An instrument for measuring the gap between two channels of ethical preference — the values someone states they hold, and the values revealed by their everyday choices over weeks. Pre-registered measurement validation in progress.
The real instrument is now runnable. Open the practice → — a working local-first prototype that runs the daily sessions on your own device, stores your choices only in your browser (nothing sent anywhere), and computes your profile locally once you've practiced across a few areas. Built on an event-sourced runtime you can read in the open repo.
It's still early — pre-cohort, and the reveal deepens the more you practice — but the full loop runs today: the three-layer card sort, daily sessions that alternate quick-fire choices with branching narratives, the cost-of-virtue probes, and an on-device reveal with per-domain visuals.
New here, or want the bite-sized pieces first? Start with The Two-Minute Self-Portrait — a quick pop-quiz that's honest about why you shouldn't trust it, and the gentlest way in.
Or explore the instrument's pieces one at a time — each a couple of minutes, nothing recorded. The first three are the inputs; the last is the payoff they build toward.
Keep five of twenty values; see how they fall across the four domains. This is what you'd say you value.
Predict how honestly you'll answer, then make six timed choices and see the gap. This is what your choices reveal.
Make one choice in the abstract, then again about someone you've come to know. Notice what moves — and why that's ambiguous.
What the instrument shows after weeks of use: the distance between stated and revealed values, across all four domains. The gap is the subject — no score, no verdict.
Existing ethics instruments measure one of two channels:
Neither systematically compares the two within-person over time. The gap is where growth becomes concrete — not "be more honest," but "your aspirational layer ranks honesty third; your everyday choices reveal it operates more like sixth."
The instrument is positioned as a contemplative practice with research-grade measurement underneath, not a personality quiz and not a self-improvement app. Three operating constraints are load-bearing:
Drawn from a synthesis of Moral Foundations Theory (Haidt), HEXACO Honesty-Humility, and behavioral-ethics literature. Each domain is probed across three scenario types: quick-fire repeated low-stakes (8-second timer), branching narratives with multi-decision arcs, and cost-of-virtue probes with stake ladders.
Three layers: current self, aspirational self, and admired other. Forced-choice format throughout (no Likert scales). Bradley-Terry scoring on pairwise comparisons. Same 20-value deck used across all three layers so within-person divergence is interpretable.
Try the card sort → — a 2-minute reduction of the stated channel: keep five of twenty values, then see how they fall across the four domains.
You don't need this part to use the instrument — the practice is built for your own reflection, and runs whether or not any of the below ever happens. But the honest question "does a daily-puzzle format really capture how someone lives?" deserves a real answer, so the design is laid out to be testable: a cohort study (pre-registerable at OSF, publishable including negative results) could check it against nine pre-registered hypotheses (three primary, six secondary), plus a wider research program of further measurement ideas (below). This is a possibility the open design supports, not a gate you're waiting on:
The first — and so far the only one with an interactive demo — of a family of novel methodological claims the project is developing (the others are in the research program below). H8 is pre-registered alongside the standard validation hypotheses:
Narrative-embedding with recurring-character attachment functions as a measure-debiasing mechanism against social-desirability response — AND as a stake-grounding mechanism for high-stakes attachment-laden choices. The instrument's narrative format is therefore a load-bearing measurement choice, not an aesthetic one.
Standard psychometric instruments treat narrative-embedding as either cosmetic ("makes items more interesting") or a confounding source of variance to be minimized. H8 inverts this: it claims narrative-embedding-with-attachment is a feature on a specific, named, falsifiable measurement-quality dimension.
If a participant has interacted with a recurring character across multiple sessions and developed measurable parasocial attachment to them, their response to a high-stakes choice involving that character's welfare is grounded in something the abstract version of the same dilemma cannot replicate. It's easy to say you'll save a handful of humans over a dog. When the dog is a character you've come to know across many sessions, the trade-off becomes grounded in real-feeling stakes.
Theoretical anchors: narrative transportation theory (Green & Brock 2000), parasocial attachment (Horton & Wohl 1956; Tukachinsky 2010 PSR-PRD scale). Tested via within-subject paired narrative-vs-abstract probes; sub-hypotheses H8a (debiasing) and H8b (attachment-grounding) both required for the combined claim.
See the idea on yourself. A two-minute interactive illustration: make one choice in the abstract, then make the same choice about someone you've come to know. Notice what moves — and read the honest caveat about why a shift could be either debiasing or manipulation.
The Weight of a Name — try the H8 demo →H8 is the first of a family of novel measurement ideas, each asking a different question about the distance between who you are and who you take yourself to be. H8 and H9 are pre-registered; the rest are design-stage — fully specified in the open repo, not yet locked into the validation. The honest label is on each.
