Why Everyone's Estimate Is Starting to Look the Same
The Consensus Flywheel: how aggregate taste becomes the standard, and how a small company can keep setting it
There is a thing that happens on YouTube and TikTok that everybody understands intuitively and almost nobody builds on purpose.
A video does numbers. Because it did numbers, more people make videos like it. Because more people make them, that shape becomes what a video in that niche looks like. And then the next big trend is not invented from scratch. It is a mutation of the thing that just won. Watch enough and you can feel the format hardening in real time: the same hook, the same cut, the same on-screen caption style. Then it turns over, and a new one hardens.
I have become convinced the same loop runs under something far less glamorous than short-form video: business documents. A contractor's estimate. A consultant's proposal. An engagement letter. There is, right now, a shape that a winning estimate has, and the people who write them are converging on it without anyone deciding they should. I want to name that loop, and I want 1nicorn to be the thing that sets it.
Sources
- Matthew effect: recognition accrues to the already-recognised (Merton, Science 1968)science.org
- Cumulative advantage formalised as success-breeds-success (Price, JASIS 1976)asistdl.onlinelibrary.wiley.com
- Preferential attachment produces scale-free, power-law popularity (Barabasi and Albert, Science 1999)science.org
- Mimetic isomorphism: firms imitate winning peers under uncertainty (DiMaggio and Powell, ASR 1983)jstor.org
- Informational cascades: rational to copy others and ignore your own signal (Bikhchandani, Hirshleifer and Welch, JPE 1992)journals.uchicago.edu
- QWERTY lock-in: a small early advantage can lock in a non-optimal standard (David, AER 1985)jstor.org
- Increasing returns and lock-in by historical events (Arthur, Economic Journal 1989)academic.oup.com
- Dominant-design cycle: ferment, dominant design, then a discontinuity (Anderson and Tushman, ASQ 1990)jstor.org
- Blind variation and selective retention as the engine of novelty (Campbell, Psych Review 1960)psycnet.apa.org
- The truth about data network effects: genuine ones are rarer than claimed (Currier, NFX 2020)nfx.com
- Algorithmic confounding increases homogeneity and decreases utility (Chaney, Stewart and Engelhardt, RecSys 2018)dl.acm.org
- AI models collapse when trained on recursively generated data (Shumailov et al., Nature 2024)nature.com
- Thompson sampling: balancing exploit and explore (Thompson, Biometrika 1933)academic.oup.com
- Finite-time analysis of the multi-armed bandit problem, UCB (Auer, Cesa-Bianchi and Fischer, ML 2002)link.springer.com