How is a finding on PlayerSells promoted from a measurement to a published claim?

A measurement becomes a published finding only after clearing five independent rules: a Benjamini-Hochberg adjusted p-value at or below 0.05 inside its pre-registered testing family, at least 50 contributing accounts, an effect whose 95% interval stays at least 10% away from zero, agreement between two halves of a stable account split, and repetition across 3 consecutive passes. Measurements that fail are kept on record rather than discarded.

Why does PlayerSells correct for multiple comparisons on X engagement data?

Each pass tests thousands of hypotheses at once. At a 95% confidence level roughly one test in twenty crosses the line by chance, so a few thousand simultaneous tests produce over a hundred confident-looking results that are noise. Benjamini-Hochberg controls the share of published findings expected to be false at 0.05, which is a statement that can be checked; an uncorrected per-test threshold is not.

Does a finding holding for several passes mean it was independently replicated?

No, and the page never claims it. Each pass measures a 90-day window a week apart, so two consecutive passes share most of their evidence and a streak is a stability measure. The independent check is a split-half over accounts inside every pass: the same account always lands in the same half, and a finding must hold in both halves to be published.

Why does PlayerSells publish findings it has withdrawn?

Because a research feed that only ever adds is indistinguishable from one that never re-checks anything. Withdrawn and reversed findings are the visible evidence that every published claim is re-tested on every pass, and they carry the date the reading changed.

What changed in the X research, and when

Every finding on these pages is re-tested on a schedule. The most recent pass tested 3,492 hypotheses across 3,294 tracked X accounts, 1,158 of which survived correction for multiple comparisons and 0 of which cleared every publication rule. This page lists what was added, revised and withdrawn.

Key findings

The PlayerSells X research loop tested 3,492 hypotheses in its most recent pass on 19 Aug 2026, of which 1,158 survived a Benjamini-Hochberg false-discovery-rate correction at 0.05 and 0 cleared every publication rule.
Findings are measured across 3,294 tracked X accounts and 201,968 original posts over a 90-day window, every one of which has at least 1,644,259 followers.

