Methodology

The Good Tech Index is an opinionated, transparent index — not a certification authority. Everything here is open so the scores can be audited and challenged. This page is generated from the live data, so it always matches what the scores actually use.

1. A product is modelled as layers

Each layer means something different by "good", so scores live per layer and roll up. Layers carry a default weight in the balanced profile.

Layer	Default weight	What it covers
Hardware	1.0
Energy	1.5
Hosting & network	2.0
Services	1.5
Software supply chain	1.0
Logistics & fulfilment	1.0

2. Every ingredient is scored on fixed axes

Each ingredient gets a 1–5 rating on each axis. The score is the vector; the single number is a weighted roll-up. Axes that "follow provenance" walk the supply chain (e.g. a self-hosted model → the lab that trained it) and take the worse of the two.

Not every axis applies to every layer — an energy source is judged on its footprint, ownership and concentration, not its data-sovereignty or software licence. Each layer counts only its relevant axes, so an ingredient isn't propped up (or dragged down) by axes that don't apply.

Data sovereignty & surveillance 1–5 higher is better

Where data lives and the surveillance capacity it grants. 5 = fully sovereign.

Environmental footprint 1–5 higher is better follows provenance

Energy source, hardware lifecycle, water/cooling. Follows provenance.

Local economic benefit 1–5 higher is better

Does the spend stay in-region (5) or leave (1)?

Market concentration 1–5 higher is better

Does using this reinforce a chokepoint or monopoly? 5 = no lock-in / many alternatives.

Openness & lock-in 1–5 higher is better

Can you leave, and can others reach it? 5 = open access with no account required and easy data export; 1 = a walled garden — account-gated, proprietary, hard to export.

Ownership model 1–5 higher is better

Who benefits from ownership? Shareholder / private-equity extraction (1) vs worker-owned / nonprofit / state or public-good (5).

Software freedom (FOSS) 1–5 higher is better

Is the software free & open-source, and under what licence? 5 = fully FOSS under an OSI-approved licence; 1 = proprietary / closed source.

Upstream labour 1–5 higher is better follows provenance

Chip fabrication, warehouse, data-labelling conditions. Follows provenance.

3. How the number is computed

Normalise each axis score to 0–1 across its scale, flipping it when lower is better.
Ingredient score — a weighted average of its normalised axis scores under the chosen profile.
Layer score — each ingredient's score weighted by the share it holds of that layer.
The product itself — a product can also be scored directly on the same axes (is the app open-source? who owns it?), folded in as its own weighted component.
App total — layer scores plus the product's own score, weighted and rendered 0–100.

Ingredients are recursive: one that is itself built on others (an ISP on its energy and hosting) has its score rolled up the same way — its own score blended with the roll-up of what it depends on. A family parent with no score of its own (e.g. a provider across many regions) defaults to the average of its children.

A letter grade is a convenience over that number: A ≥ 80, B ≥ 65, C ≥ 50, D ≥ 35, else E.

4. Evidence and confidence

Every score points at evidence with a source, a date, and a quality tier. Confidence is derived from that evidence and shown alongside the score, so a high score built on weak evidence reads as low-confidence rather than authoritative. On any label, click an axis to expand the reasoning — each ingredient's (and the product's own) score, confidence, rationale and source links.

Evidence quality tiers

Verified — a checkable fact (grid data, filings)
Credible — a reputable third-party report
Self-reported — the ingredient's own claim
Inferred — reasoned from indirect signals

Confidence

High — backed by verified evidence
Medium — credible but not independently verified
Low — self-reported or inferred

Confidence propagates: the app-level figure carries the weighted-down aggregate of the evidence beneath it.

5. Flags are not averaged in

Some involvements deserve a visible marker rather than a small drag on an average. Flags come in three severities — info, concern, and hard fail — and a hard fail is surfaced prominently regardless of the numeric score.

6. Values differ — so the index is a family of scores

A climate-first reader and a labour-first reader weight the same data differently. These system profiles ship by default, and signed-in users can build their own:

Balanced
Climate first
Labour first