Signal-by-signal hypothesis grounded in X's open-sourced heavy ranker and Meta's public statements about Threads. Your numbers are facts; claims about what the ranker is doing are hypotheses.
Highest-ROI lever HYPOTHESIS
Reply to every inbound comment within 60 minutes of posting
Mechanism · On X's open-source heavy ranker, an author reply that the original commenter re-engages with scores +75 weight — approximately 150× a like and 5.6× a plain reply; Meta has publicly stated replies are the strongest signal on Threads; responding to early comments plausibly triggers conversation chains that compound the reply-velocity signal in the first-hour engagement window, which is the period most likely to determine whether the ranker surfaces a post beyond the follower graph
Expected impact · If the account converts its median 7-reply posts into 2-3 reply chains each, the weighted conversation signal per post would plausibly increase substantially; over 30 days this could be consistent with raising the median view count from 506 toward the 1,000–2,000 range and reducing the zero-reply fraction below 10%, though exact outcomes cannot be predicted without knowing Threads's specific weights
X heavy ranker open-source weights: author reply that user re-engages with = +75; plain reply = +13.5; like = +0.5 (source: twitter/the-algorithm, GitHub, 2023); Meta/Threads blog: 'replies are the strongest signal on Threads'
Diagnosis HYPOTHESIS
Yosephgratika's account shows a dramatically bimodal distribution: two posts (19,561 and 8,734 views) account for the overwhelming majority of 30-day views, while the median post sits at only 506 views against a 369-follower base — a reach rate of 137% that is inflated by those outliers. The reply-to-like ratio of 0.71 at the account level is genuinely strong and consistent with Meta's documented preference for conversation-driving content, suggesting the ranker has at some point surfaced these posts broadly. However, 20% of posts received zero replies and 15% zero likes, and the extreme variance between floor (221 views, 0 engagement) and ceiling (19,561 views, 112 replies) is consistent with a ranker that is still assigning an uncertain or volatile quality prior to this account rather than a stable boosted one. The pattern is plausibly consistent with an account that occasionally breaks out of its distribution but has not yet established the consistent per-post engagement floor that would signal a reliably high quality prior to the ranker.
Inferred signal weights
HYPOTHESIS
Claude's estimate of how much each signal is currently dragging your ranking. 0 = no impact, 1 = dominant.
Reply velocity (first 30-60 min)
BOOSTED
Mean replies per post = 14.9; median replies = 7.0; account-level reply-to-like ratio = 0.711; top post received 112 replies on 19,561 views; another received 78 replies on 7,216 views
On X's open-source ranker, a reply carries +13.5 weight vs. +0.5 for a like (27× multiplier); Meta has publicly stated replies are the strongest signal on Threads, so this account's above-average reply rate likely contributes positively to distribution on posts that achieve early reply velocity. However, the floor posts (0 replies) likely drag the account's aggregate quality signal down.
Conversation depth (replies vs likes)
NEUTRAL
No thread-depth data is available in the dataset (no self-reply chain counts or nested reply metrics); median replies = 7.0 but we cannot distinguish whether those replies generated back-and-forth conversation chains or were single-level responses
On X's open-source ranker, a reply that the author re-engages with scores +75 vs. +13.5 for a plain reply — a ~5.6× premium for conversation chains; if Threads weights similarly (unconfirmed), deeper conversation chains would plausibly deliver outsized distribution lift, but the data here does not allow a confident classification of this signal.
Self-reply behavior (author → commenter)
NEUTRAL
No self-reply or thread-continuation data is present in the dataset; cannot determine whether posts are standalone or part of threaded chains
Self-replies that extend a post into a thread could plausibly increase dwell time and surface additional engagement opportunities, but without data this signal cannot be classified; the impact is unknown.
Zero-reply penalty loop
PENALIZED
4 of 20 posts (20%) received zero replies; 3 of 20 posts (15%) received zero likes; the zero-reply posts include a 5am TEXT_POST (468 views, 0 replies, 0 likes), a VIDEO (163 views, 0 likes, 1 reply), an IMAGE (390 views, 0 replies, 0 likes), a TEXT_POST (182 views, 0 replies, 0 likes), and two TEXT_POSTs with 221 and 221 views and 0 engagement
A zero-reply post provides no positive conversation signal to the ranker; consistent with quality-signal ranking systems, repeated zero-engagement posts plausibly lower the account's aggregate quality prior, making it harder for subsequent posts to achieve initial distribution — though Meta has not documented a specific 'zero-reply penalty loop' mechanism, the pattern is consistent with one.
Format diversity (text vs image/video)
NEUTRAL
11 TEXT_POST, 6 IMAGE, 3 VIDEO out of 20 posts; TEXT_POST includes both the highest-performing post (19,561 views, 112 replies) and several of the lowest (221 views, 0 engagement); IMAGE includes the second-highest performer (7,216 views, 78 replies) and the lowest IMAGE (390 views, 0 engagement); VIDEO posts average 252 views with minimal engagement
Format diversity exists but does not appear to be a reliable performance differentiator for this account; the data does not support classifying format as a boosted or penalized signal — content quality and timing appear to dominate over format type in this sample.
20 posts in the analyzed window; hour distribution is spread across 7am, 9am, 11am (most frequent at 4 posts), 6am, 4am, with some posts at 4am and 5am receiving near-zero engagement; no clear clustering around a single consistent posting window
Posting at 4am and 5am local time (2 posts) correlates with the lowest-engagement posts in the sample, which is plausibly consistent with reduced audience availability suppressing first-hour engagement velocity — a signal that Bluesky's Discover documentation confirms matters for surfacing, and which is likely relevant on Threads as well; however, the data sample is too small to establish cadence as a standalone penalized signal with confidence.