What this category's data looks like
In the scenario of signing research project task books, the materials include project task books, subject task breakdown tables, and budget books. The project task book carries the research objectives, cycle, and signature fields; the subject task breakdown table carries the sub-project names, responsible persons, and progress milestones; the budget book carries subject amounts, quantities, and dates. The material source is the park's research management archive, and the update rhythm changes with signing batches and version revisions. Field characteristics are concentrated in names, quantities, dates, and signature receipts, and the same record needs to be compared item by item across multiple materials. When entering the vector model and index function, the project task book, subject task breakdown table, and budget book are imported as independent data rows, each generating vectors and recording index fields. During retrieval, they are merged back to the original data by vector identifier, supporting basis entries that can point back to the source text.
Configuration and implementation checks
This step is executed by the index service and vector database, outputting vectors and index status (whether recall is possible and whether data in the same group is merged). Basis entries that can point back to the source text are then completed by subsequent AI nodes, code nodes, or manual work (retrieval and citation trimming only take supporting materials).
| Object | Check | Verification action |
|---|---|---|
| Project task book | Names, quantities, dates, and signature fields must be aligned item by item with the subject task breakdown table and budget book | Open the data row to check the index fields and confirm whether vector identifiers in the same group are merged |
| Subject task breakdown table | Sub-project names, responsible persons, and progress milestones must be consistent with the project task book | Run one cross-material query and check whether the recalled content points to the same group of original data |
| Budget book | Subject amounts, quantities, and dates must match the project task book and subject task breakdown table | Check the data scope covered by the rebuild task and confirm the index status after changing models |
| Data row index fields | Multiple vector identifiers for one piece of data are all present, and the status is indexed | Check the budget of limit in the cited text, and record the array count separately |
Three common errors
- Symptom: The same record's names, quantities, and dates do not match between the project task book and the budget book. Inspect the following: The two materials were imported separately, and retrieval merging was not aligned.
- Symptom: After changing models, recall quality drops, and some data rows have content but cannot be retrieved. Inspect the following: The rebuild task is triggered only for data that has already completed indexing; data rows not covered have not completed indexing and do not participate in recall.
- Symptom: Signatures or receipts are missing, and basis entries cannot point back to the source text. Inspect the following: Versions are not synchronized, current materials and archived materials are mixed.
How to verify the setup
This function's acceptance: sample one data row and check the index fields to confirm that multiple vector identifiers are all present and the status is indexed; then run one retrieval and see whether the same group of original data is merged, with the vector score taking the highest score.
Business test: ask a cross-material basis question, see whether the answer gives a source that can be pointed back to, and check the citation count and the tokens budget of the cited text, recording the array count separately.
- In the knowledge base settings, check the index model, confirm it is consistent with the vector model actually used by the knowledge base, and check the data scope covered by the rebuild task.
- Open one data row and check the index fields, confirm the budget of
limitin the cited text, and record the array count separately.