About X Radar

Who builds this

X Radar is built by Tarun Kumar (Founder, X Radar). You can reach me at @TarunDevYT on X, or at help@xradar.dev. It is an independent product, operated from India - not a team, not a content farm. Every article on this site is written by me.

Where the numbers come from

X publishes its recommendation algorithm at xai-org/x-algorithm. X Radar is built by reading that source directly - the ranking scorers, the candidate filters, and the Grox content classifiers that screen a post before it is ever ranked.

The scorer is pinned to a specific commit rather than "the algorithm" in the abstract. The version live right now is xai-2026-07-04, built from commit 0bfc2795d3. A watcher tracks that repository, so when the source changes the product is re-checked against it instead of quietly going stale.

What we claim, and what we don't

This is the part most tools skip. X Radar models 31 ranking signals, and each one is labelled by how well it is actually evidenced:

• 27 are confirmed mechanics - the signal provably exists in the published source, and the file it lives in is named in the app.
• The rest are community estimates- observed behaviour that fits the code but isn't published. They are shown as estimated, never dressed up as fact.

One honest limit worth stating plainly: X publishes the ranking code, but the file holding the exact weight constants is not in the public repository. So the heads - which actions the ranker scores - are verifiable, while several of the exact multipliers are informed estimates. Anywhere that distinction matters, the app says so.

How the product uses it

When you paste a draft, X Radar asks a model for the per-impression probability of each engagement the ranker predicts, then multiplies those by the real open-source weights and sums them in our own code. The arithmetic is deterministic and auditable - the model estimates inputs, it does not hand back a score. Projections are then grounded in your own account's median impressions, so the number reflects your account rather than an abstract benchmark.

We also grade ourselves. Every reach projection is checked against the real impressions the post earned about 24 hours later, and the hit rate is shown back to you - misses included.

Why trust any of this

Because you can check it. The repository is public, the commit is pinned, and every signal in the app names the file it comes from. If something here is wrong, it is wrong in a way you can point at - which is the opposite of how most growth advice works.