What the documents look like
For product questions in Amazon FBM fulfillment, this page examines FastGPT retrieval. Compare the material received with the sources used in the answer. The examples are hypothetical.
Constraints at the retrieval step
A new FBM picking note says, “Left-side accessory; installation tools excluded.” Item and tool questions need the position wording and exclusion respectively. Matching a tool noun alone does not establish inclusion. Compare retrieval modes with the literal accessory term and the full contents question, inspecting whether selected sources contain that statement. Left-side is a hypothetical item description, not a software setting or support for universal installation.
Configuration and implementation checks
| Documented item or implementation check | Material and condition | Evidence action |
|---|---|---|
| Source scope (implementation check) | Inspect picking note and confirmation separately | Accessory pack is not an itemized tool list. |
| Search modes: semantic, full-text, hybrid | Left-side wording and tool-inclusion question | Inspect retrieval of the actual exclusion. |
| Citation token limit | Installation tools excluded | Do not lose the exclusion in citations. |
| Minimum relevance | Inspect filtering under semantic search or reranking | High relevance does not establish interchangeable positions. |
| Question optimization | Right-side installation follow-up | Inspect any position change introduced by optimization. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Associated knowledge bases | Knowledge bases associated with one search node must use the same embedding model. |
Search mode | Semantic, full-text and hybrid modes are available. |
Citation limit | Citation limits are measured in tokens. |
Minimum relevance | Range 0–1; active for semantic search or reranking. |
Question optimization | Resolves references and expands questions. |
Three common errors
A separate proposed shipment confirmation says only “accessory pack,” without a tool list. Inferring a complete kit from pack lacks contents evidence; preserve that candidate beside the picking statement. Also check whether token trimming removes the exclusion, allowing repeated broad labels to obscure it. Keep independent source identities. Whether actual delivery contained tools needs corresponding records, beyond a product category appearing in search results. Compare a proposed erroneous output, “tools complete,” with the original exclusion. Inspect whether retrieval supplied only a pack label or citations dropped the qualifying sentence. Retain original, candidates and answer to distinguish a condition missing from recall from an unsupported expansion during generation.
How to verify the setup
A right-side installation follow-up has no compatibility support in the left-side description. If optimization makes left and right interchangeable, it introduces an unsupplied condition; compare the full original and rewritten questions. After obtaining actual fit material, repeat the question and trace added candidates and citations, recording supplementation separately from settings changes. Check minimum relevance only under its conditions. Similar names, earlier positions or a longer answer do not fill the gap in right-side compatibility evidence.
| Material context | Scope |
|---|---|
| Document types | Order Picking List, Shipment Confirmation, Return Address Settings and Logistics Tracking Sheet |
| Candidate sources | Amazon Seller Central, Carrier System, ERP Orders and Customer Service Email |
| Technical baseline | FastGPT v4.17.2 |