Turtle WoW knowledge base

An archive and analysis of turtlecraft.gg (official website), forum.turtlecraft.gg (phpBB forums, English and German) and the community wiki turtle-wow.fandom.com, captured on 2026-09-10, five weeks before the announced closure of the website and forums on 2026-10-16. Turtle WoW's game servers went offline on 2026-05-15; the final content patch was 1.18.1 "Nightmares of Ursol" (released 2026-03-20). It also folds in the archived Turtle WoW Europe Discord (see Provenance and limits), which supplies the deep class, PvP, twinking and addon knowledge the forums lacked.

Purpose: (1) a reference for a local Turtle WoW 1.18.1 server (tortoise-wow core) to find and fix discrepancies; (2) source material for human-like bots that play and talk like real Turtle WoW players.

Start here: the class playbooks

Players and bots should start with guide/, the per-class playbooks for Turtle WoW 1.18.1, published on the site (https://stevemcqueenz.github.io/turtle-knowledge-base-/):

• guide/classes/<class>/: for each class, an index.md overview (spec viability for raid, dungeon, PvP, leveling and farming; 1.18.1 changes; races and professions), plus leveling.md (a validated talent order for every level from 10 to 60, the pull loop, gear and dungeons by bracket), one page per endgame spec (talent build with a working calculator link, stats and caps, single-target and AoE priority lists, cooldowns, gear, enchants, imbues/oils/poisons, consumables, raid notes and common mistakes), pvp.md and sources.md (the experts relied on, and the gaps).
• guide/instances/: 27 dungeon pages and 12 raid pages (Turtle custom and vanilla), each with boss-by-boss role duties, resistances and immunities.
• The bot playbooks in structured/classes/<class>/*.yaml match the guide. Each has a guide: key, talent points with spell_ids, a validated leveling_order, and rotation priority lists.

The guide is written mainly from a full read of the archived Discord for the 1.18.1 era (every message from 2025-10-03 on, plus highlights from 2024-11 to 2025-10), combined with the forum and wiki archive below. Every non-obvious claim is cited, and a QA pass checked each Discord citation against the claim it supports. How it was built is described in tools/discord-pipeline/.

Layout

directory	content
guide/	the class playbooks and dungeon/raid pages (see "Start here"); the source of the site's class, playbook, leveling, sources and instance pages
research/discord-notes/	cited reader notes, one file per Discord chunk (122 chunks, 10 channels) under fixed topic headings: the raw material the guide was written from. Verify a note's citation before reusing it
tools/discord-pipeline/	the chunker, the reader, writer, QA and instance briefs, and the validation tools (talent_tool.py, check_notes.py, cite_pairs.py, collect_evidence.py, ...)
site/	the React class-guide site (GitHub Pages); site/scripts/build-data.py builds its data from guide/, structured/ and synthesis/
recon/	Phase 0 site map, forum index with sizes, robots findings
scrape/	crawler (crawl.py), extractors (extract.py, extract_site.py, extract_wiki.py), entity index builder, validator, coverage report, agent prompts, and the crawl manifest manifest.jsonl (every URL with status, agent and timestamp)
raw/	original HTML as fetched, gzipped: raw/forum/f<forum>/t<topic>-p<page>.html.gz, raw/forum/f<forum>/list-<start>.html.gz, raw/site/<slug>.html.gz; raw/wiki/pages/<pageid>.json (wikitext via the MediaWiki API)
extracted/	clean Markdown: one file per forum topic (all pages merged, per-post metadata, [STAFF] tags, era markers), one per site page, one per wiki page
structured/	machine-readable data: forum/posts/*.jsonl (one record per post), forum/topics.jsonl, forum/authors.json (authority levels), forum/timeline.json, external-links.jsonl, talent-links.jsonl, patches/ (1.18.1 changes, Turtle-vs-vanilla, timeline), classes/<class>/*.yaml (bot playbooks), content/ (zones, quests, factions, vendors, drop rates, recipes, instances), wiki/ (NPC/quest/zone/faction records), entity-index/, issues.yaml, staff-statements.jsonl, glossary.jsonl, contradictions.yaml, plus the Discord-mined twinking.yaml, level-one-lunatic.yaml, addons.yaml, macros.yaml, pvp.yaml
structured/talents/	decoded 1.18.1 talent trees, one <class>.yaml per class (tabs, talents, rank spell_ids, req/req_rank), and builds-<class>.yaml: 1,719 distinct 51-point builds decoded from forum and Discord build codes and validated against those trees, each allocation carrying talent_id and rank spell_id; pre-1.18.1 codes that no longer fit the 1.18.1 rules are kept in builds_with_tree_drift
structured/discord/	the Discord citation resolver: evidence-<channel>.jsonl, one file per channel holding the verbatim cited messages (id, ts, author, content, attachments, reactions, pinned), plus README.md; every [[d:<channel>#<id>]] in the KB must resolve to a record here. The collector that rebuilds these files, pipeline/collect_evidence.py, lives outside this repo
synthesis/	analysis documents with a source link for every claim: 1.18.1-master-changelog.md, turtle-vs-vanilla-timeline.md, classes/ (spec-role matrix, per spec-role playbooks, leveling guides), content/ (zones, questlines, factions, NPCs/vendors, drop rates, professions, dungeons and raids with per-instance files), staff-statements.md, issues-summary.md, glossary.md, contradictions.md, coverage-report.md, CONVENTIONS.md, and the Discord-sourced twinking/, challenges/level-one-lunatic.md, guides/addons-and-macros.md, pvp/matchups-cross-class.md
behavior/	player-behavior corpus for the bot project, usernames pseudonymized; popular-addons and pvp-culture now carry Discord (#code-corner, #pvp-general) sections alongside the forum corpus

