Citation Source and Traceability for Airport Financial

What This Category of Data Looks Like

Airport financial report data primarily comes from official annual reports and quarterly operation briefings of airport groups, as well as publicly available statistical materials from civil aviation authorities. The update schedule is one full annual financial report released annually, and quarterly operation data updated 1 to 2 months after the end of each quarter. The document structure includes fields such as takeoff and landing cycles, passenger throughput, cargo and mail throughput, non-aeronautical business revenue, and unit operating costs. Units are mostly cycles, person-times, tons, and ten thousand yuan. A single complete financial report document is usually dozens of pages long, with attached monthly detailed operation breakdown attachments.

Constraints on Citation Source and Traceability from These Data Characteristics

Scattered data sources require clear field mapping rules for different data sources, to avoid mixing up the data calibers of official financial reports and civil aviation authority statistics. Differences in the update schedules of quarterly and annual data require the traceability link to distinguish the recall scope across different time windows, to ensure that the timeliness of cited data matches analysis needs. The long length of single documents and attached detailed attachments require reasonable segmentation rules to be configured, to avoid damaging the integrity of core fields such as takeoff and landing cycles and throughput during splitting, while ensuring that detailed data in attachments can be accurately traced.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`Recall count`	`top 8-12 entries`	Core fields of airport financial reports are concentrated. Excessive recall will increase traceability noise, and financial report data calibers are fixed, so no redundant results are needed
`Similarity threshold`	`0.65-0.75`	Financial report data mostly consists of structured numbers and fixed terminology. A threshold that is too low will introduce irrelevant industry general expressions, while a threshold that is too high may miss detailed operation data
`Chunk size`	`800-1200 characters`	Airport financial reports include long sections of operation descriptions and detailed tables. This length can retain the complete context of a single piece of business data, avoiding field breakage after splitting
`PARSE_FILE_TIMEOUT_SECONDS`	`120 seconds`	Single financial report documents usually have a large number of pages, so sufficient time is required for parsing to complete text splitting and metadata extraction
`Citation source display format`	`Display file name + page number + field name`	Financial report traceability requires clear corresponding documents, page numbers, and specific data fields to facilitate auditing and verification
`Rerank result count`	`top 5-8 entries`	After sorting by relevance for core financial report data, the top 5-8 entries can cover all core indicators required for analysis

The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis, and it is recommended to test on your own samples before finalizing settings.

Three Common Misconfigurations

The phenomenon is that after adjusting the Similarity threshold to the minimum and Recall count to the maximum, the number of recall results does not change and remains fixed. The cause is that the 数据源范围 is not configured to all uploaded airport financial report documents, and the system only recalls files in the specified folder by default.
The phenomenon is that the output content contains original Markdown syntax and is not rendered into rich text format. The cause is that the 输出内容格式化 parameter is not enabled, or the prompt does not explicitly require converting the result into a renderable display format, only generating syntax text.
The phenomenon is that the sourceid field in the traceability result cannot be associated with the file metadata of the corresponding file. The cause is that the 提取文件元数据 parameter was not enabled when uploading the document, or the 元数据映射规则 was not configured to bind the sourceid to the file storage field.

How to Verify Proper Configuration

Upload a single test airport financial report document, initiate a query containing clear operation indicators, and check whether the returned results are accompanied by traceability information of the corresponding document.
Adjust the Similarity threshold and Recall count parameters, repeat the same query, and observe whether the number of recall results changes correspondingly with the parameter adjustments.
View the traceability details panel, confirm that the sourceid field of each returned result can be associated with the storage metadata of the corresponding file.
Check the format of the output content, confirm that Markdown syntax has been correctly processed into a readable display form.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.