—
—
—
WHAT WE OBSERVED
UNKNOWN
WHO HAS POWER?
Ethical complication.
EXPERIMENTAL ATLAS / BEHAVIOUR × POWER × DESIGN
THE AGENCY
MACHINE.
An intention is not a lever that moves the world.
Enter a fictional situation. Manipulate its conditions. Watch five competing explanations disagree—and discover what each explanation leaves out.
◈ Conceptual indices, not probabilities, diagnoses, causal estimates or judgments of worth.
18FICTIONAL WORLDS
05TRANSPARENT MODELS
12INTERVENTIONS
“A useful hypothesis must survive the possibility of being wrong.”
01 / CASEBOOK
The situation comes first.
Every case is invented. Before turning a dial, ask what actually happened—and what remains unknowable.
Select a world
18 storiesSix possible turns
Isolated contextual shiftsSelect a moment to inspect it in the Model Bench. Each event perturbs the same baseline; the events are not a cumulative fictional biography.
CONTROL ROOM
What conditions would change the explanation?
15 explicit constructs in four groups. These are creative, defeasible assumptions, not questionnaire responses.
Select a group to open its dials. All changes propagate across the machine.
02 / MODEL BENCH
A disagreement is a clue.
Five different lenses, one set of inputs. A larger index does not mean a theory is “more true.”
At this moment
0–100 arbitrary display indexSeven parallel opportunities
Baseline + six contextual eventsEach episode is compared with the same chosen baseline. Lines connect hypothetical conditions, not measured human lives.
Open one moment
How the variables change| Opportunity | Identity–Action | Fogg | If–then | COM-B | Access |
|---|
Sensitivity / change one variable
Every other input stays fixed; inspect the model’s response as a construct varies from 0 to 1.
03 / INTERVENTION STUDIO
Who is allowed to move the dials?
Design combinations under a fictional budget. Some levers belong to managers, service providers or institutions—not the person facing the obstacle.
0 / 5 used
Token costs are fictional exercise values, not real prices or estimates of implementation difficulty. A numerically improved index is not proof of a better or more ethical intervention.
Intervention catalogue
What the machine says changes
| Model | Before | After | Difference |
|---|
Who carries the burden?
04 / CROSS-EXAMINATION
Make a claim. Risk being wrong.
Commit to a prediction before revealing the arithmetic. Be careful: correct prediction of an invented equation is not empirical confirmation.
PRE-REGISTER A MINI PREDICTION
No score is kept. Reconsidering your view is the point.
THE CROSS-EXAMINATION RECORD
The record is sealed.
Make your prediction and the five numerical explanations will testify.
What would evidence actually require?
05 / THE FIELD NOTEBOOK
Leave a record of your doubts.
Snapshots and written objections stay only in this browser’s local storage until you export or delete them.
Experiment record
The current situation, all 15 dials, five computed indices and the selected interventions will be captured together.
Saved experiments
0 recordsImports are validated and merged locally. Neither the page nor the builder receives the contents.
06 / THE METHOD IS PART OF THE ART
This machine should be challenged.
Its outputs are invented mathematics. The scientific work begins when definitions, measurement and rival predictions become precise enough to fail.
EPISTEMIC CONTRACT
What has been established—and what has not?
The project is an educational interactive thought experiment, not a behavioural assessment. Every scenario and event is fictional. Every numerical coefficient was chosen for illustration. No model has been fitted to observations. The 0–100 numbers are not probabilities, behavioural risks, clinical scores, forecasts, nor rankings of people or theories. Even agreement among all five lenses would not be empirical evidence.
This work is anchored in the creator’s public Identity–Action Gap inquiry, which describes identity–action mismatch as an object of investigation, and Agency Lab, which treats agency as situated influence rather than magical control. A proposed conceptual framing is not an established law. This page does not claim to reproduce unpublished IAG manuscript coefficients, definitions, or propositions.
“We can take our intentions seriously and remain willing to revise them.” — paraphrase of the public Identity–Action Gap position; not a measurement result
MODEL GALLERY
Five lenses, five blind spots.
σ(x) = 1/(1 + e−x). All input values are normalised to 0–1. F = 0.31 × time + 0.29 × cognitive + 0.24 × social + 0.16 × emotional friction. R = cost of adverse consequences. Input letters overlap across formulas only as specified in the equation glossary below.
