An ice-cutter pond office writes a polished greet / ask / show script but has no stills or captions. How should that script be filed?
Select an answer to reveal the explanation.
Short Explanation
A beautiful script is agent choreography, not a shelf of stills and captions. Don’t file it as a CLIP or fusion corpus when the media arms are empty.
Full Explanation
A polished greet / ask / show script with no stills or captions should be filed as agent-side instructions, separate from a paired multimodal corpus. That filing is correct for the ice-cutter pond office: scripts direct orchestration; they are not CLIP image–text pairs or early-fusion datasets. Civic playbooks without media remain useful runbooks but must not be counted as multimodal training gold.
Calling the script a complete CLIP image–text corpus fails without stills and captions. Calling it an early-fusion dataset ready for concatenation fails for the same missing-media reason. Claiming agent scripts replace paired multimodal corpora fails conceptually: instructions do not substitute for paired examples.
Exam caveat: keep agent scripts and paired corpora in separate civic inventory buckets. Operational check: store scripts under runbook/agent docs; require actual still–caption (or other modality) pairs before claiming a multimodal corpus. Keep municipal reviewers in the loop whenever modality claims, fusion labels, or agent scripts are promoted from lab notes into the production corpus used by public-facing services. Record the decision, owner, and date beside the dataset manifest.