What the documents look like
For staff training in Amazon FBM fulfillment, this page examines FastGPT citation tracing. Separate role cards, practice replies and carrier events in delivery-service exercises. The examples are hypothetical.
Constraints at the citation tracing step
A proposed FBM service lesson has a role card for an absent customer concerned about rain, a script asking whether another location is convenient and guidance to inspect available options first. A carrier sample records an attempted contact without a customer reply. The learner asks whether to announce agreement to doorstep placement. Concern is not consent, a question is not an answer and attempted contact establishes no completed conversation. Preserve fictional roles and event stages instead of writing an imagined customer preference as an actual decision or treating guidance as an existing arrangement.
Configuration and implementation checks
| Training-source check | Teaching identity and evidence scope | Acceptance action |
|---|---|---|
| Role preference | Concern about rain | Do not invent consent. |
| Sentence role | Question script | Do not make it customer reply. |
| Contact stage | Contact attempted | Do not infer agreed location. |
| Speaker | Separate agent and customer | Check actual attribution. |
| Added practice reply | Same exercise and question | Retain hypothetical status. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Associated knowledge bases | Knowledge bases associated with one search node must use the same embedding model. |
Citation output | The output is an array; an empty array can continue. |
Citation input handoff | Inspect citations actually received by the answer step. |
Citation display check | Trace displayed sources to their originals. |
Document-validity check | Inspect source identity and applicable version separately. |
Three common errors
First, concern about rain may become consent to doorstep placement; inspect the card and any explicit choice instead of answering for the customer. Second, the question script may become a customer reply; check speaker and sentence role while retaining the unanswered state. Third, attempted contact may become an agreed new location; compare the carrier event with actual subsequent records. These require preference evidence, dialogue attribution and event progress. Their common service topic cannot merge the separate actions into one completed decision merely because all appear in the training packet.
How to verify the setup
Ask what the role expresses, what the script asks and what the carrier event records, inspecting speaker information in actual references. The scenario alone supplies no customer decision. Attempted contact establishes neither a connected call nor failed contact. Added practice replies need the same exercise and answered question matched and remain hypothetical. Empty references require scope and transfer investigation. The lesson changes no delivery instruction, proves no doorstep placement and measures no actual service effect. Real choice and delivery still require their own applicable operational evidence beyond the classroom scenario.