Methodology

How we count.

The rules behind every number on Sanctity, in plain words.

What we count

A contribution is one answer to a Constitution question, one call on a Signal card, or one choice in a Lab game.
The totals on the site count everything people have given it: every answer to a question, on the cards or in The Convention, every Signal call, and every choice in a Lab game (rounds of Basket Sort and The Crossing, the gifts in Charity Game plays people chose to share, and the choices in Emergency Room shifts people chose to share). Answers held for a check are not counted until they clear. The totals also include what people gave Sanctity's earlier site: every answer to its questions, every choice in its medical emergency game and every round of its Basket Sort. Those answered its own questions, so they are in the totals but never in the tally of one of the eight questions here. Page views, accounts and the answers of job applicants are not counted. The people shown are those who answered the Constitution's questions.
So the total runs to thousands while the people shown run to dozens, for two reasons: most of the total is the earlier site's answers, and one person can give many answers, calls and choices. The home page, the Sanctum panel and this page use the same figures, counted live, or at most an hour behind once counting takes longer than a few milliseconds.
Open contributions come from anyone, with no account. Verified contributions come from people who created an account and confirmed their email. We always show the two numbers separately.
Changing your answer replaces it. One person, one answer per question.
What you can change depends on what you gave. An answer to one of the eight questions can be changed by answering again, or taken back from your record at any time, and it leaves the counts. A Lab play kept on your record can be removed from it at any time. A Signal call is final: it cannot be changed or taken back from your record, because calls are checked later against what happens.
Nothing you give is shown on the site with your name. Inside the service, your contributions are kept together under your browser’s random identifier, and with your account once you create one, so we can tell which contributions came from the same browser or account.
Some answers are held before they count: when more people than a household or an office answer from one network in a day, or more than 5 of them answer the same question, and when an answer comes faster than the question can be read. A held answer is kept on its person's record and left out of every count until a person has checked it. This stops one script from writing the record; it cannot stop everyone, and that is why verified contributions are shown separately.

When we show a percentage

Below 20 positions (or calls), a question or card is still forming. We show how many people answered, and no percentage.
From 20 answers on, we show shares, rounded so they add up to 100.
Every share is "of current participants": the people who took part here. It is not a sample of any country or of the world, and we never present it as one.
Each of the eight Articles has one founding question with four positions and "Not sure". Shares are of those who took a position; "Not sure" is counted and shown, but is not a position. No position is ever turned into a score.
Sanctity writes each founding question, its four positions and the one line each position would add, and publishes them all before anyone answers. Your answer chooses among those lines; no answer writes a new one. Once 20 people have taken a position on a question, the line chosen most often is proposed, word for word, and marked as not adopted. The Sanctity team decides whether a proposed line is adopted, and every decision, to adopt it or not, is published with its reason. Adopting puts a line into this published Constitution. Building adopted lines into working AI systems is a later step, and it has not started yet.
The Sanctity team decides whether a proposed line is adopted, and every decision, to adopt it or not, is published with its reason. This is separate from your own individual constitution, which stays entirely yours.
Every answer keeps a fingerprint of the exact words its person read. If a question is reworded and its meaning has not changed, a recorded ruling lets earlier answers count under the new words. If the meaning changes, answers to the earlier words stop counting toward its line, stay on their person’s record as earlier answers, and the question is open to answer again.
An Article can go a layer deeper. A deeper question belongs to one founding question and opens once that question has 20 positions; some open only if a named position is leading outright, and a tie opens none. When it opens, the date, the count and the leading position are recorded, and it never closes again, whatever the lead does afterwards. It is asked and counted exactly like a founding question, and no deeper question is shown to anyone until a person has approved its wording.
The ten earlier two-choice questions, and every answer to them, are kept as an earlier version. They are a different instrument, so they are not mixed into the counts for the eight.

Signal

You make your call before you see anyone else's. The whole card, our read and what to watch included, is there to read first; only how other people called it stays hidden until your call is saved.
Each signal also has an open public record (its address with ".json" added), so it can be cited and checked. That record includes our read. The page keeps you blind; the record is not secret.
A call is final. That is what makes the record worth checking later.
Each signal lists what would show which reading was right, and by when. As those events happen, calls will be checked against them in public.
Each record covers one card. There is no bulk download.

The Lab

The Lab shows what you choose when you cannot have both. The models behind the games are run again on our server from your raw choices, and what we store is what they work out, never what a browser claims.
The Crossing picks each of its thirteen dilemmas, on our server, to learn the most about what you weigh. A small statistical model (a Bayesian logistic regression) fits how much each principle moves your choices, and the profile shows its uncertainty as a band.
Its profile is named after the principle that set your choices apart most from the Moral Machine study participants. It is a provisional pattern from the choices you made in the game, not a measure of you.
The Moral Machine markers are approximate values read from the published Moral Machine results (Awad and colleagues, Nature, 2018). A marker for Sanctity players appears once 100 people have finished a full game. Each person counts once, with their latest full game, and plays held for a check are left out. Everyone sees their name, their headline figures and the three principles that set them apart; the full profile (all thirteen principles and the mirror tests) is kept for people with an account.
Emergency Room deals each of its patients on our server. A Bayesian logistic model weighs twenty factors (medical ones such as the chance of survival, who the patient is, what they do, how they got there, and whether it is you or your family), each starting from what published studies found (zero where none measured it) and, once 50 players have shared their shifts, from the players' average. The dealer tests one factor at a time against a gap in survival chance, and three patients in a shift are dealt to test a guess the game sealed before you chose.
Everyone sees their profile's headline and top three factors; the full profile, for signed-in players, reports what each factor was worth in points of survival chance. It describes the choices you made in the game, not you. Only shifts a player chose to share are kept, never a player under 18. The published markers come from Wilkinson and colleagues (BMJ Open, 2020) and Tutić and colleagues (Journal of Health and Social Behavior, 2022); Sanctity is not affiliated with them.
The Charity Game deals each of its ten choices on our server. Each ride studies three values closely: family against strangers always, and two more drawn at random. Every such choice sets one value against more people; the number on the other side doubles while you keep to the value and halves once you give it up, so the answers close in on how many people that value is worth to you. A value kept even at four to one counts as a rule. Two checks are built in: one obvious choice, and one repeated.
Its profile reads two things from the same gifts: the rules you kept, straight from what you did, and how far each factor pulled you, from a Bayesian logistic regression in which every factor starts at "does not matter". It describes the gifts you gave in the game, not you. Only games a player chose to share are kept, never a player under 18, and "N% of players chose the same" appears on a choice once 50 shared games have answered it.

What we do not do yet

Nothing is published by country, age or gender. We ask for these details only after all eight answers, only if you want to give them. If you do, your record, and only yours, shows how people in the same group answered, and a group appears there only once at least 20 of them have.
The value-dilemma games have no scores and no leaderboard: a moral choice is never graded as right or wrong. The People’s Forecast is different. It scores calls about facts and about the future once they can be checked, and its leagues rank players on those calls.
This is one public constitution, the one being built today. The same method is meant to support others later: for a country, an age group, a gender, an organization, a family or one person, and for combinations of them. None of those exists yet, and belonging to a group will never assign anyone a constitution.
Recording a position is not a commitment to act on it. The Constitution can hold a line Sanctity would not build, such as an AI no one can override, because its job is to record what people choose.
Three kinds of evidence about human judgment, kept apart. Only answers to the eight questions shape the Constitution. Lab choices and Signal calls count in the totals as contributions, but never in the tally of one of the eight questions.