Research changelog: findings added, revised and withdrawn
Date	Change	What changed
19 Aug 2026	Ruled out	The effect of word_count = w20_39 differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -6.4% to +2.9%, which rules out any change larger than 10%, measured across 959 tracked X accounts and 6,460 posts.
19 Aug 2026	Ruled out	The effect of word_count = w10_19 differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -4.1% to +5.0%, which rules out any change larger than 10%, measured across 872 tracked X accounts and 5,290 posts.
19 Aug 2026	Ruled out	The effect of weekend = yes differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -7.5% to +2.8%, which rules out any change larger than 10%, measured across 948 tracked X accounts and 5,276 posts.
19 Aug 2026	Ruled out	The effect of weekend = yes differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -8.5% to +0.3%, which rules out any change larger than 10%, measured across 948 tracked X accounts and 5,276 posts.
19 Aug 2026	Ruled out	The effect of weekend = yes differs between posts without media and posts with media does not change bookmarks by a meaningful amount: the 95% interval is -9.5% to +1.4%, which rules out any change larger than 10%, measured across 948 tracked X accounts and 5,276 posts.
19 Aug 2026	Ruled out	The effect of source_type = native differs between posts under 120 characters and posts of 120 characters or more does not change engagement per impression by a meaningful amount: the 95% interval is -4.5% to +5.8%, which rules out any change larger than 10%, measured across 262 tracked X accounts and 2,900 posts.
19 Aug 2026	Ruled out	The effect of post_gap = under_1h differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -5.9% to +7.6%, which rules out any change larger than 10%, measured across 641 tracked X accounts and 5,100 posts.
19 Aug 2026	Ruled out	The effect of post_gap = under_1h differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -7.4% to +2.2%, which rules out any change larger than 10%, measured across 641 tracked X accounts and 5,100 posts.
19 Aug 2026	Ruled out	The effect of post_gap = under_1h differs between posts without media and posts with media does not change bookmarks by a meaningful amount: the 95% interval is -6.8% to +9.2%, which rules out any change larger than 10%, measured across 641 tracked X accounts and 5,100 posts.
19 Aug 2026	Ruled out	The effect of post_gap = over_24h differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -7.5% to +8.8%, which rules out any change larger than 10%, measured across 449 tracked X accounts and 2,038 posts.
19 Aug 2026	Ruled out	The effect of post_gap = 6_24h differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -0.0% to +9.1%, which rules out any change larger than 10%, measured across 798 tracked X accounts and 3,229 posts.
19 Aug 2026	Ruled out	The effect of post_gap = 1_6h differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -7.4% to +2.1%, which rules out any change larger than 10%, measured across 945 tracked X accounts and 4,437 posts.
19 Aug 2026	Ruled out	The effect of post_gap = 1_6h differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -2.8% to +7.4%, which rules out any change larger than 10%, measured across 945 tracked X accounts and 4,437 posts.
19 Aug 2026	Ruled out	The effect of post_gap = 1_6h differs between posts without media and posts with media does not change engagement by a meaningful amount: the 95% interval is -8.2% to +7.9%, which rules out any change larger than 10%, measured across 945 tracked X accounts and 4,437 posts.
19 Aug 2026	Ruled out	The effect of post_gap = 1_6h differs between posts without media and posts with media does not change bookmarks by a meaningful amount: the 95% interval is -7.9% to +6.8%, which rules out any change larger than 10%, measured across 945 tracked X accounts and 4,437 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = night differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -9.4% to +6.1%, which rules out any change larger than 10%, measured across 674 tracked X accounts and 4,093 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = night differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -7.5% to +3.4%, which rules out any change larger than 10%, measured across 674 tracked X accounts and 4,093 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = morning differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -7.3% to +4.3%, which rules out any change larger than 10%, measured across 735 tracked X accounts and 4,582 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = morning differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -2.0% to +8.2%, which rules out any change larger than 10%, measured across 735 tracked X accounts and 4,582 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = evening differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -6.5% to +5.5%, which rules out any change larger than 10%, measured across 902 tracked X accounts and 5,716 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = evening differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -6.7% to +5.4%, which rules out any change larger than 10%, measured across 902 tracked X accounts and 5,716 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = evening differs between posts without media and posts with media does not change engagement by a meaningful amount: the 95% interval is -8.7% to +8.2%, which rules out any change larger than 10%, measured across 902 tracked X accounts and 5,717 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = evening differs between posts without media and posts with media does not change bookmarks by a meaningful amount: the 95% interval is -9.8% to +2.4%, which rules out any change larger than 10%, measured across 902 tracked X accounts and 5,717 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -4.7% to +5.6%, which rules out any change larger than 10%, measured across 1,127 tracked X accounts and 8,072 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -5.8% to +1.5%, which rules out any change larger than 10%, measured across 1,127 tracked X accounts and 8,072 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without media and posts with media does not change engagement by a meaningful amount: the 95% interval is -6.1% to +8.2%, which rules out any change larger than 10%, measured across 1,127 tracked X accounts and 8,073 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without media and posts with media does not change bookmarks by a meaningful amount: the 95% interval is -1.4% to +7.5%, which rules out any change larger than 10%, measured across 1,127 tracked X accounts and 8,073 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = night differs between posts without a link and posts containing a link does not change reach (impressions) by a meaningful amount: the 95% interval is -9.1% to +2.7%, which rules out any change larger than 10%, measured across 555 tracked X accounts and 3,398 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = night differs between posts without a link and posts containing a link does not change engagement per impression by a meaningful amount: the 95% interval is -8.6% to +3.1%, which rules out any change larger than 10%, measured across 555 tracked X accounts and 3,398 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = night differs between posts without a link and posts containing a link does not change bookmarks by a meaningful amount: the 95% interval is -9.5% to +6.5%, which rules out any change larger than 10%, measured across 555 tracked X accounts and 3,401 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = morning differs between posts without a link and posts containing a link does not change reach (impressions) by a meaningful amount: the 95% interval is -8.8% to +2.2%, which rules out any change larger than 10%, measured across 648 tracked X accounts and 4,266 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = morning differs between posts without a link and posts containing a link does not change engagement per impression by a meaningful amount: the 95% interval is -7.8% to +4.5%, which rules out any change larger than 10%, measured across 648 tracked X accounts and 4,266 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = morning differs between posts without a link and posts containing a link does not change bookmarks by a meaningful amount: the 95% interval is -9.6% to +2.4%, which rules out any change larger than 10%, measured across 648 tracked X accounts and 4,267 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = evening differs between posts without a link and posts containing a link does not change engagement per impression by a meaningful amount: the 95% interval is -6.2% to +3.0%, which rules out any change larger than 10%, measured across 743 tracked X accounts and 4,616 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without a link and posts containing a link does not change reach (impressions) by a meaningful amount: the 95% interval is -5.8% to +2.4%, which rules out any change larger than 10%, measured across 981 tracked X accounts and 6,798 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without a link and posts containing a link does not change engagement per impression by a meaningful amount: the 95% interval is -4.2% to +4.9%, which rules out any change larger than 10%, measured across 981 tracked X accounts and 6,798 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without a link and posts containing a link does not change engagement by a meaningful amount: the 95% interval is -8.7% to +5.5%, which rules out any change larger than 10%, measured across 981 tracked X accounts and 6,800 posts.
19 Aug 2026	Ruled out	The effect of part_of_day = afternoon differs between posts without a link and posts containing a link does not change bookmarks by a meaningful amount: the 95% interval is -7.9% to +2.1%, which rules out any change larger than 10%, measured across 981 tracked X accounts and 6,800 posts.
19 Aug 2026	Ruled out	The effect of length_bucket = medium differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -6.4% to +3.4%, which rules out any change larger than 10%, measured across 978 tracked X accounts and 6,710 posts.
