What this category of data looks like
Declaration materials and file review data for shipbuilding mainly originate from submission documents at each stage of ship construction, classification society inspection files, production process documents, and material procurement records. The data is updated in phases according to construction milestones, covering core stages such as block construction, pre-launch submission, and sea trial acceptance.
Each individual file includes fields such as ship identification number (IMO number), construction batch number, ship type parameters, submission date, and review opinions. Some attached documents like process specifications and inspection reports can reach tens of thousands of characters. Units involved include deadweight tons, ship length in meters, construction cycle in days, and other exclusive identifiers.
What constraints these characteristics impose on the "citation sources and traceability" link
Shipbuilding file data is updated by construction milestone, and includes exclusive ship identification numbers and batch numbers. Traceability must accurately associate materials corresponding to the relevant construction stage to avoid cross-batch confusion.
The length of individual documents varies greatly. The token usage of long documents far exceeds that of general scenarios. It is necessary to strictly control context length to avoid overflow.
Exclusive parameters in fields such as IMO number and deadweight tonnage must be used as core matching items for retrieval and traceability. Without these, original files cannot be accurately located.
The phased update feature also requires knowledge base synchronization to be tied to construction milestones. This ensures that cited materials are always valid for the current submission stage.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum Context | 8000–12000 token | Covers multiple associated files for a single batch submission, avoids truncating core review content |
quoteMaxToken | 15000–20000 token, calibrated via actual testing | Adapts to the token usage of long shipbuilding files, ensures complete display of review opinions and process parameters |
responseData | Includes ship identification number, submission stage, review conclusion | Matches core fields of files, provides clear positioning basis for traceability |
| Single Search Limit | Top 8–12 retrieval results | Corresponds to the number of associated materials for a single batch submission, avoids interference from irrelevant information in AI generation |
detail | Enable full content display | Ensures that review details of original files can be fully traced |
Three common mistakes
- Phenomenon: The number of model context entries displayed in the interface does not match the number of citations actually sent to the AI. Cause: The matching relationship between
quoteMaxTokenand knowledge base chunk length is not correctly configured, resulting in truncated long document chunks not being correctly counted in context statistics. - Phenomenon: Knowledge base chunks exceeding the set citation limit are still cited. Cause: Retrieval filtering is not performed based on the construction batch field of shipbuilding files, causing irrelevant cross-batch materials to be recalled and included in citation statistics.
- Phenomenon: Retrieval results passed via HTTP requests cannot be correctly associated for traceability by the AI. Cause: Core identification fields such as ship identification number and submission stage are not included in the
responseDataconfiguration, resulting in missing traceability information.
How to confirm the configuration is correct
- View the knowledge base retrieval preview interface, confirm that the returned results include core traceability fields such as ship identification number and submission stage.
- Trigger an AI conversation and open the context popup, check that the displayed citation sources correspond one-to-one with actual file numbers.
- Adjust the value of
quoteMaxToken, verify that the truncation logic of retrieval results matches the length characteristics of individual files. - Simulate a workflow to pass retrieval results, confirm that the traceability data format carried in the HTTP request meets the receiving requirements of AI conversations.
The values provided here are common starting points and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.