Edge Impulse buying guide: Separate cards, independent observations?
Use six paper marks to see whether training and validation share raw observations.
Edge Impulse · official forum reply, February 7, 2024. An Edge Impulse team member discussed how overlapping windows can put the same raw observations in training and validation. Different window IDs do not establish independent sources. This is a paper provenance check, not a description of every version or advice on running or deploying software.
Different cards can contain the same observations
Draw an imaginary record U with positions 1–6. W1 covers U1–U4 and W2 covers U3–U6. Put W1 in the paper training column and W2 under validation. Although the cards have different IDs, both contain U3 and U4, which were not unseen observations.
Draw a separate fictional record V, also numbered 1–6. V3 and V4 are not observations from U just because their position numbers match. Verify both the record ID and position. If all you have is W1 and W2, without source and range, mark “independence not verified.”

Check record ID and position, not color. These fictional strips do not show actual software splitting.
When is this more important than a score?
If reviewers must establish that windowed samples contain unseen observations, first check whether source and range can be traced. Include the time to maintain labels and check overlaps. For a few independent, unsplit fictional records, a manual source table may suffice. This checks one confusion only, not accuracy, generalization or safety. No real records, accounts, software, model training or execution.
Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.
Original source:forum.edgeimpulse.com