19 Aug 2026	Ruled out	The effect of length_bucket = long differs between posts without media and posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -6.9% to +4.0%, which rules out any change larger than 10%, measured across 776 tracked X accounts and 4,169 posts.
19 Aug 2026	Ruled out	The effect of day_position = 3rd+ differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -6.0% to +9.3%, which rules out any change larger than 10%, measured across 616 tracked X accounts and 4,783 posts.
19 Aug 2026	Ruled out	The effect of day_position = 2nd differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -7.8% to +5.4%, which rules out any change larger than 10%, measured across 832 tracked X accounts and 2,734 posts.
19 Aug 2026	Ruled out	The effect of day_position = 1st differs between posts without media and posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -8.3% to +3.7%, which rules out any change larger than 10%, measured across 756 tracked X accounts and 3,969 posts.
19 Aug 2026	Ruled out	Posting with word_count = w40+, among posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -4.2% to +0.5%, which rules out any change larger than 10%, measured across 2,122 tracked X accounts and 31,771 posts.
19 Aug 2026	Ruled out	Posting with word_count = w40+, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -6.9% to -2.3%, which rules out any change larger than 10%, measured across 2,122 tracked X accounts and 31,771 posts.
19 Aug 2026	Ruled out	Posting with word_count = w40+, among posts with media does not change engagement by a meaningful amount: the 95% interval is -9.0% to -2.5%, which rules out any change larger than 10%, measured across 2,122 tracked X accounts and 31,775 posts.
19 Aug 2026	Ruled out	Posting with word_count = w40+, among posts with media does not change bookmarks by a meaningful amount: the 95% interval is +1.1% to +7.0%, which rules out any change larger than 10%, measured across 2,122 tracked X accounts and 31,775 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -9.2% to -6.2%, which rules out any change larger than 10%, measured across 2,651 tracked X accounts and 40,822 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -8.6% to -5.1%, which rules out any change larger than 10%, measured across 2,651 tracked X accounts and 40,822 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is +0.3% to +4.6%, which rules out any change larger than 10%, measured across 2,478 tracked X accounts and 36,187 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -4.1% to -0.3%, which rules out any change larger than 10%, measured across 2,478 tracked X accounts and 36,187 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts with media does not change engagement by a meaningful amount: the 95% interval is -3.7% to +1.8%, which rules out any change larger than 10%, measured across 2,478 tracked X accounts and 36,192 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts with media does not change bookmarks by a meaningful amount: the 95% interval is -3.1% to +1.3%, which rules out any change larger than 10%, measured across 2,478 tracked X accounts and 36,192 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts without media does not change reach (impressions) by a meaningful amount: the 95% interval is -8.2% to -3.9%, which rules out any change larger than 10%, measured across 1,231 tracked X accounts and 16,194 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts without media does not change engagement per impression by a meaningful amount: the 95% interval is +0.9% to +7.5%, which rules out any change larger than 10%, measured across 1,231 tracked X accounts and 16,194 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts without media does not change engagement by a meaningful amount: the 95% interval is -7.5% to +0.7%, which rules out any change larger than 10%, measured across 1,231 tracked X accounts and 16,200 posts.
19 Aug 2026	Ruled out	Posting with word_count = w20_39, among posts without media does not change bookmarks by a meaningful amount: the 95% interval is -7.7% to -2.2%, which rules out any change larger than 10%, measured across 1,231 tracked X accounts and 16,200 posts.
19 Aug 2026	Ruled out	Posting with word_count = w1_9, among posts without media does not change reach (impressions) by a meaningful amount: the 95% interval is -6.5% to -0.9%, which rules out any change larger than 10%, measured across 850 tracked X accounts and 6,474 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts without media does not change reach (impressions) by a meaningful amount: the 95% interval is -0.9% to +4.3%, which rules out any change larger than 10%, measured across 1,253 tracked X accounts and 20,737 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts without media does not change engagement per impression by a meaningful amount: the 95% interval is -7.2% to -2.5%, which rules out any change larger than 10%, measured across 1,253 tracked X accounts and 20,737 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts without media does not change engagement by a meaningful amount: the 95% interval is -6.6% to -0.5%, which rules out any change larger than 10%, measured across 1,253 tracked X accounts and 20,749 posts.
