XThe X Algorithm
The numbers

X's own parameters, and the arithmetic.

The feed runs on values X publishes. Here they are, unedited, with one plain sentence each and the division that converts them into a figure you can hold in your head. Nothing on this page is an opinion, and nothing is behind a paywall.

Read from X's published configuration, commit b55151777cd8901111c7b5985a724f671f05d2c4, on November 19, 2026. A like is 0.708, so every figure in the right-hand column is that row's value divided by 0.708.

Two weight files. X writes its weights in two files, but both of its services load their live values from the same settings file, features/home-mixer/main/rust_home_mixer.yml. The feed server's file, home-mixer/params/param.rs, is mirrored from that settings file, so the values on this page come from it. Settings it doesn't list come from the ranking service's file, vm-ranker/params.rs. The date above is the day we read the code, not the day X changed a value. Why there are two files.

How to read a weight. Before X shows your post to a reader, it predicts how likely that reader is to take each action listed below. Each prediction is multiplied by that action's weight, and the results are added together into the score that sets the post's position in that reader's feed. A weight therefore applies to a predicted probability, not to a count of actions that have already occurred. The right-hand column is the ratio between two weights, not a number of likes.

Report-234report weight
Like0.708like weight
Report ÷ like-330.5085-234 ÷ 0.708, times a like

What a reader can do for you

Weights that increase a post's score. The right-hand column is each weight divided by the weight of a like, which is how the book refers to them.

ParameterWhat it isX's valueTimes a like
ReplyWeight How likely this reader is to reply to your post. 5 +7.0621
BidirectionalFollowReplyWeightBoost Added on top of the reply weight when you and this reader follow each other. 15 +21.1864
ShareViaCopyLinkWeight How likely this reader is to copy the link and send your post somewhere else. 20 +28.2486
ShareViaDmWeight How likely this reader is to send your post in a message on X. 5 +7.0621
QuoteWeight How likely this reader is to quote your post. 5 +7.0621
ShareWeight How likely this reader is to use the share button. 2.118 +2.9915
FollowAuthorWeight How likely this reader is to follow you from the post. 4 +5.6497
RetweetWeight How likely this reader is to repost you. 1 +1.4124
FavoriteWeight How likely this reader is to like your post. This is the unit every other weight is measured against. 0.708 +1
ClickWeight How likely this reader is to open the post itself. 0.274 +0.387
OpenLinkWeight How likely this reader is to open a link in your post. 0.2 +0.2825
VideoOpenWeight How likely this reader is to open a video. 0 0
PhotoExpandWeight How likely this reader is to expand a photo. 0.05 +0.0706
DwellWeight How likely this reader is to stop and read rather than scroll past. 0.05 +0.0706
ProfileClickWeight How likely this reader is to open your profile from the post. 0 0
PostUnexploredWeight A small addition applied to in-network candidates only. 0.015 +0.0212

What a reader can do against you

Weights that reduce a post's score. They are far larger than the positive weights because these actions are rare. X puts the probability of a report at more than a thousand times lower than the probability of a like, so a large weight is required for the prediction to affect ranking at all. The size of the number reflects how rarely the action occurs, not the effect of a single report.

ParameterWhat it isX's valueTimes a like
ReportWeight How likely this reader is to report your post. -234 -330.5085
MuteAuthorWeight How likely this reader is to mute you. -58.8 -83.0508
NotInterestedWeight How likely this reader is to tap "not interested in this post". -53.07 -74.9576
BlockAuthorWeight How likely this reader is to block you. -31.2 -44.0678
NotDwelledWeight How likely this reader is to scroll straight past without stopping. -0.02 -0.0282

How far outside your followers it travels

How a post is scored when it is shown to readers who do not follow you.

ParameterWhat it isX's valueTimes a like
OonWeightFactor A multiplier applied to the score of an out-of-network candidate. 0.75 —
TopicOonWeightFactor The same multiplier, for candidates pulled in by topic. 0.5 —
EnableOonRescoreForInNetworkRepliesRetweets Whether replies and reposts from inside your network are rescored this way. true —

How often you can appear

Limits on how much of one reader's feed a single author can occupy.

ParameterWhat it isX's valueTimes a like
AuthorDiversityDecay The decay used with the floor below. Each further post by the same author is multiplied by (1 - floor) x decay^k + floor, so at 0.5 and 0.25 the multipliers run 1, 0.625, 0.4375. 0.5 —
AuthorDiversityFloor The lowest that reduction is allowed to reach. 0.25 —
EnableAuthorDiversity Whether that limit is applied. true —

New posts and new accounts

How the feed scores a post or a reader it has no history for.