Conventions

• Every claim in synthesis/, structured/ and behavior/ cites a post URL (viewtopic.php?p=<id>#p<id>), a site page, a wiki page, or a Discord message ([[d:<channel>#<id>]]). python3 scrape/validate.py checks that every cited post and Discord message exists in the archive and that all data files parse.
• Authority: staff = rank "Turtle WoW Team" (21 accounts; green username), otherwise player; guide writers are called out where relevant. Staff statements about intended behavior outrank player claims.
• Time: all dates UTC. Eras: pre-1.18.1 (before 2025-10-03), 1.18.1-announced-pre-release (2025-10-03 to 2026-03-19), post-1.18.1 (from 2026-03-20). See structured/forum/timeline.json.
• Discord claims cite the archived message as [[d:<channel>#<message_id>]] (e.g. [[d:mage#818530154769940561]]), where <channel> is the normalized channel name and <message_id> the Discord snowflake. python3 scrape/validate.py resolves every such citation against structured/discord/evidence-<channel>.jsonl, which reproduces each cited message verbatim.
• Agreement: consensus / single source / contested, marked per recommendation.
• Nothing is invented: gaps are listed as gaps. In-game IDs appear only where a source states them (the wiki gives quest IDs for 182 quests; the forum almost never gives IDs; the item/spell database subdomain was already offline).

Provenance and limits

• Crawl: one fetcher at about one request per second, descriptive user agent, robots.txt honored, no login, nothing gated fetched. 22,449 topic pages, 624 listing pages, 34 site pages; 21,301 of 21,308 English+German topics (4 deleted topics returned 404). Russian, Chinese, Spanish and Ukrainian forums were not crawled by decision.
• Discord: the archived Turtle WoW Europe server export (category "Achived Soon"), 2021-02 to 2026-05, 3,004,388 messages across 14 channels — #warrior, #paladin, #hunter, #rogue, #priest, #shaman, #mage, #warlock, #druid, #code-corner, #pvp-general, #turtle-wiki, #twinking, #level-one-lunatic; the export's theorycrafting channel was deliberately not fetched (the message count is a provenance claim from the export manifest; the norm/ export lives outside this repo and is not independently checkable from within it). It is a player source with no green-name/staff signal: staff and recognised guide writers are inferred from post content and reception and cross-checked against structured/forum/authors.json where a name matches.
• Not available: database.turtlecraft.gg (DNS gone), talents.turtlecraft.gg (origin down; build links are cataloged undecoded in structured/talent-links.jsonl), the images under turtlecraft.gg/images/ (all 404), the site pages /hardcore and /bug-tracker (server errors), the forum staff list (login-gated).
• Wiki pages were fetched through the public MediaWiki API because the HTML front is behind a Cloudflare challenge; the wiki is a community source and is marked as such everywhere it is used.

Regenerating

python3 scrape/extract.py --rebuild      # forum topics -> extracted/ and structured/forum
python3 scrape/extract_site.py           # site pages
python3 scrape/extract_wiki.py           # wiki
python3 scrape/build_entity_index.py     # entity index
python3 scrape/coverage.py               # coverage report + spot-check
python3 scrape/validate.py               # citations and data files

• scrape/prompts/discord-local-agent.md: the full brief for extending the knowledge base from the Discord server backup (ingester in scrape/discord/, talent-tree tooling in scrape/talents/).
• scrape/discord/SURVEY.md: what the backup actually holds — 14 channels (the nine class channels, pvp-general, code-corner, turtle-wiki, twinking, level-one-lunatic), 3,004,388 messages, 73,446 images, guild 466622455805378571 — which staff roles exist in it (only Moderator, 469 messages), the size policy, and which domains the backup cannot support.

Discord evidence files are rebuilt by pipeline/collect_evidence.py, which lives one directory outside this repo; it resolves every [[d:<channel>#<id>]] citation in the KB against the normalized norm/<channel>/messages.jsonl export and rewrites structured/discord/evidence-<channel>.jsonl. See structured/discord/README.md.