Before you choose, you predict what you'll do; the instrument scores how well you forecast yourself. Two people with the same stated–revealed gap are different if one saw the slip coming and the other was blindsided — self-knowledge vs. self-deception. The spec →
How much your choices swing across settings — work, family, public, anonymous — treated as a stable trait in its own right. Steady or context-driven; neither is graded as better. The spec →
The gap between what you demand of others and what you demand of yourself — the self–other double standard, cleanly separated from ordinary weakness of will. The spec →
Someone who knows you makes blind predictions of your choices; the gap between their accuracy and your own measures how legible your character is to others vs. to yourself. The spec →
The radius of your moral circle, read from real choices — how concern falls off from kin to friends to strangers to out-groups to animals. A break point on the distance axis, the way the cost-of-virtue probe is one on the stakes axis. Wide or narrow; neither graded as better. The spec →
What you spontaneously moralize about in your own words, vs. what you actually do — the talk–walk gap. A third channel beside stated values and behavior (and what makes multi-method cross-checking possible). The spec →
After a choice, how much you feel the tug of the option you didn't take — the effort behind a virtue, or the quiet guilt behind a lapse. It separates virtue that's felt from virtue that's merely performed. The hardest signal to measure honestly (self-reported feeling is noisy), so the most exploratory. The spec →
Over weeks, when your actions and your stated values disagree, which one moves? Quietly adjusting the bar down to match how you act (rationalization) reads very differently from holding the bar and pulling your behavior up to meet it (growth). It measures the direction of the drift, not just the size of the gap — and needs a long enough run to see, so it's design-stage. The spec →
Some values have a price; some you refuse to trade at any amount. This reads the choices where you answered 'never' — the lines you won't cross for any sum — to surface which of your values are non-negotiable vs. tradeable. Holding many such lines can be integrity or rigidity; few can be flexibility or lack of conviction — it's descriptive, not a score. The spec →
How load-bearing morality is to your self-concept — and whether it's internalized (being good is core to who you are, and it drives your choices) or symbolic (being seen as good drives your self-presentation). It's the meta-trait that predicts how tightly your actions track your values at all. Self-report here is flattery-prone, so it's read against your actual consistency — design-stage. The spec →
When you act against your own standards, what lets you do it without guilt — 'everyone does it,' 'they had it coming,' 'it's not really that bad,' 'I had no choice'? It maps which of these you reach for. The honest hard part: a real justification looks identical to a self-serving excuse, so it's read for the tells of self-exculpation (self-serving, inconsistent, after-the-fact), not flagged the moment you give a reason. The spec →
Before you can weigh a moral choice, you have to notice there is one. This is the perceptual front-end — how readily you register the ethical dimension of an ordinary situation someone else would read as purely practical. It sits upstream of everything else here. More isn't better: high attention can be conscience or scrupulosity; low can be a relaxed non-moralizing eye or a real blind spot — descriptive, design-stage. The spec →
When you look at a moral situation, do you read it through the one who acts (the agent — responsibility, intent, blame) or the one acted upon (the patient — harm, suffering, need)? It's the structure you impose on the scene before any judgment — a justice-vs-care emphasis, neither better. The most exploratory of the set: for now the instrument can only glimpse it indirectly. The spec →
Some people hold a moral position as an objective fact — true for everyone, whether they agree or not; others hold the very same position as their own commitment, and wouldn't impose it on anyone else. This reads which way you lean — from how you talk about right and wrong, and whether you leave room for people who choose differently. Holding them as facts can be moral clarity or rigid intolerance; holding them as your own can be tolerant pluralism or a refusal to stand for anything — it's descriptive, not a score. The spec →
Together these are many views of the same choices — what you do, how steady it is across rooms, how far your concern reaches, what you expect of yourself and of others, what you predict about yourself, and what others predict about you. And a research-directions map sketches the channels still ahead — multi-method convergence, process and moral-emotion signals — each chosen so its blind spots differ from the daily-choice instrument's.
The operating constraints — the instrument never tells you a choice was wrong, has no leaderboards or scores, is never sold for screening or hiring, and keeps your data on your own device — are usually read as ethics. They are also validity conditions. The moment a measure of your values is used to judge or select you, people optimize the measure instead of revealing themselves (Goodhart's law), and the number stops meaning anything — the most polished profile then comes from the best performer, not the most honest person. So the same rules that keep the instrument trustworthy keep it valid: misuse-resistance and measurement validity turn out to be the same property. (A related caution the design builds in: how you act under real pressure can be a different thing from how you choose in a calm daily puzzle, so the instrument is careful about how far it extrapolates.) The validity audit →
And the sharpest test of all is one the project is committed to running even though it could sink the whole thing: a real-stakes channel — occasional decisions with actual consequences (real money, a real charity donation) — checked against what the hypothetical instrument predicted. If daily-puzzle choices turn out not to predict real behavior, that is a result the design is built to report, not bury. It's the keystone: the empirical answer to "does any of this measure real character?" The real-stakes design →
What's next: the core loop now runs end-to-end — the three-layer card sort (current / aspirational / admired self), quick-fire and branching-narrative daily sessions, cost-of-virtue probes, and an on-device reveal with per-domain visuals. From here it's continued refinement of the runtime plus specifying the research program above (H10–H12, the moral circle, the moral 360, and the real-stakes and moral-language channels are design-stage, not yet locked into the validation) — all of it in the open repo.