What the documents look like
For product questions in Shopify fulfillment, this page examines FastGPT document chunking. Compare the material received with the sources used in the answer. The examples are hypothetical.
Constraints at the document chunking step
A new proposed independent-store order says, “Ceramic cup, capacity class A; the gift box excludes a lid.” Parsing needs to preserve the product attribute and the package exclusion, beyond the isolated nouns “gift box” and “lid.” Compare original and parsed text, then ask separately about capacity and included contents. Check whether the citation drops the exclusion. Class A is a proposed label, without a supplied capacity value or an actual product specification.
Configuration and implementation checks
| Item or implementation check | Input or condition | Traceable evidence |
|---|---|---|
| Parsed output | Cup attribute and package exclusion | Do not turn a capacity label into a numerical value. |
| Chunk completeness | Gift box excludes a lid | Keep the exclusion with its object. |
| Original data and indexes | Separate order and shipping label | A set name does not supply an accessory list. |
| Sample retrieval | Capacity and included-content questions | Trace their actual supporting wording separately. |
| Post-update review | Product-information attachment absent | Inspect parsing and indexes after supplementation. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Original-document check | Compare original files with parsed output. |
Chunk completeness | Keep conditions traceable to their subjects. |
Data and indexes | Retain original-data and index evidence separately. |
Sample retrieval | Retest literal wording and natural questions. |
Post-update check | Inspect training or index state after updates. |
Three common errors
A separately proposed shipping label says only “cup gift-box set,” without listing its contents. Parsing that name must not supply accessories absent from the order wording. Inspect independent original data and indexes so a search summary does not combine the set label with a lid into a new statement. If only the shipping label is retrieved, return to the order’s exclusion before deciding whether the source text says a lid is included. Distinguish the set name from an itemized contents list, which the current label does not supply. A shared reference to the cup still requires returning to the order exclusion, without using the common noun to fill in package contents.
How to verify the setup
The order note refers to product-information material that was not supplied, leaving detailed material and use conditions without sources. After adding it, inspect original text, parsed output and processing or index state separately, then repeat the question to trace the added content. Preserve a failed sample where parsing omits a condition present in the original. Material never supplied remains missing; an upload or a matching product name does not establish searchable evidence for every attribute.
| Material context | Scope |
|---|---|
| Document types | Order Details, Shipping Rate Template, Shipping Label and Refund Records |
| Candidate sources | Shopify Admin, Payment Gateway, Carrier System and Inventory Location Data |
| Technical baseline | FastGPT v4.17.2 |