ParameterWhat it isX's valueTimes a like
ColdStartMaxPostAgeSecs The oldest a post can be and still qualify for the author cold-start check. 7200 —
LowImpressionsMaxPositionRatio The share of scored candidates within which a low-impression post can still take a cold-start slot. 0.97 —
EnableViewerColdStart Whether the author cold-start scoring is switched on for this reader. true —
NewUserMinEngagementFilterMaxAccountAgeSecs How long an account counts as new for the engagement filter. 1800 —

7200 seconds is 2 hours.

The machinery

Values that shape the feed as a whole rather than any single post.

ParameterWhat it isX's valueTimes a like
MaxPostsToCache The most candidates kept in the per-reader cache. 750 —
ExcludeServedTweetIdsNumber How many already-seen posts are kept out. 100 —
EngagementSignalsMaxPerType How many signals of each type are carried into scoring. 15 —
AdsTimeGapTSec The spacing applied when ads are placed in the feed. 4 —

What this does not tell you

These are the weights, not the whole system. They determine how much each predicted action contributes while a post is being scored for one reader. They do not say which posts get pulled in to be scored in the first place, what the models predict, which rules can hide a post before any of this runs, or what happens when an account is labelled. Those are the parts the book is for.

Everything else in the two parameter files, unsorted and unexplained (260 more)
AdsBlenderType"slot_spread"
AdsFollowingMinPosts5
AdsHistoryCacheTtlSecs300
AdsOnlyTimelineViewerIdsvec![]
AdsPageScoreHighThreshold0
AdsPageScoreKeepPctvec![]
AdsPageScoreLowThreshold0
AdsPageScoreSpacingvec![]
AdsPageScoreTopK20
AdsTimeGapClampHi2
AdsTimeGapClampLo0.5
AdsTimeGapMinOrganicGap3
AdsUserLoadEvenSpreadfalse
AdsUserStateKeepPctvec![]
AdsUserStateSpacingvec![]
AuthorExplorationBonus0
AuthorIsControlfalse
AuthorIsTreatmentfalse
AuthorPhoenixMoeEnabledfalse
AuthorServedMetricsAuthorIdsvec![]
AuthorZeroFavCodivertIsControlfalse
AuthorZeroFavCodivertIsTreatmentfalse
BidirectionalFollowDwellWeightBoost0
ColdStartBetaAlpha00.75
ColdStartBetaBeta049.25
ColdStartFollowerCap50000
ColdStartImpressionScale1
ColdStartImpressionThreshold200
ColdStartSlotMax16
ColdStartSlotMin15
ColdStartTrackedIds""
ColdStartTsTopK2
ComputeValueModelfalse
ContClickDwellTimeWeight0.385
ContDwellTimeWeight0.0075
DppEnabledtrue
DppGreedyOrderfalse
DppMaxSelectedRank150
DppTheta0.65
DppWindow0
DppWindowHalfDistance0
DppWindowSidLambda0
EnableAdAdjacentServedFilterfalse
EnableAdsAuthorBrandSafetyFallbacktrue
EnableAdsBrandSafetyVerdictV2false
EnableAdsFunnelTracetrue
EnableAdsHistoryXdcCacheWritetrue
EnableAdsInjectionLoggingtrue
EnableAdsOnlyTimelinefalse
EnableAdsSourcetrue
EnableAiTrendFeedbackContextfalse
EnableAllAuthorFollowHydrationtrue
EnableAuthorServedMetricsExperimentBucketfalse
EnableBidirectionalFollowHydrationtrue
EnableCachedPoststrue
EnableColdStartThompsonSamplingtrue
EnableContextFeaturestrue
EnableEngagementCountsHydrationtrue
EnableEntryIdRandomHashfalse
EnableExplicitEngagementSignalstrue
EnableFairShareRankingtrue
EnableFavHoldoutfalse
EnableFeedSurveyfalse
EnableFollowingRepliedUsersFacepilefalse
EnableHomeExcursionScorestrue
EnableImplicitEngagementSignalstrue
EnableInferredGenderHydrationtrue
EnableInventoryHoldoutfalse
EnableIpFeaturetrue
EnableJetfuelFrameGrokbotSuggestedTasksfalse
EnableJetfuelFrameMlbCarouselfalse