Letter key for the formulas
I = identity salience; T = specific intention; M = momentary motivation; B = practical task ability; A = real room to choose; O = practical opportunity; S = social support; R = adverse risk; P = prompt; L = plan specificity; C = cue; F = friction composite. This notation is for the teaching adaptations here, not a claim about Sahil Tayal’s unpublished manuscript symbols.
A RESEARCH PROGRAMME, NOT A RESEARCH RESULT
How one could actually test explanatory value.
Specify one observable action
Define the person, action, context and observation window without assuming every intended action is desirable. Record a separately coded “changed mind,” “declined,” or “not feasible.”
Operationalise competing constructs
Define specific measures of intention, perceived control, opportunity and each proposed friction dimension in advance. Test reliability; do not invent convenient categories after seeing outcomes. Include perceived and externally verified conditions when appropriate.
Design repeated contextual observation
Use a consent-based within-person diary or ecological momentary assessment across opportunities. Avoid collecting unnecessary identities, employer details or sensitive narratives. Evaluate burden, missingness and selection bias.
Fit models against rivals on held-out opportunities
Compare calibrated baseline models (e.g., intention-only, Theory of Planned Behavior-derived predictors, cue/plan interactions, COM-B measures) with a preregistered IAG candidate. Account for temporal dependence and participant variation; assess predictive discrimination and calibration out of sample.
Test the intervention mechanism, not a success slogan
If ethically feasible, randomise a low-risk cue intervention or natural change in accessible resources. Measure actual exposure and opportunity. Do not randomise retaliation, denial of access or harmful constraints. Examine whether the intervention changed the intended mediator.
Precommit to falsification and heterogeneity
Ask whether identity adds explanatory value beyond intention and context, whether friction retains predictive meaning across settings, and whether results differ by institutional authority. Publish nulls, confounding concerns, exclusions, and negative controls.
THREE STRONG OBJECTIONS
01 / Is “friction” an explanation for everything?
If every negative outcome can be retrospectively labelled friction, the idea cannot be falsified. Prospectively measured types and rival model comparisons are essential.
02 / Are we smuggling in a preferred outcome?
Not acting may mean consent is respected, risks are recognised, the goal is revised, or rest is chosen. Labeling all non-execution as deficiency can turn ethical judgment into arithmetic.
03 / Who gets assigned the burden?
A manager can demand agency from an employee while withholding authority; an inaccessible venue can send reminders indefinitely. The right intervention may belong to the institution, not the individual.
RESEARCH LITERATURE & SOURCE TRANSPARENCY
Read the work behind the question.
Links below are primary materials or creator essays inspected in preparation for this build. Being linked means intellectual relevance, not that a paper endorses or validates this simulation.
The Identity–Action Gap ↗
Unvalidated proposal, multidimensional friction, counterarguments and careful distinction between intention and action.
Agency Lab ↗
Situated agency, institutional power, fairness and the costs of speaking up or refusing a request.
The Attention Lab ↗
Attention and context; resists equating meaningful life with completed tasks.
Human Capital for Humans ↗
Workplace dignity, capability development, institutional opportunity and freedom beyond output metrics.
BJ Fogg — Fogg Behavior Model ↗
Motivation, ability and prompt convergence; the source is conceptual, not this installation’s equation.
Gollwitzer (1999), Implementation Intentions ↗
American Psychologist, 54(7), 493–503. If–then cues, goal pursuit and implementation plans.
Michie, van Stralen & West (2011), Behaviour Change Wheel ↗
Implementation Science, 6, 42. COM-B’s capability, opportunity and motivation system; no numerical formula imported.
Ajzen (1991), Theory of Planned Behavior ↗
Organizational Behavior and Human Decision Processes, 50(2), 179–211. Intention, norms and perceived behavioural control.
CONCEPTS & DEFINITIONS
Words have consequences.
CREATOR’S NOTE
Why make a machine that distrusts itself?
Because the most tempting mistake in a beautiful interactive laboratory is to confuse its precision of drawing with precision of knowledge. A curve can be exquisitely smooth and entirely unsupported. This project deliberately puts the assumptions where visitors can question them, and the power relations where numerical models tend to hide them.
The simulation is playful. Human agency is not a video-game stat. A meaningful life does not owe the machine proof of relentless execution.
Built as an independent educational prototype. Reviewed against public sources on 23 October 2026. The unpublished Identity–Action Gap manuscript was not provided to this build and has not been verified.