Underplaced is designed to find talented, intelligent Americans whose ability is stronger than their résumé, credentials, or current job would suggest. AI broadens the search; people make the consequential decisions.
AI searches the public web for signs of unusual ability. The search covers blogs, Substacks, social posts and replies, YouTube videos and transcripts, podcasts, SoundCloud and Bandcamp releases, visual work, software, technical forums, independent projects, and game communities.
Games are unusually rich evidence. Esports, Minecraft, Space Station 13, Garry’s Mod, and their modding communities can reveal systems thinking, spatial reasoning, leadership, persistence, and technical craft that formal credentials miss.
The first instrument is eight sources we can actually score, in this order: GitHub taste, Paradox forum prose, health-department PIC and ABC lists, romhacking/pret, SS13/GMod makers, dark-pool prose, FAA airmen plus FCC Extra, then flyover USPTO inventors. See the ranked queue on the map →
Popularity decides where to look first. It is never g or A. Everything else on the source menu is a prior until Stage 2 assigns loadings.
The first pass favors small and midsized creators. A real but modest audience means YouTube, Twitter, or another platform has already performed a cheap, noisy first round of discovery. People with 200,000 followers usually have enough connections and public proof that finding them adds little.
That is a search priority, not a cutoff. Underplaced must still look for the 60-follower reply account, tiny channel, obscure forum writer, or brilliant builder who is bad at feeding an algorithm. Popularity can save compute; it cannot define intelligence.
People without a useful public trail can apply directly or be nominated. They can submit work samples and, if the evidence warrants it, receive the same invitation to take the assessment.
A signal creates a lead for human review. It does not reject, rank for employment, or hire anyone.
A person checks whether the work is authentic, substantial, and actually belongs to the candidate. The reviewer also asks whether the candidate’s demonstrated ability appears mismatched with their current opportunities. This stage produces an invitation, not a judgment about the person.
Underplaced does not reduce every kind of accomplishment to one vague talent score. It keeps two estimates separate. g is evidence of general reasoning: abstraction, problem decomposition, transfer, error correction, and the ability to hold difficult constraints together. A is evidence of agency: choosing a goal, finishing it, improving it, and sustaining responsibility for the result.
A dedicated small YouTuber can provide strong evidence of A through years of self-directed production even when the videos say little about g. An original SoundCloud album can show composition, production, completion, and iteration. Its musical sophistication is not automatically general intelligence; the g estimate rises only when the work reveals structure, problem-solving, transfer, or unusually fast learning.
Every creative medium can contribute to both estimates, but the evidence is weighted according to what it actually shows. Popularity affects where the crawler looks first. Followers, views, likes, and downloads do not become g or A.
g remains primary. Agency can prioritize candidates whose reasoning evidence already clears the Stage-1 bar; it cannot compensate for weak evidence of g. The assessment supersedes the search estimate, and a paid work trial provides the stronger test of agency.
Promising candidates can choose to take a difficult assessment of verbal, quantitative, and fluid reasoning. It is designed with enough headroom to distinguish exceptional performance rather than placing everyone near the top at the same ceiling.
The idea comes from talent searches such as Duke TIP, which identified unusually able children by giving them tests intended for older students. Underplaced applies the same principle after formal schooling: search broadly, then use a harder opt-in measure.
Strong performers could be introduced to employers, paid work trials, or fellowships that let them demonstrate what they can do. Search estimates are not sold to employers and never substitute for the candidate’s own participation.
This site is a working product concept and technical prototype. It contains a fictional candidate walkthrough, a browser-based writing-analysis demo, and interactive maps of the proposed system. It does not currently crawl or score real people.