Learning Loop keeps Markdown and Properties as the only source of truth. It has no AI client, prompt history, chat storage, proprietary learning database, or dependency on another community plugin.
The initialization command creates missing folders only:
00-Inbox/
10-Topics/
20-Nodes/
30-Sources/{Papers,Books,Documents}/
40-Records/{Experiments,Changes,Incidents}/
50-Assets/{Servers,Services,Databases}/
60-Cards/
70-Maps/
80-Daily/
90-Templates/
99-Attachments/
Every managed note has ll_id, ll_type, ll_title, and ll_created.
Relationships use stable IDs in ll_topic, ll_parent, and ll_related.
Ordering, current-node state, confidence, verification, mastery, and review
selection are ordinary Properties. The plugin modifies Properties only through
FileManager.processFrontMatter().
Create topic from Markdown outline parses headings and task lists while ignoring fenced examples. It creates one stable Markdown node per item, preserves hierarchy and task completion, selects the first incomplete node, and writes a linked learning tree into the topic.
Node commands create inline questions, promote a question, move to another parent or the topic root, reorder siblings, split selected material, merge without deleting the source note, append timestamped corrections, select the current node, set confidence, mark verification, add to review, and mark mastery. Cycle checks prevent a node from becoming its own descendant.
Each node contains:
当前理解
为什么
边界条件
示例
实验或证据
容易混淆
待探索
相关知识
来源
修正记录
English-term notes include technical and cross-domain meanings, original and cloze sentences, examples, a user-authored sentence, and confusing words. A new term also creates an English review card.
Paper notes include metadata, research question, method, experiments, results, author claims, limitations, original text, translation, user understanding, questions, and distilled nodes. A paper or incident can be distilled into a node while retaining a source link.
Operations commands create server, service, database, change, incident, and runbook notes. Their templates include prechecks, rollback, health or recovery verification, and post-incident knowledge extraction as appropriate. A conservative scanner rejects likely private-key blocks, bearer credentials, access keys, and password/token assignments. Markdown should contain only a password-manager or SecretStorage reference.
Cards support exactly these types:
定义 原理 比较 步骤 诊断 代码预测 英文术语 论文结论
The only grades are 不会, 模糊, and 掌握. The deterministic scheduler
stores ll_due, ll_interval, ll_ease, ll_repetitions,
ll_last_reviewed, and ll_last_grade in the card's Properties. An interval
is capped at ten years and ease remains between 1.3 and 3.0. The daily
dashboard is a normal Markdown file with the current node and all due-card
links.
For each topic, the generator creates deterministic structure, relation, mobile-path, and focused Canvas files from node Properties. Current nodes are highlighted, ordering is stable, duplicate relation edges are removed, and large desktop maps collapse after a bounded node count with a visible placeholder. The mobile path contains ancestors, siblings, and children around the current node.
Thinking.canvas is created once and never overwritten. Automatic map names
are the only files the generator updates. Deleting an automatic Canvas triggers
reconstruction from Markdown; deleting or editing the manual thinking map does
not.
The single-column mobile panel exposes today's review, continue current node, quick question, record term, append one understanding, current topic path, runbook, unlock, foreground sync, and lock. These are lightweight views over the same Markdown data used on desktop.