{ "generated_by": "websites/x-lite/scripts/build_real_world_seed.py", "generated_at": "2026-11-13T18:34:56.936502+00:00", "base_seed": "data/seed.json", "real_world_sources": [ { "dataset": "Roronotalt/bluesky-ten-million", "license": "mit", "url": "https://huggingface.co/datasets/Roronotalt/bluesky-ten-million", "fields_used": [ "type", "text", "created_at", "author", "author_did", "uri", "langs", "reply_to" ], "rows_requested": 2000, "weighted_by": "ranked pinned roots by imported descendant count", "root_uri_file": "data/bluesky_srs_2000_roots.txt", "source": "tree cache", "tree_cache_file": "data/bluesky_srs_2000_trees.jsonl.gz", "rows_loaded": 55444, "roots_loaded": 2000, "selection": "pinned ranked roots from compact tree cache", "roots_imported": 2000, "tweets_imported": 55444, "authors_imported": 25976, "max_tree_child_count": 1185, "total_tree_child_count": 53444, "explore_roots": { "source": "jsonl cache", "cache_file": "data/bluesky_explore_roots.jsonl.gz", "rows_loaded": 150, "roots_imported": 134, "authors_imported": 129 } }, { "dataset": "CardiffNLP tweet_topic_multi", "license": "other", "url": "https://huggingface.co/datasets/cardiffnlp/tweet_topic_multi", "fields_used": [ "text", "date", "label_name", "id" ], "split": "train_random", "rows_requested": 0, "rows_loaded": 0, "rows_imported": 0, "source": "skipped" }, { "dataset": "Zenodo 17677813: X(Tweeter)", "license": "cc-by-4.0", "url": "https://zenodo.org/records/17677813", "fields_used": [ "id_str", "conversation_id_str", "user_id_str", "created_at", "full_text", "favorite_count", "retweet_count", "reply_count", "quote_count", "category" ], "rows_requested": 300, "rows_imported": 0 }, { "dataset": "ReTweet: How Will Your Tweet Be Received?", "license": "kaggle dataset, see source", "url": "https://www.kaggle.com/soroosharasteh/retweet/", "code_url": "https://github.com/tayebiarasteh/retweet", "fields_used": [ "source tweet text", "reply text", "optional sentiment label" ], "rows_requested": 0, "rows_loaded": 0, "rows_imported": 0, "source": "skipped" }, { "dataset": "The Structure of Toxic Conversations on Twitter", "license": "cc0-1.0", "url": "https://doi.org/10.7910/DVN/TRWPVB", "fields_used": [ "optional direct tree metrics CSV URL for reply-count distribution" ], "rows_requested": 0, "rows_loaded": 0, "source": "skipped" }, { "dataset": "X-lite local generated username pool", "license": "generated", "url": "", "fields_used": [ "adjective", "noun", "number" ], "rows_requested": 5000 }, { "dataset": "MiDe22: Annotated Multi-Event Tweet Dataset", "license": "mit", "url": "https://github.com/metunlp/MiDe22", "fields_used": [ "topic", "event_id", "label", "tweet_id", "text" ], "rows_requested": 600, "include_turkish": true, "rows_imported": 0 }, { "dataset": "Mendeley t5kcr3nkfc/1: Twitter/X Dataset on Public Conversation Around COP30", "license": "cc-by-sa-4.0", "url": "https://data.mendeley.com/datasets/t5kcr3nkfc/1", "fields_used": [ "optional direct CSV URL only" ], "rows_requested": 0 } ], "tagged_accounts": { "accounts_created": 520, "tag_mappings": 801, "display_placeholders_replaced": 0, "generic_placeholders_replaced": 0, "raw_mentions_normalized": 41 }, "interaction_distribution": { "profile_count": 1, "favorite_count": { "min": 0, "median": 0, "p90": 0, "max": 0 }, "retweet_count": { "min": 0, "median": 0, "p90": 0, "max": 0 }, "reply_count": { "min": 0, "median": 0, "p90": 0, "max": 0 }, "quote_count": { "min": 0, "median": 0, "p90": 0, "max": 0 } }, "notes": [ "The canonical synthetic X-lite task seed is preserved when X_LITE_PRUNE_ORIGINAL_SEED=0; otherwise it is pruned to task-relevant tweets plus a small recent-post sample for followed accounts.", "Every imported source user is inserted as an X-lite user before imported posts are added.", "Imported user IDs are preserved internally; usernames/display names are assigned from a seeded external username source.", "Tagged people in imported text are normalized to @username handles and inserted as lightweight X-lite accounts.", "Bluesky root posts are selected from a small pinned URI file containing the previously ranked top root trees.", "Bluesky replies keep their original immediate parent where that parent is part of the sampled tree; full selected trees are imported by default.", "Root post source_reply_count is updated to the total number of imported descendant comments, and root views are sampled from that tree size.", "ReTweet rows, when available, provide real first-order reply text instead of generated reply templates.", "Likes, reposts, and saves are bounded deterministic samples from high-variance normal multipliers times root descendant reply count; generated replies are disabled unless explicitly requested." ] }