Researchers say Twitter's open sourced For You code is missing context to fully explain the service's recommendations, despite a supposed push for transparency
part of the CEO's new approach to media inquiries.” https://www.theatlantic.com/ ... @theatlantic : Twitter's decision to “open source” its algorithm is unprecedented—and filled with problems, @mimbsy writes. https://on.theatln.tc/nBAnB7h @igorbrigadir : @mimbsy Yep, pretty spot on. Digging in to the code is way more fun for people who work with these kinds of systems, but for most people, not very exciting at best, or misleading and confusing at worst. A few contributors are keeping technical notes here tho: https://github.com/... James Fawcette / @thefawcette : “#Twitter has become less transparent since #Musk, not more, despite showy announcements such as this one.” .. policies once used to safeguard users “are now being rewritten in obviously nonsensical ways to fit with the whims of its owner.” https://www.theatlantic.com/ ... @lizjosullivan : My first time in @TheAtlantic stating the obvious: “If Mr Musk truly valued transparency or the equal expression of all voices in this so-called town square, he would invite outside auditors to conduct & publish independent reviews of the technology” but somehow I think he won't. https://twitter.com/... Damon Beres / @dlberes : Such a great piece about Twitter's bogus “open sourcing,” the meaning of true algorithmic transparency, and the many ways Musk has only made this platform more chaotic and opaque. (NPR as state-affiliated media? Come on!) by @mimbsy https://www.theatlantic.com/ ...
Context & Ripple Effects
This lands almost exactly a year into a promise arc: in April 2022, Musk pledged to make the algorithms open source to increase trust, and skeptics warned at the time that releasing code alone wouldn't deliver transparency because recommendation engines run on an immense data firehose. Earlier proposals went further still, suggesting a marketplace where anyone could swap in their own algorithms.
Days ago Twitter published part of that code on GitHub — the For You recommender but not the ad algorithm — and The Atlantic's piece plus independent engineers digging through the repo now supply the first verdict: the release is real but incomplete, with community technical notes filling gaps the code itself leaves.
First-order effects
- Researchers and outside contributors auditing the GitHub repo can read the For You ranking logic but cannot fully explain what users actually see, because the training context and data pipeline are absent from the drop.
- The ad recommendation algorithm stays closed, so the most commercially sensitive half of Twitter's recommendation stack remains unauditable despite the transparency framing.
Second-order effects
- Pressure shifts to Twitter to publish the missing layers — the data corpus and feature context — or accept that the release functions as branding rather than verification; competitors facing similar calls now have a template showing partial code dumps as a low-cost response.
- A volunteer ecosystem of external note-takers and fork-maintainers becomes the de facto explanation layer, moving interpretive authority over the feed away from the company.
Third-order effects
- If this pattern holds across platforms, 'open source' claims will be judged on whether outsiders can reproduce outcomes, not on whether code exists in public — pushing regulators and auditors toward demanding data access and reproducibility rather than repository access.
- Recommendation transparency risks bifurcating into a two-tier norm: consumer-facing feeds get symbolic releases while revenue-bearing systems like ad targeting stay proprietary indefinitely.
The trend: Platform 'open sourcing' is becoming a test of nominal versus usable openness, where publishing code without the data context behind it satisfies the announcement but not the audit.