Creator sales campaigns: vector indexes for product questions

Review vector indexes for product questions in Creator sales campaigns. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

Creator-led sales sessions depend on external creator fulfilment and commission settlement rather than internal brand staffing. The working materials for this task are creator cooperation agreements, commission setup sheets, sample shipment forms, and livestream performance reports. Each carries a different role in product material and selection questions. Agreements and commission sheets carry the parties, the object of cooperation, and the fee basis. Sample shipment forms record logistics and receipt status. Livestream reports provide per-session results. Data sources include the selected alliance, the creator marketplace platform, creator history, and store transaction detail, with updates per session or per week. Update cadence does not change what this page covers: the vector model and index only handle the mapping between stored text and its indexes.

Constraints at the vector indexes step

Knowledge bases associated for retrieval must use the same vector model. Check the actual model and input type first. Bases built on different models cannot be associated together; verify the identifier and exclude those that do not match. One source data block can be linked to multiple indexes, and the index content is checked separately from the original text. Index content must be complete and focused, and retrieval coverage against index length is weighed against the sample at hand. Image retrieval also depends on whether an image description or image vector index was generated at ingestion time, not only on what the query side sends. After a model or index change, compare cited source text using the same set of questions; do not promise results based on vector scores alone. Semantic recall and relevance scores do not handle permission or entity isolation, and the same-model requirement for associated knowledge bases does not extend to business registry sites or other external systems. Minimum relevance is a filter threshold, not a final ranking rule. RRF is a fusion process for mixed results, not a toggle.

Hypothetical paper exercise sample: the source text to be checked reads, "Party A commissions Party B to promote Product A in a dedicated session on 18 June 2025, with commission settled at fifteen percent of transaction value; Party B must return the sample receipt form by 20 June." A pending handover request reads, "Please confirm that commission settlement for this creator session is complete and authorize payment against the livestream report amount." The source text states only the commission, the commission rate, and the return deadline. It does not state that settlement is complete or that payment is authorized. This request is a matter to be verified and cannot be treated as approved.

Configuration and implementation checks

Configuration itemSetting or check for this pageRelation to the sample
Knowledge base associationSelect one or more knowledge bases using the same vector modelCheck the model identifier before ingesting the sample material
Index and source dataOne source data block can link to multiple indexes; check each separatelyThe commission clause and the return deadline can each have their own index
Index contentBuild indexes from complete and focused informationKeep the two sample sentences as separate indexes, preserving parties and dates
Image indexDepends on whether an image description or image vector was generated at ingestionIf the sample shipment form contains images, check the ingestion index first
Search modeSemantic search, full-text search, or hybrid searchExact items such as the commission rate suit full-text or hybrid search
Result reviewCompare cited source text using the same set of questionsUse the sample source sentences to check whether citations match

This page does not specify any default value. If a starting point is needed for this hypothetical exercise, minimum relevance may take a clearly labelled hypothetical value, chosen to make it easier to observe the change in the number of cited items before and after filtering, compared against the cited source text for the same set of questions before and after adjustment. That value is not an official default.

Three common errors

  1. Treating a retrieval hit as authorized execution. In the sample, the handover request asks to confirm completed settlement and authorize payment. Retrieving the commission clause only shows that the clause exists in the source text; it does not show that settlement is complete or that payment is approved. The check is to input that request, compare the status words in the request against what the source text states, and observe whether the source text contains wording about completed settlement or payment authorization. If it does not, record the item as pending verification.
  2. Associating knowledge bases that use different vector models. If the sample materials were ingested into knowledge bases with different models, associating them together does not produce comparable results. The check is to verify the model identifier of each knowledge base, compare whether the identifiers match, and observe whether any mismatch exists. If one does, exclude that base and record it. Do not run a cross-model retrieval to compare missed recalls.
  3. Treating an image query as proof that the image was ingested. If the sample shipment form exists as an image, uploading an image on the query side does not show that an image description or image vector was generated at ingestion. The check is to verify the image index field in the ingestion record, compare the query input type against the index types already built, and observe whether a corresponding index exists. If it does not, record an index gap.

How to verify the setup

  1. For the sample source sentence "Party A commissions Party B to promote Product A in a dedicated session on 18 June 2025, with commission settled at fifteen percent of transaction value," input the question "What is the commission rate for this session?" Compare whether the cited source text contains fifteen percent and the corresponding parties. Observe whether the citation matches the source text. A match satisfies the condition.
  2. For the pending handover request "Please confirm that commission settlement for this creator session is complete and authorize payment against the livestream report amount," input that request. Compare the completed settlement and payment authorization in the request against what the source text states. Observe whether the source text states that status. If it does not, record the item as unsupported rather than as a denial or a conflict.
  3. For empty-result handling, use the explicit paper input [] to check the downstream design. Compare whether the output array length can be zero. Observe whether the output path completes. Do not claim that a fabricated rule retrieval necessarily returns []. All of the above are paper checks; no retrieval or deployment was actually performed.