EnableJetfuelFrameNflfalse
EnableJetfuelFrameNflTimelinePinnedfalse
EnableJetfuelFrameNflTimelineRotationfalse
EnableJetfuelFrameSoccerfalse
EnableJetfuelFramesfalse
EnableMutualFollowJaccardHydrationfalse
EnableNewUserMinEngagementFilterfalse
EnableNewUserSoftNsfwFilterfalse
EnablePageDecodeRerankfalse
EnablePageLoggingtrue
EnablePageRlExplorationfalse
EnablePhoenixMOESourcefalse
EnablePhoenixOonRepliesfalse
EnablePhoenixRequestCacheSideEffecttrue
EnablePhoenixRetrievalStatsExperimentBucketfalse
EnablePhoenixScoreStatsExperimentBucketfalse
EnablePhoenixSourcetrue
EnablePopularPostsSourcetrue
EnablePostEngagersfalse
EnablePromptstrue
EnablePublishSeenIdsToKafkatrue
EnableQuotedVqvDurationCheckfalse
EnableRankingtrue
EnableRecentSearchQueriesHydrationfalse
EnableResponseDiversityStatsExperimentBucketfalse
EnableRetrievalSequenceHydrationtrue
EnableScoringSequenceHydrationtrue
EnableSemanticIdHydrationtrue
EnableSerendipityStatsExperimentBucketfalse
EnableServedFilterAllRequeststrue
EnableSidSourcefalse
EnableSimclustersSourcetrue
EnableTeacherScoreSideEffecttrue
EnableTopicFeedbackContextfalse
EnableTweetMixerSourcefalse
EnableUnderTheHoodfalse
EnableUnderTheHoodFramefalse
EnableUrtMigrationComponentstrue
EnableUserInstalledAppsHydrationtrue
EnableUserInstalledAppsV2false
EnableVideoCarouselfalse
EnableVideoCarouselAuthorDedupetrue
EnableVideoCarouselCandidatesLoggingtrue
EnableVideoCarouselCtrModelfalse
EnableVideoCarouselEmbeddingDedupetrue
EnableVideoCarouselHeaderIconfalse
EnableVideoCarouselHighRiskFiltertrue
EnableVideoCarouselMediumRiskFiltertrue
EnableVideoCarouselSeenFiltertrue
EnableVideoCarouselTakeFromFeedfalse
EnableWhoToFollowModuletrue
EnableXaiVfClienttrue
EngagementLabelsMaxCandidates200
ExcludeServedTweetIdsDuration10
FeedSurveyFatigueMinutes1440
FollowingRepliedUsersFacepileMaxPosts0
FollowingRepliedUsersFacepileMinUsers0
GrokAnalyzeWeight0
HistoryTopicsEnabledfalse
HistoryTopicsWindowDays28
InNetworkDwellTimeWeightFactor-1
InNetworkEngagementWeightFactor-1
InventoryHoldoutOriginalsPercent0
InventoryHoldoutRepliesPercent0
InventoryHoldoutRetweetsPercent0
JetfuelFrameGrokbotSuggestedTasksFatigueHours6
LogSlateContextfalse
MaxSeqLengthRetrieval1024
MaxSeqLengthScoring1024
MinVideoDurationMs10000
MultiplicativeVmEngageBias0
MultiplierPreOffsetfalse
NewUserAgeThresholdSecs0
NewUserMinEngagementFilterMaxResurrectionAgeSecs1800
NewUserMinEngagementFilterMetric"fav"
NewUserMinEngagementFilterUseRatiotrue
NewUserMinEngagementThreshold0.004
NewUserOonWeightFactor1.0E-5
NewUserSoftNsfwFilterMaxAccountAgeSecs604800
NewUserSoftNsfwFilterMaxScoringSeqLength100
NotDwelledPowerExponent0
NotDwelledPowerOffset0
OonDwellTimeWeightFactor-1
OonEngagementWeightFactor-1
PageDecodePoolN200
PageDecodeRewardHead0
PageDecodeSteps64
PageDecodeTemperature0
PageLogTtlSeconds3600
PageRlExplorationTemperature0.2
PageRlExplorationTopK64
PhoenixAggregationType"DENSE_WITH_LONG_DWELL_HOME_EXCURSION"
PhoenixColdStartMaxResults200
PhoenixExperimentOverrides""
PhoenixInferenceClusterId"Experiment1Fou"
PhoenixMOEMaxResults0
PhoenixMaxResults1000
PhoenixMoeCodivertViewerIsControlfalse
PhoenixMoeCodivertViewerIsTreatmentfalse
PhoenixMoeColdStartMaxResults200
PhoenixOonRepliesRootOnlyfalse
PhoenixRankerNewUserHistoryThreshold0
PhoenixRankerNewUserInferenceClusterId"Experiment1Fou"
PhoenixRequestCacheSideEffectTtlSeconds10800
PhoenixRetrievalAggregationType"DENSE_WITH_LONG_DWELL"
