Deep dossier · 7 pages · 13,000 words · 30 August 2026

What to build first

Contra Labs has published eight datasets, thirty-nine studies and two papers in five months. Combined, the datasets have about 5,100 downloads and the papers have zero citations. A single release from a company nobody in this market was watching got 1,784 downloads in fourteen days. This is the record of what the field has actually shipped, what it earned, and what nobody has built yet.

Start with the first artefact

01

The read

What publishing actually bought the incumbents, which is less than anyone assumes.

What publishing actually bought them

Thirty-nine studies, two papers, eight datasets and no GitHub organisation bought Contra Labs 5,133 all-time downloads and zero citations. Patronus out-published them on their own moat in fourteen days; AfterQuery's 148-row FinanceQA has 17 citations and a repo. Volume bought nothing, and the pattern that did work is small, sharp, cited and shipped with code.

02

Start here

The one artefact to build first, specified closely enough to begin on Monday.

Rebuild the Human Creativity Benchmark

Take the flagship benchmark of the company being modelled, rebuild it properly, and make the contribution the reliability methodology rather than the leaderboard. Nobody has published how many designers per item an aesthetic ranking needs before it stabilises, by criterion — Google says more than ten and Contra's published rubric says three. That sweep is a re-analysis, not a new field, and it is the single most valuable unclaimed artefact in the space.

03

What already exists

Every dataset, paper, benchmark and repo the field has shipped, with its download counts.

Every public artefact, by organisation

The full record for six organisations: every dataset with repo id, date, licence, size, downloads and likes; every paper with arXiv id and citation count; every repository with stars. Design Arena holds a paid Hugging Face org that is completely empty, Taste Labs has published nothing at all, and Gamma's org is empty while the Lica repos it bought are still live.

The eleven problems a rival published

Taste Labs published eleven open problems on 16 August 2026, the same day it opened a fellowship. It is a hiring instrument and it is also a public statement that a funded rival has not solved any of them — two of which are Contra's live products. Answering a named competitor's published open problem is the cheapest citation available in this market.

04

What does not exist

The unclaimed artefacts, ranked by how much attention each would earn per week of work.

What nobody has built

Eight absences, ranked, each searched specifically and each with what it would take and what it would earn. Zero design reward-model weights exist from any of the six organisations examined; no design benchmark anywhere is held out; nobody has run designer-panel against crowd-vote on the same items though both companies whose business models depend on the answer have the data; and the trajectory position is now contested.

05

Mechanics

How an artefact actually travels, and what this page could not establish.

How an artefact travels

Licence, bundle, host, cadence. Three of AfterQuery's six datasets carry no licence at all, which is a real defect; every cited artefact in the record ships with a repository and every uncited one does not; and Contra's 39 studies could not be stacked into anything because no two of them used the same instrument.

What this plan could not establish

Ranked by how much the answer moves the build. The top item is that the technical stack was not researched at all — environment interfaces, capture APIs, how FigmaTrace was actually made, and what an aesthetic reward model costs to train — which means no environment item can be costed in front of a buyer yet. Then four smaller questions, each with the specific action that settles it.