How this notebook thinks
Understanding attention should make your decisions more informed, not replace one persuasive certainty with another.
01 / Label the claim, not the whole idea
Labels describe where a claim comes from. They are not a confidence score, a quality ranking or a guarantee that a finding transfers to your audience.
Official documentation can describe a surface, signal family or system stage. It is not an independent audit, a complete model or a current list of weights. Historical disclosures stay marked as historical; restricted access stays visible.
Experiments can support comparisons within their design and context. In headline datasets, readers may be randomized to whole headlines without independently randomizing each word attribute. An analysis of negativity or concreteness is not automatically an isolated causal effect of that feature.
Null and replication findings matter. The arousal entry does not turn an unreliable sharing effect into a universal virality rule.
Observed associations can motivate questions but do not establish that changing one feature causes an outcome. The moral-emotional diffusion research concerns historical political Twitter networks; network, author and topic differences can travel with the words.
An editorial suggestion or creative template is a proposition to examine, not a published result. Sources can motivate it without validating it. This label is especially important when moving from news headlines or a music-market experiment to an email or an offer page.
02 / Keep context attached
03 / Understand → Apply → Test
The Lab uses deterministic, research-informed templates in your browser. It does not call a model, measure your audience, fetch platform data or publish content. Your topic, audience description, objective, surface, mechanism and supplied evidence shape an editable brief.
Define one changed variable, hold other features constant, choose a primary metric with an eligible denominator, and record quality or trust guardrails, an observation window, stopping rules and confounders before looking at outcomes.
Manually entered nonnegative integer counts are checked so outcomes cannot exceed eligible observations. The calculator reports rates and their percentage-point difference. Relative lift requires a positive baseline; zero denominators remain unavailable.
Those calculations do not assign a winner, establish statistical significance or prove causation. Separate social posts and before/after SEO comparisons are descriptive unless an appropriate randomized design actually exists. Even randomization needs sound execution, stable definitions and a suitable analysis.
04 / The decoder boundary
The free checker is a separate, existing detection system. Its rule matches and optional configured model can identify language patterns worth inspecting, not establish the sender’s intent, diagnose a reader’s emotion or prove that an offer is false.
Its older category reference labels are not this new research registry. The Atlas does not silently redefine detector categories, thresholds or the ruleset version. Selected scarcity and social-proof categories link to related research using stable IDs only; the checker does not claim to detect all six Atlas mechanisms.
Scam and phishing warnings retain their own priority. They are not repackaged as strategies to try. Checker text is not carried into the Lab through URLs or saved drafts.
05 / Demand needs context
The Trends notebook is manual. It does not fetch a live feed, crawl pages or infer keyword volume. Record the source, topic, geography, date range, search type, comparison context and what you actually observed.
Google Trends’ 0–100 values describe normalized, sampled relative interest, not absolute search volume. Independently normalized exports are not automatically comparable; a zero can reflect insufficient data. A spike does not identify its cause.
Keep a contextualized note06 / Useful without coercion
No invented testimonials, resetting timers, fake stock counts, guaranteed rankings or universal virality scores. Missing evidence stays visibly unfilled. A public claim still needs your verification before publication.
Moral-emotional research is included for media literacy, not targeted political persuasion or outrage engineering. The proposed application is a general, nonpolitical clarity exercise; no sensitive-identity inference belongs in the brief.
The free, no-login checker remains available. An optional audit service is separate from these research tools.