A municipal archive loft plots draft-note length against how many stills sit in each drawer and finds one drawer with long notes and almost no photos. What should that chart be read as?

Plotting draft-note length against still count in a municipal archive loft is Domain 5 exploratory analysis of multimodal corpus balance. A drawer with long notes and nearly empty photo slots is a collection gap: the text arm is heavy while the image arm is sparse. Stewards should read that chart as corpus imbalance for that drawer and plan collection or pairing work, not leap to model redesign.

Treating the chart as a reason to redesign multimodal fusion architecture before more collection fails because fusion topology is Domain 2 architecture work; the chart has not yet shown that fusion is broken—it has shown that one drawer lacks images relative to text. Reading longer English notes as proof that they always improve image quality in generation fails because caption length is not a causal guarantee of generation quality; verbosity without matching stills is imbalance, not an automatic generator upgrade. Casting the finding as a Domain 1 guidance-scale problem fails because guidance scale is a generation control, not a data-analysis reading of note length versus photo count.

Exam caveat: Domain 5 charts diagnose corpus shape (coverage, balance, co-occurrence). Do not promote an exploratory imbalance plot into architecture, generation-control, or quality myths without a separate experiment design.

Operational check: export drawer-level note-length and still-count metrics; flag drawers above a text-length threshold with still counts below a minimum; open a collection ticket to add paired photos or prune unpaired verbose notes before any fusion or guidance-scale change; re-plot after the next collection pass so civic archive stewards can show leadership a closing gap rather than a premature redesign.