A few small files and one formula, written down so any app can rank choices by a person's own values, and anyone can rerun the result to check it.
Version 0.1, a working draftReference engine v0.11Used in full by 1 app, in demos by 2
Builders: skip to Fig. 2 and try to break it. On purpose.
open-values-passportWhat one person cares about. Weights only, no name.open-values-lensThe options, the axes they are measured on, and each fact's source.engine.js · score()One formula. No accounts, no network, no hidden state. The same inputs always give the same order.score · coverage · why · capA score for each option, how much of it was known, and what moved it.Plate IFig. 1. Four example files from the spec.
The numbered marks point at the parts builders most often get wrong.
What matters to one person, carried from app to app.
{ "format": "open-values-passport",1 "version": "0.1", "values": { "planet": 5, "openness": 4,2 "autonomy": 5, "access": 3 } }
One thing an option is measured on, and how it is known.
{ "key": "environment",1 "label": "Green financing", "tier": "assessed"2 }
environment can share a values file without translating it.measured is an open-data fact, shown exactly. certified is a third-party label. assessed is a researched judgement, which must cite a source and is shown as a band, never as a precise-looking number.Scores on the declared axes, each with where it came from.
{ "code": "example-bank", "name": "Example Bank", "scores": { "environment": 98 },1 "provenance": { "environment": { "note": "What the source says, in one line", "source": "https://example.org/report",2 "asof": "2026" } } }
A number, plus how much of it was known and what moved it.
{ "score": 49, "coverage": 0.57,1 "why": ["Ethics", "Transparent"], "facts": 4, "wanted": 7, "cap": { "label": "Green financing", "v": 12 }2 }
Plate IIFig. 2. The scoring formula, running the reference engine.
This runs the reference engine, the same engine.js file the app uses. One option, three axes. Change a weight, change a fact, or mark a fact as unknown, and see which of the three rules below applies.
Fair pay is unknown, so 3 of the 8 weight points had no fact, and the score is pulled a little toward 50.
What score() returned
{"score":64,"coverage":0.625,"why":["Climate","Price"],"facts":2,"wanted":3,"cap":null}
An unknown fact is left out. The score is pulled toward a neutral 50 by how much is unknown.
Below a quarter of what you care about, the engine returns nothing rather than a guess.
An axis you weight 4 or 5 that scores 20 or less caps the score at 49. A certification is the exception, because a 0 there means no label was found, which is missing paperwork, not harm.
What's happening. One option, three axes, and the reference engine itself, the same file the app runs. Each Try it button sets the inputs so that one rule applies.
Why it matters. An average is the obvious way to combine what someone cares about, and it misleads in two predictable ways: it treats a missing fact as a zero, and it lets a disaster on one axis hide behind good marks on the others. The first two rules answer the first; the cap answers the second.
Devil's advocate. The cap is crude on purpose. It is a flat 49, whatever the rest says. It used to fire on certifications too, where a 0 only means no label was found. On the open-data food comparisons that was 10,702 of 14,945 ethics scores in September 2026, so version 0.11 stopped treating missing paperwork as a verdict.
Rabbit hole. The engine is one plain file. Its score() function is 27 lines, six of them comments explaining the rules. Read it on GitHub.
An app can look and sound however it likes. These six stay the same, and five of them are checked by a validator anyone can run.
| Rule | What it means | How it is checked |
|---|---|---|
| Fit to someone's values | A score says how well an option fits someone's stated values. It never says an option is good or bad for everyone. | Same facts with other weights must give another order |
| Every judgement cites | An assessed claim carries a note, a source and the date it was true. | Validator: provenance shape |
| Missing is not zero | Unknown facts are left out, pulled toward neutral, and withheld when too little is known. | Validator: score keys and 0 to 100 range |
| No pay-to-rank | Nothing in the files can encode sponsorship or paid placement. | Validator: no paid-placement fields |
| Values travel | A person's values file carries no identity and works in any compatible app. | Validator: values file shape |
| Any community's values | An app can bring its own value names and map them to the shared ten. The format favours none. | Validator: value-name mappings point at real axes |
To check a lens you built: clone the source and run npm run audit:conformance, or add --passport=yourfile.json to check a values file.
| App | How far |
|---|---|
| Conscious Consuming | In full. The reference implementation, ranking 140+ everyday decisions. |
| Kosplora | Demo. The second app. It needed its own value names, which is why version 0.1 exists. |
| Where to Message | Demo. Messengers ranked on privacy, openness and reach. |
67 formats are registered: 24 stable, 31 drafts and 12 internal contracts used while building.
stabledraftinternalThe registry, as JSON →
One maintainer today: Bentley Moon, at the Futurism Institute. To propose a change, open an issue on GitHub or write to futurisminstitute@gmail.com.
If it ever publishes descriptions of organisations, it will use Murmurations, an existing open protocol, rather than invent a format.