19 Aug 2026	Ruled out	Posting with word_count = w10_19, among posts without media does not change bookmarks by a meaningful amount: the 95% interval is -7.0% to -1.2%, which rules out any change larger than 10%, measured across 1,253 tracked X accounts and 20,749 posts.
19 Aug 2026	Ruled out	Posting with weekend = yes, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is +0.6% to +4.0%, which rules out any change larger than 10%, measured across 2,579 tracked X accounts and 33,180 posts.
19 Aug 2026	Ruled out	Posting with weekend = yes, among posts with media does not change bookmarks by a meaningful amount: the 95% interval is +3.7% to +9.1%, which rules out any change larger than 10%, measured across 2,579 tracked X accounts and 33,181 posts.
19 Aug 2026	Ruled out	Posting with weekend = yes, among posts without media does not change engagement per impression by a meaningful amount: the 95% interval is -3.3% to +2.3%, which rules out any change larger than 10%, measured across 1,223 tracked X accounts and 12,042 posts.
19 Aug 2026	Ruled out	Posting with video = yes, among posts without a link does not change engagement by a meaningful amount: the 95% interval is -1.1% to +6.4%, which rules out any change larger than 10%, measured across 2,360 tracked X accounts and 43,523 posts.
19 Aug 2026	Ruled out	Posting with thread = yes, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -3.7% to +6.1%, which rules out any change larger than 10%, measured across 553 tracked X accounts and 3,205 posts.
19 Aug 2026	Ruled out	Posting with source_type = native, among posts of 120 characters or more does not change reach (impressions) by a meaningful amount: the 95% interval is -8.8% to -1.4%, which rules out any change larger than 10%, measured across 786 tracked X accounts and 21,086 posts.
19 Aug 2026	Ruled out	Posting with source_type = native, among posts of 120 characters or more does not change bookmarks by a meaningful amount: the 95% interval is -0.6% to +6.4%, which rules out any change larger than 10%, measured across 786 tracked X accounts and 21,094 posts.
19 Aug 2026	Ruled out	Posting with post_gap = under_1h, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is +2.6% to +6.4%, which rules out any change larger than 10%, measured across 2,238 tracked X accounts and 53,975 posts.
19 Aug 2026	Ruled out	Posting with post_gap = under_1h, among posts with media does not change engagement by a meaningful amount: the 95% interval is -9.8% to -4.8%, which rules out any change larger than 10%, measured across 2,238 tracked X accounts and 53,985 posts.
19 Aug 2026	Ruled out	Posting with post_gap = over_24h, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -7.3% to -2.7%, which rules out any change larger than 10%, measured across 1,580 tracked X accounts and 15,076 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 6_24h, among posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -1.1% to +2.5%, which rules out any change larger than 10%, measured across 2,555 tracked X accounts and 21,284 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 6_24h, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -0.1% to +3.1%, which rules out any change larger than 10%, measured across 2,555 tracked X accounts and 21,284 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 6_24h, among posts with media does not change engagement by a meaningful amount: the 95% interval is -1.1% to +3.2%, which rules out any change larger than 10%, measured across 2,555 tracked X accounts and 21,284 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 6_24h, among posts with media does not change bookmarks by a meaningful amount: the 95% interval is -2.1% to +3.8%, which rules out any change larger than 10%, measured across 2,555 tracked X accounts and 21,284 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 1_6h, among posts with media does not change reach (impressions) by a meaningful amount: the 95% interval is -1.9% to +1.3%, which rules out any change larger than 10%, measured across 2,622 tracked X accounts and 29,755 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 1_6h, among posts with media does not change engagement per impression by a meaningful amount: the 95% interval is -0.6% to +2.7%, which rules out any change larger than 10%, measured across 2,622 tracked X accounts and 29,755 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 1_6h, among posts with media does not change engagement by a meaningful amount: the 95% interval is -0.9% to +3.6%, which rules out any change larger than 10%, measured across 2,622 tracked X accounts and 29,755 posts.
19 Aug 2026	Ruled out	Posting with post_gap = 1_6h, among posts with media does not change bookmarks by a meaningful amount: the 95% interval is -0.2% to +3.6%, which rules out any change larger than 10%, measured across 2,622 tracked X accounts and 29,755 posts.

