How Nudgeswipe rates teardowns and evidence

How a teardown is captured

• 121 teardowns are screenshots. The caption is the host, the capture date, and the words “logged out”. The Economist prices a year by the week, and preselects it is captioned as a logged-out capture on 16 Oct 2026.
• The About page states that screenshots are our own dated captures from a headless browser, logged out, cropped to what the teardown discusses, with no personal data.
• When a teardown quotes a price, it adds a line naming the currency and saying the prices are exactly as shown on the page or email we captured.
• 26 teardowns are emails. They are not re-hosted. Each one links to Email Love, and the caption says the subject lines are as listed there. The About page says emails quote subject lines and visible copy.
• The steps are the notes stored for that capture.

How a verdict is assigned

Each teardown stores one verdict. The label is not calculated from a score. The ethics sentence on the teardown says why that label was used.

• Honest nudge (87). In the rules shipped with every teardown prompt: true, visible, easy to reverse, and in the user’s interest.
• Conditional (41). The findings page describes this set as honest only if something on the page is true, usually a deadline or a “was” price. The condition is the ethics sentence.
• Grey area (19). A stored label of its own, beside honest nudge, conditional and dark pattern. The reason is the ethics note. Each one is listed on grey-area teardowns.
• Dark pattern (0). In the same rules: hides costs, fakes scarcity or popularity, shames, or makes leaving harder than joining.

How evidence is rated

The same four lines are on the bias directory. Today the library stores 56 strong, 16 mixed and 24 limited. 31 ratings are also marked provisional.

• Strong: several independent replications, or a meta-analysis or multi-lab replication that supports it; no credible large failed replication; an empirical source is cited, not only a theory paper.
• Mixed: credible evidence both ways, such as a large or pre-registered failed replication, a meta-analysis near zero or heavily affected by publication bias, or a live dispute that it’s an artefact. The critique is always cited. Those 16 biases are listed on mixed evidence.
• Limited: one paper, or a few from one lab; no independent replication found by us; or only lab or hypothetical evidence for the marketing use described.
• Provisional: any rating without a cited replication or review. A rating based only on a theory paper can’t be Strong. The bias page then says the rating cites only the original paper so far.

How evidence is cited

• Each bias stores a citation and a DOI. The bias page prints the citation and links the DOI. A second DOI is linked when one is stored.
• A replication or critique is stored as a sentence plus a URL. The bias page lists those under “Replications and critiques”.
• When a critique sentence is stored, it is shown next to the rating.
• A teardown’s source list starts with the capture, or the Email Love links, then the bias citation and its DOI, then any public statements, extra sources or deceptive.design links stored on that teardown.
• The prompt on each teardown and bias page says not to cite a retracted or failed-to-replicate study as support. The example named there is Ariely & Wertenbroch 2002. It also says not to state a study figure unless it is on the page or in a source you can link.

Limits of nudge effect sizes

The bias directory already notes that nudge effects shrink after correcting for publication bias (Mertens et al. 2022; Maier et al. 2022). This page does not add figures from those papers. Effect sizes belong on the bias page that cites them.