The Chat Test

Every app idea now competes with a free chat window that already knows the domain. This is a research sweep built around that one question. Three analysts independently derived a rubric for when a purpose-built app genuinely beats asking a chatbot; it came back with 11 axes and 22 ways to die. Then 81 concepts across 27 skill domains — from game telemetry, ceramics and darkroom printing to foraging, weld auditing and tonal language learning — were each attacked by two independent skeptics, one asking what chat gives you free right now, one asking whether the thing survives real file formats, real error bars and real incumbents. Both defaulted to killing it. Seventeen concepts survived. A deliberate negative control (prose craft, recipe substitution, code review — territory chat should obviously win) killed all three of its own concepts at the lowest scores in the sweep. That is the result that makes the rest readable: the rubric discriminates rather than blessing whatever it is shown. Inside: the full rubric and kill list; eight cross-cutting patterns that predict a verdict before you design anything (byproduct ledgers live and homework ledgers die; bundled third-party facts rot invisibly; the error bar must be smaller than the decision threshold; provable refusal is the product); a domain-by-domain results table; an appeals docket where a completeness critic's attacks on the sweep's own verdicts are ruled on one by one; a ranked top twelve, each with the caveat most likely to kill it; and an unusually blunt section on how much of this to believe — including the fact that no chat transcript was ever actually executed.