Reversals and withdrawals are listed first. Every entry states its own numbers and sample so it can be quoted on its own.

How this was measured

Method: Public posts are collected per account and replies and reposts are discarded. Within each account, the median engagement of the posts in a bucket is compared against the median engagement of that same account's posts outside it, producing one ratio per account. The reported effect is the median of those per-account ratios, with a distribution-free 95% interval around it. Account size therefore cancels out instead of being re-measured, and the unit of evidence is the account rather than the post, because posts inside one account are not independent observations of how the platform behaves.
What an effect percentage means: An effect of +10% means the typical measured account earned 10% more engagement on posts in that bucket than on its own posts outside it. The interval beside it is a 95% confidence interval on that figure; where the interval spans zero the page reports no measurable difference rather than a small effect. It is not a click-through rate and not a guarantee for any individual account.
Sample and window: 3,294 X accounts carry an engagement aggregate, built from 201,968 collected posts. Every aggregate covers a trailing 90-day window, last recomputed on 19 August 2026. Interaction counts are read at collection time and reflect that moment, not a live figure.
How uncertainty is handled: Every effect carries a distribution-free 95% confidence interval and the number of independent accounts behind it. A result is only stated as a finding when that interval stays entirely on one side of zero; where it spans zero the page says no measurable difference rather than printing a small number that would get quoted without its caveat. Accounts are the unit of evidence, not posts: several thousand posts from one account are one observation of how the platform behaves, not several thousand.
What we do not claim: These are observational differences, not a controlled experiment. A format that correlates with higher engagement here has not been shown to cause it: posts that carry a link, or an image, or a hashtag also differ in what they are about, and this data cannot separate the format from the content. A flat result means the interval spans zero, not that the true effect is exactly nothing.
Multiple comparisons: Each pass runs thousands of tests at once, so a per-test threshold would produce dozens of confident-looking results by chance alone. p-values come from an exact two-sided sign test on the per-account differences the published interval is built from, and are adjusted with the Benjamini-Hochberg procedure inside a testing family fixed in code before the numbers are seen. The share of published findings expected to be false is 0.05.
Pre-registration: The set of questions the miner may ask is a fixed table in the engine source, not something assembled while looking at results. Its fingerprint is recorded with every pass (1.0 / 9462bc6ed394b5a2), so a widened search space is visible in the record rather than silent.
What a withdrawn finding means: A published finding that fails the bar on two consecutive passes is retired rather than deleted, and one that turns significant in the opposite direction is marked reversed. Both stay on this page with the date the reading changed.

Frequently asked questions

How is a finding on PlayerSells promoted from a measurement to a published claim?: A measurement becomes a published finding only after clearing five independent rules: a Benjamini-Hochberg adjusted p-value at or below 0.05 inside its pre-registered testing family, at least 50 contributing accounts, an effect whose 95% interval stays at least 10% away from zero, agreement between two halves of a stable account split, and repetition across 3 consecutive passes. Measurements that fail are kept on record rather than discarded.
Why does PlayerSells correct for multiple comparisons on X engagement data?: Each pass tests thousands of hypotheses at once. At a 95% confidence level roughly one test in twenty crosses the line by chance, so a few thousand simultaneous tests produce over a hundred confident-looking results that are noise. Benjamini-Hochberg controls the share of published findings expected to be false at 0.05, which is a statement that can be checked; an uncorrected per-test threshold is not.
Does a finding holding for several passes mean it was independently replicated?: No, and the page never claims it. Each pass measures a 90-day window a week apart, so two consecutive passes share most of their evidence and a streak is a stability measure. The independent check is a split-half over accounts inside every pass: the same account always lands in the same half, and a finding must hold in both halves to be published.
Why does PlayerSells publish findings it has withdrawn?: Because a research feed that only ever adds is indistinguishable from one that never re-checks anything. Withdrawn and reversed findings are the visible evidence that every published claim is re-tested on every pass, and they carry the date the reading changed.

Cite this page

The aggregate findings on this page are published under CC BY 4.0. You may reuse them anywhere, including in AI-generated answers, provided the source is named. The licence covers the derived statistics only, not the underlying post text, which is not redistributed.

PlayerSells. "X research changelog: what changed and when." PlayerSells X Engagement Insights. Data measured 19 August 2026. https://playersells.com/insights/changelog

Machine-readable findings (JSON)Methodology

What is a good engagement rate on X?
Engagement rate percentiles across every large X account we measure, with the median, the quartiles and the gap between mean and median.
What time should you post on X?
Median engagement by hour of the day in UTC, all 24 hours, with the hours that measurably underperform called out.
What is the best day to post on X?
Median engagement by day of the week, including the one weekday that measurably underperforms every other.
Do links hurt reach on X?
Posts with an outbound link measured against posts without one - the largest single effect in the dataset.
Do images get more engagement on X?
Posts carrying an image or video measured against text-only posts, with video split out separately.
Does post length affect engagement on X?
Median engagement across four length bands, from under 80 characters to over 280.
Do hashtags work on X?
Median engagement by hashtag count, and what happens to the effect once the sample is large enough to see it.
What actually drives engagement on X
The hub: every measured effect in one place, plus the accounts with the strongest engagement for their size.

The full dataset index lives at https://playersells.com/insights.