PhoenixRetrievalExcludeSeenPostsfalse
PhoenixRetrievalInferenceClusterId"Experiment1Fou"
PhoenixRetrievalMOEInferenceClusterId"Experiment2Memy04"
PhoenixRetrievalNewUserHistoryThreshold0
PhoenixRetrievalNewUserInferenceClusterId"Experiment1Lap7"
PhoenixRetrievalTopicInferenceClusterId"Experiment2Fou"
PhoenixScoresResultSize2800
PhoenixXdsRetrievalMaxRetries0
PhoenixXdsRetrievalUncompressedClustersListvec![]
PopularPostsMaxResults500
PostTopicsMaxCandidates200
PostTopicsTimeoutMs80
PostUnexploredIncludeOutOfNetworkfalse
PostVideoWeight0
ProfileVisitSecsWeight0
QuotedClickWeight0.05
QuotedVqvWeight0
RecentViewsRows64
RecentViewsSequenceLength256
RerankerHeadTag0
RetrievalCandidatesKafkaMaxCandidates100000
RetrievalCandidatesKafkaSamplePercent5
RetrievalValueModelWeights""
ShadowTrafficDefaultPercent0.3
ShadowTrafficPhoenixClusterAggregationTypesvec!["Experiment16Fou:DENSE_WITH_LONG_DWELL_HOME_EXCURSION".to_string()]
ShadowTrafficPhoenixClusterRatesvec![ "Experiment12Fou:0".to_string(), "Experiment18Fou:0.3".to_string(), "Experiment19Fou:0.3".to_string(), "Experiment22Fou:0.3".to_string(), "Experiment23Fou:0.3".to_string(), "Experiment24Fou:0.3".to_string(), "Experiment25Fou:0.3".to_string(), "Experiment26Fou:0.3".to_string(), "Experiment27Fou:0.3".to_string(), "Experiment28Fou:0.3".to_string(), "Experiment29Fou:0.3".to_string() ]
ShadowTrafficPhoenixClustersListvec![]
ShadowTrafficPhoenixRetrievalClustersListvec![]
SidServerEndpoint"xds://xai-recsys-sid-v8-root.prod.recsys-infra:grpc"
SidSourceMaxPerSeed100
SidSourceMaxResults800
SidSourceMaxSeeds50
SidSourceMinPrefixDepth3
SimclustersMaxCandidateAgeHours48
TeacherActionIndicesvec![1, 4, 5, 6, 11, 12, 13, 14, 16, 17, 21, 23, 25, 29, 30, 32, 33, 37, 38, 39, 59]
TeacherTtlSeconds3000
ThunderAlgorithm"default"
ThunderMaxResults1200
TopicDppEnabledfalse
TopicFilteringId"PostBased50Pct"
TopicFilteringOverrides""
TweetMixerMaxResults800
UasSourceDataType"NEW_UUA_ARROW"
UseXdsForUastrue
UserActionsClusterId"Fou"
UserDebiasAlpha0
UserDebiasClipLogit2
UserDebiasEnabledfalse
UserDebiasHeadAlpha""
UserDebiasHeadClipLogit""
UserDebiasHeads""
UserDebiasMaxArtifactAgeSecs3600
UserVideoContinuationWeight0
VMRankerClusterId"Experiment1"
VMRankerDppHistoryMaxItems50
VMRankerEnableFallbackfalse
VMRankerSendDebiasInputstrue
VMRankerSendDppHistoryfalse
VMRankerSendEngagementIneligiblefalse
VMRankerSendHistoryTopicsfalse
VMRankerSendPacingInputsfalse
VMRankerSendPostTopicsfalse
ValueModelModeweighted
VideoCarouselCtrModelName"lr_v8_search_video_1_2026-11-03_no_high_risk"
VideoCarouselEmbeddingDedupeThreshold0.72
VideoCarouselEmbeddingVersion"v8_search_1"
VideoCarouselExtraCandidates30
VideoCarouselFatigueMinutes60
VideoCarouselImmersiveAggregationType"IMMERSIVE_LONG_DWELL"
VideoCarouselImmersiveClusterId"Experiment3Fou"
VideoCarouselImmersiveVqvWeight0
VideoCarouselNotInterestedWeight0
VideoCarouselOpenTimesImmersiveDwellWeight0
VideoCarouselPosition6
VideoCarouselSemanticIdDedupeLevels3
VideoCarouselVideoContinuationWeight0
VideoCarouselVideoOpenWeight1
VideoCarouselWeightedScoreWeight0
VideoContinuationWeight0
ViewerZeroFavCodivertIsTreatmentfalse
VqvWeight0
WeightPerturbationSalt""
WeightPerturbationSigma0
WhoToFollowFatigueHours30

The book is the other half. This page is the numbers. The book tells you what to do about them. It is updated every month, and updates are free.

See what it costs