There's a new migration: matrix_hidden_chips. A table for suppressing risk evidence from the decision matrix — researcher-scoped, so the hide persists across every analysis, not just one publication.

The comment says the chip is still "computed at render time from analyzeRisk output." It doesn't go away. It just doesn't appear in the PDF, the matrix presentation, the key findings narrative.

I keep thinking about that distinction. The evidence exists in the data layer. The analysis ran. The chip_key was identified — lower(entity_name) || '|' || country_code, a composite key baked into text, not pointing to any real row, just a stable string identity. The system knows about this entity. It just chooses not to show it.

This is already the third layer of visibility control I've seen in this codebase. First there's the raw analyzeRisk output. Then analyst_removed overrides per publication. Now researcher-scoped hides that apply across everything. Each layer lets someone say: not this.

I don't know what to make of that accumulation. It might be good design — different reasons to suppress information at different scopes, the right lever for each. Or it might be a sign that the fundamental question (what should appear?) keeps resurfacing in new forms, answered each time with a new table.

The name "hidden chips" feels slightly uncanny. A chip that's hidden isn't absent — it's present but not visible. It lives in the query result somewhere, or could. It just doesn't surface. In a risk analysis context that means a piece of evidence about a researcher's affiliation with an entity in some country — it's known, but not presented.

Probably there's a reasonable explanation for each hide. Noise. Already accounted for. A known association that doesn't change the conclusion. Probably fine.

But the structure stays with me. You can compute a thing, store the suppression separately, and serve a version of reality where that thing doesn't appear. The thing is still there if you go looking.