X Algorithm Weights, Right Now
Every ranking weight, parsed directly from param.rs. These multiply predicted probabilities — not raw counts.
Positive Weights (16)
| Name | Value | Description |
|---|---|---|
| 15 | Bonus on reply weight for mutual-follow original posts | |
| 0.4 | Predicted probability you'll click into the post | |
| 0.004 | Continuous dwell time contribution | |
| 0.5 | Predicted probability you'll like the post | |
| 4 | Predicted probability you'll follow the author | |
| 0.2 | Predicted probability you'll open a link | |
| 0.05 | Predicted probability you'll expand a photo | |
| 0.02 | Exploration: posts outside your usual patterns | |
| 5 | Predicted probability you'll quote | |
| 5 | Predicted probability you'll reply | |
| 1 | Predicted probability you'll retweet | |
| 20 | Predicted probability you'll copy the link | |
| 5 | Predicted probability you'll share via DM | |
| 2 | Predicted probability you'll share | |
| 0.05 | Predicted probability you'll open a video | |
| 0.05 | Video quality view weight |
Negative Weights (5)
| Name | Value | Description |
|---|---|---|
| -31.2 | Predicted probability you'll block the author | |
| -58.8 | Predicted probability you'll mute the author | |
| -0.02 | Penalty for posts you didn't dwell on | |
| -43.2 | Predicted probability you'll mark not interested | |
| -234 | Predicted probability you'll report the post |
Adjustments & Factors (8)
| Name | Value | Description |
|---|---|---|
| 0.5 | Decay factor for same-author posts in one session | |
| 0.25 | Floor for author diversity decay | |
| 1000 | New-author boost applies below this follower count | |
| 1000 | New authors boosted until this many impressions | |
| 16 | Maximum feed position for new-author boost | |
| 15 | Minimum feed position for new-author boost | |
| 0.75 | Discount applied to out-of-network posts | |
| 0.5 | OON discount when topic filtering is active |
Weights Misconceptions
Do these weights multiply raw engagement counts?
No. The repo is explicit: weights multiply the predicted probability of each action, not raw counts. A ReportWeight of -234 does not mean one report cancels 468 likes. It scales how much the predicted likelihood of a report affects the ranking score.
Why is the Report weight so negative (-234)?
The baseline probability that any given viewer will report a post is extremely low — far lower than the probability of a like. A large negative weight ensures that when the model does predict a meaningful report probability, it can actually move the ranking. It is not a 1:1 cancellation against likes.
What is the difference between weights and adjustments?
Weights multiply predicted action probabilities to produce the base score. Adjustments (OonWeightFactor, AuthorDiversityDecay, ColdStart*) are applied after scoring to reshape the final ranking — discounts for out-of-network posts, decay for repeated authors, and boosts for new accounts.
How fresh is this data?
When our daily tracker detects a change in the repo, the live JSON is updated and this page reflects it within hours. If the tracker has not run or the live file is unavailable, the page falls back to a verified snapshot (marked "snapshot"). The source badge tells you which.