How data was harvested, filtered, categorized, and quantified; precisely what it represents; and, with identical scientific vigilance, what conclusions one cannot validly extrapolate. I delineate every phase sequentially. An empirical finding is only as robust as the methodology that produced it, and omitting the methodology would be, frankly, an intolerable breach of scientific etiquette.
For each data channel, the sequential filtration pipeline from raw search returns to the validated corpus. The highlighted bar denotes items included in final analyses.
Each platform exposes divergent telemetry schemas. An unyielding rule governing our analysis: never aggregate physically disparate dimensions. Video views and follower counts are strictly non-additive scalars.
Video views officially reported by YouTube and TikTok on harvest date. These represent cumulative snapshots: older publications naturally accumulate larger exposure over time. Instagram restricts view telemetry on harvested image/carousel pages; there we calculate potential audience (aggregate follower counts of profiles posting on-topic). A follower is unequivocally not a view; hence, these metrics are never conflated or summed.
On YouTube and TikTok, these reflect platform-reported totals. On Instagram, likes displayed on the post (some accounts elect to suppress public like counts) and comments successfully collected via scraping, which may represent a subset of platform totals. On X (formerly Twitter), API/scraper returns yielded zero values across likes, replies, and reposts for all 427 harvested tweets—a manifest interface parsing anomaly, as the probability of 427 on-topic tweets yielding exactly zero interactions is statistically infinitesimal. Consequently, X is represented strictly in publication volume and collected reply threads.
A subset of data harvesting was executed at profile scope (retrieving all recent publications of identified creators), inevitably introducing off-topic material. Consequently, an Instagram, TikTok, or X publication is admitted into the corpus if and only if caption text, post copy, or Whisper-transcribed audio contains at least one of the following canonical tokens: superdotação, superdotado, altas habilidades, AH/SD, QI, IQ, Mensa, gifted, superinteligente, gênio, prodígio, neurodivergência, dupla excepcionalidade, or inteligência. YouTube videos previously underwent rigorous manual decontamination to eliminate homonym artifacts (e.g., Ghanaian pastor Mensa Otabil and American hip-hop artist Vic Mensa).
Every algorithmic filter entails classification trade-offs, operating along a precision-recall Pareto frontier. Ours is moderately permissive, as terms such as "inteligência" and "gênio" occasionally admit tangential philosophical discussions; simultaneously, it is moderately restrictive, excluding creators discussing twice-exceptionality without explicitly deploying defined keywords.
Our natural language system operates strictly as a discrete classification pipeline. It receives normalized text, evaluates semantic embeddings against an invariant rubric, and assigns a single category from a predefined closed schema while computing a confidence probability score. It generates no conversational prose, injects no editorial commentary, and invents no ad-hoc classes.
No automated classification has yet been double-blind benchmarked against human inter-rater reliability. A stratified sample of 275 items is currently structured for subsequent inter-coder validation. In the interim, percentage values reflect model classification distributions rather than certified ground truth. This distinction is scientifically critical, and we prefer absolute transparency over pretending to possess a precision metric not yet experimentally established.
11 curated token dictionaries (Mensa; IQ & Psychometrics; ADHD, Autism & Mental Health; Clinical Assessment; Schooling & Education; Legislation & Policy; Childhood & Family; Humor & Trivia; Genius & Prodigies; Elitism; Adult Identification). A publication intersects a domain if it contains at least one dictionary token, establishing an identical measurement scale across platforms, outlets, and creators.
Creators with ≥ 3 on-topic publications. For each profile, we compute the multi-dimensional vector representing the proportion of content intersecting each lexical domain. Applied K-Means clustering (k=6) with 25 random restarts. Cluster nomenclature reflects principal statistical centroids distinguishing each cohort from universe mean.
Creators with ≥ 3 on-topic publications and ≥ 50,000 reach (video views or followers). Quintile scores (1 to 5) across three objective criteria: reach magnitude; editorial affinity (proportion of substantive clinical/school/family/policy content minus 50% trivia weight); and conversation depth (comments per post).
Adjacent token pairs co-occurring across documents, deduplicated per document. Purged of syntactic stop words, numerals (quiz answers), tokens shorter than three characters, and platform interface artifacts. Displays the top 20 most frequent adjacent pairs per facet appearing across at least 3 distinct documents.
| Methodological Constraint | Analytical Consequence |
|---|---|
| Search Sample | Annual volumes reflect sample capture rather than absolute universe size. Commercial search APIs bias toward recent and viral items. |
| 2026 Harvest Date | Instagram and TikTok were scraped in October 2026, naturally skewing volume toward 2025 and 2026 items. |
| YouTube Comment Timestamps | Scraped as relative strings ("N years ago") and converted relative to harvest date. Reliable at annual resolution only, with a ±1 year margin. |
| X Telemetry Null Values | Harvested engagement metrics returned zero due to scraper parsing limitations. X is incorporated strictly for post counts and reply text. |
| Instagram Video Views Omitted | Instagram reach reflects potential audience (profile followers) rather than verified impressions. |
| Unsupervised Model Outputs | Thematic and sentiment classifications originate from NLP inference without double-blind human verification. |
| YouTube Comments per Video | Comments originate from 262 videos with active comment threads; highly viral videos may reflect partial API harvests. |
| Mensa Membership Counts | Figures originate from journalistic press releases rather than audited administrative records. |
Author identities for public YouTube comments were irreversibly pseudonymized via salted one-way SHA-256 cryptographic hashes and do not appear anywhere in this observatory. No personal comment is attributed to an identifiable author, and strings containing phone numbers, URLs, or personal tags were purged. Exclusively verified public profiles of creators, media outlets, and registered institutions appear by name. The key bridging pseudonymized tokens to original identifiers is strictly air-gapped on restricted local storage.