How we grade
Every library page displays one dated evidence badge. Whoever writes a page never stamps it. Every quarter the reviewer re-checks stamps; they can go down.
The six evidence badges
Several good studies or a formal result; the direction holds; boundaries mostly known.
Real only inside named conditions, stated on the page.
Famous, but the evidence fights itself.
Widely used by practitioners, little systematic evidence.
A popular story whose core claim is disputed, e.g. Rat Park.
Still quoted, doesn't survive; appears only in the graveyard.
Where it has been shown
Four steps, from the lab to India.
- 1Lab only
Shown in the lab only.
- 2Field
Outside the lab, in any domain.
- 3Field-tested in business
Two independent tests, or one pre-registered at scale.
- 4India-tested
At least one Registry entry in India.
Case grades (G1–G4)
Every case carries a last-checked date. Anything that came from memory (a TV show, a talk, an AI answer) stays G4 until it is traced to a primary source.
Tested
A clear comparison or pre-registered test, with numbers.
Documented
Several independent sources; count originators, not rewritten press releases.
Actor-told
A founder's or operator's own story.
Folklore
Unverifiable; may illustrate, never carries a page alone.
Atom types
| Type | Description |
|---|---|
| Bias or heuristic | A predictable mental shortcut. Examples: anchoring, processing fluency. |
| Social process | What happens between people. Examples: authority cues, information cascades, committed-minority tipping. |
| Formal model | A mathematical or normative rule. Examples: Bayes' rule, Condorcet's jury theorem. |
| Statistical artifact | An error in the data you look at, not in the mind. Examples: survivorship and selection bias, regression to the mean. |
| Memory and perception | Sensing, time and memory. Examples: peak-end, spacing, mere exposure. |
| Motivation and self | Goals, identity, habit. Examples: implementation intentions, goal-gradient. |
Default expectations
The replication policy. Until a page's evidence note shows otherwise:
Across 126 trials by two US government nudge units, the average effect was about 1.4 percentage points, against 8.7 in academic journals (DellaVigna & Linos, 2022). We use it as a rough reference for product decisions; that use is our extrapolation, not their finding.
Expect a famous social-psychology finding may not replicate. The 2015 Open Science Collaboration project is the reference point; we check its exact figures before quoting them.
Assume a Western, student-heavy sample until the evidence note says otherwise.
For cue-based effects (eyes, scarcity, social proof), assume decay until shown otherwise.
How to check a stamp
Click the badge, read the evidence note, open the sources, see who checked it and when, read the changelog.
“If you can't check it, it isn't a stamp.”
Whoever writes a page never stamps it. Every quarter the reviewer re-checks stamps; they can go down.