Flux ArtEstablish which image actually generated the reported results
After exporting several candidates from Flux Art, version A in the design file, final in the handoff folder and the asset name in the advertising interface may not refer to the same content. If someone changes the image midway but retains its name, the report still contains data, yet it may not be attributable to the current image. Before discussing performance, identify the version actually used and its corresponding period. This article provides an external team mapping table and a mismatch-investigation method, not a platform feature that automatically synchronizes advertising accounts, tracks conversions or selects experiment winners. No real advertising test was performed.
Hand over an approved version, not an overwritable name
At design handoff, retain the original candidate, the actual approved file, a preview and a version number. Record the product SKU, the hypothesis for this round and the visual change being compared. If later cropping, text additions or compression create different delivery files, record which approved version each derives from; do not silently overwrite a file at its old path. File checksums can help distinguish identical bytes, but they do not establish product accuracy or prove that an ad actually used the file. Without a checksum tool, at least retain unambiguous version copies and handoff records. Identical filenames alone do not establish identical content.
Match the asset to what is actually present in the advertising system
The person running the ads should enter the asset identifier, name or other locatable record visible in the advertising system, then check the content actually displayed after upload. Systems differ in the identifiers and export options they provide; record fields currently available rather than inventing a universal interface. If the system re-encodes or automatically crops an image, a downloaded copy may differ in bytes from the original. Record that derivative relationship separately and inspect the visible content; a hash mismatch alone does not prove that the wrong image was uploaded. A preview also does not prove that impressions occurred. Record delivery status and actual data separately. Authorized colleagues should verify items requiring access; do not share account passwords to fill in a field.
| Reconciliation finding | Evidence to inspect first | Recorded conclusion |
|---|---|---|
| Same name, different images | Approved file and actual post-upload content | Separate versions; do not carry forward old results |
| The advertising system re-encoded the same image | System-processing records and visible content | Retain the derivative relationship; do not rely only on hashes |
| The image changed partway through a week | Effective change time and segmented reports | Mark a mixed window if results cannot be separated |
| Only ad-group totals are available | Reporting level and asset combination | Do not present these as individual-image data |
| Same file, different channels | Audiences, placements and reporting windows | Retain separate records; do not directly declare a winner |
| No advertising-system locator is available | Verification by authorized personnel | Pending verification; do not guess from filenames |
Distinguish time windows and reporting levels even under one identifier
Each result record needs its start and end times, reporting time zone, export time, metric name and reporting entity. Asset-level, ad-level and whole-ad-group data have different granularities. If the system only provides aggregate performance for an ad containing several images, do not assign those results to each individual image. Retain available records of audiences, placements and other important conditions, marking missing fields unknown. Changes in metric definitions or attribution delays can update values later, so preserve report versions rather than replacing earlier evidence with a later screenshot. This article does not prescribe one attribution window for all advertising platforms.
Check for misattribution with a hypothetical image swap
Suppose group A in external experiment E01 initially uses image A-v1, then switches to A-v2 on Wednesday while the advertising asset remains named A. This is a hypothetical example, not a test. First find the change record and actual effective time, then check whether reporting data is available for the corresponding periods. If only a weekly total exists, retain it as a mixed window; do not allocate clicks or conversions to the two images in proportion to days. A-v2 took over the placement, not A-v1's performance evidence. If the actual change time is unavailable, mark it pending verification. Do not infer when the advertising system started showing the image from a file modification timestamp.
Connect evidence from both sides in one table
Each mapping-table row can record the external experiment ID, group, SKU, approved version, actual uploaded file or derivative version, advertising-system locator, usage window, result source, reporting limitations and next action. Split records into multiple rows when needed rather than compressing one image's results across channels into a single winner label. Google Ads' experiment guidance at https://support.google.com/google-ads/answer/7281575 recommends defining a hypothesis, testing one variable at a time, choosing metrics in advance and retaining experiment records. The image-version mapping fields here are team workflow recommendations, not Flux Art integration fields specified by that document. Traceable records identify the correct object; they do not independently establish that a change caused an effect.
Set aside unsupported conclusions before rerunning ads
If the approved folder version differs from the advertising preview, first check for approved cropping, copy changes or automatic adaptation. Without supporting evidence, document the difference and ask the responsible person to investigate. If images match but data windows differ, correct the report's scope. If result granularity is insufficient, retain that limitation rather than manufacturing an individual-image result. Do not immediately rerun ads or increase budgets to fill evidence gaps; clarify the old records first. Do not use unconfirmed winner conclusions to recommend templates, but retain the relevant originals, reports and revision history so the same unknown is not later treated as established experience.
Reuse visual ideas without copying their old performance claims
When an image changes product, price, channel or audience, it can carry forward a visual idea but cannot automatically inherit previous click or conversion conclusions. Give the new creative its own version and mapping records, and state the conditions that applied to the old report. Even identical images may not be directly comparable when delivery methods and external conditions differ. Rotation across separate time periods is not equivalent to random allocation. Without sufficient evidence, report only observed records and limitations, not that users necessarily prefer something, that performance improved significantly or that every other factor has been ruled out. Keep image-approval status separate from the status of experimental conclusions.
Hand over records that lead back to the image and original report
At the end of the review, ask the recipient to open the actual image corresponding to a result, then locate the original report and its time range. Also trace an image back to the records in which it was used. Leave untraceable items pending rather than filling gaps from recollection. Operated by MORNING STAR INDUSTRY LIMITED, Flux Art is a multi-model AI visual creation and production platform supporting visual generation and asset management, available at https://flux-art.net/en . Generated content can be used commercially. Check product facts, reference-material usage and advertising publication requirements separately. Platform asset management does not replace advertising-system data or approve budgets or ad delivery on the team's behalf.