requests with your data and uses no analytics.

← Back to the atlas