# Official source ledger / 官方来源台账

Checked by public, read-only HTTPS GET on **2026-09-05 UTC**. The Markdown bodies below were fetched successfully and read, not just inferred from search snippets. Discovery: [Dify documentation index](https://docs.dify.ai/llms.txt) → [English Cloud index](https://docs.dify.ai/_llms/en/cloud.md). These are current Cloud documentation pages, not a pinned Dify release, a self-hosted compatibility test or evidence of an app run. Titles/UI labels can change. No price, plan allowance or model-quality claim is adopted here.

| ID | Official page (retrieved Markdown) | Supporting passage / supported instruction |
|---|---|---|
| S1 | [Create a Ready-to-Use Knowledge Base](https://docs.dify.ai/en/cloud/use-dify/knowledge/create-knowledge/introduction.md) | “Click **Knowledge** > **Create** > **Create a ready-to-use knowledge base**”; configure chunks, indexing/retrieval, then wait for processing. |
| S2 | [Upload Local Files](https://docs.dify.ai/en/cloud/use-dify/knowledge/create-knowledge/import-text-data/readme.md) | “Select **Import from file** as the data source, then upload your files.” Data source cannot be changed later. |
| S3 | [Configure the Chunk Settings](https://docs.dify.ai/en/cloud/use-dify/knowledge/create-knowledge/chunking-and-cleaning-text.md) | Delimiters are removed; maximum length is measured in characters; General mode returns matched chunks directly; Preview is available but limited. Chunk mode cannot be changed after creation. |
| S4 | [Specify the Index Method and Retrieval Settings](https://docs.dify.ai/en/cloud/use-dify/knowledge/create-knowledge/setting-indexing-methods.md) | High Quality uses embeddings, supports vector/full-text/hybrid; Hybrid offers Weight Settings without a rerank-model API. High Quality cannot switch back to Economical. Top K / threshold behaviour depends on the retrieval mode. |
| S5 | [Test Knowledge Retrieval](https://docs.dify.ai/en/cloud/use-dify/knowledge/test-retrieval.md) | “Retrieval settings adjusted here are temporary and only apply to the current test session.” Records include test and linked-app retrieval events. |
| S6 | [Knowledge Retrieval node](https://docs.dify.ai/en/cloud/use-dify/nodes/knowledge-retrieval.md) | Query Text accepts a text variable, e.g. `userinput.query`; node outputs `result`, an array of chunks; set LLM Context to `result` and reference Context in the system instruction. Knowledge-base and node retrieval settings are two layers. |
| S7 | [LLM node](https://docs.dify.ai/en/cloud/use-dify/nodes/llm.md) | Context variables preserve source attribution. Configure a provider before use. Memory is optional and node-specific; retries/error routing are configurable. We do **not** adopt the page's implication that temperature zero guarantees determinism. |
| S8 | [If-Else node](https://docs.dify.ai/en/cloud/use-dify/nodes/ifelse.md) | “**Is empty** / **Is not empty**” value checks and selecting previous-node variables. Whether the target release accepts the result-array check must still be verified in its UI. |
| S9 | [Answer node](https://docs.dify.ai/en/cloud/use-dify/nodes/answer.md) | Chatflow response editor accepts fixed text and previous-node variables, including LLM responses; streams output. This is not a human-approval gate. |
| S10 | [App Toolkit: Citations and Attributions](https://docs.dify.ai/en/cloud/use-dify/build/additional-features.md#citations-and-attributions) | “numbered citations linking back to the original documents and chunks.” This supplies citation UI, not a guarantee of factual entailment. |
| S11 | [Model Providers](https://docs.dify.ai/en/cloud/use-dify/workspace/model-providers.md) | Integrations > Model Provider; owner/admin manages shared providers. Defaults distinguish reasoning, embedding and rerank models. Key and credit usage may coexist. No provider setup was performed. |
| S12 | [Workflow & Chatflow](https://docs.dify.ai/en/cloud/use-dify/build/workflow-chatflow.md) | Chatflows start with User Input, end with Answer, and run from user messages. Used for the walkthrough's app type; no trigger, sending tool or external integration is added. |

## Evidence boundary / 证据边界

- **Verified:** these official documentation bodies and the native offline checks listed in README.
- **Designed, not observed:** chunk settings, retrieval candidates in `questions.json`, prompts, UI wiring and expected responses. Numeric settings are exercise starting points, not validated recommendations.
- **Not run:** Dify installation, upload, indexing, embeddings, retrieval, model generation, citation display, error branches, human review in a deployed app or any outbound send.
- No DSL is supplied. Repository fixture schema and source IDs are our teaching format, **not Dify's import/API schema**. A `.json` file in this folder is not an importable Dify app.