Skip to content

Ref as Python dict/UUID repr in buyer chrome

The corpus prior

Agents optimise emit coverage and byte floors. After person×queue_meta Avatar work, preferring FK entity dicts for Avatar initials is correct for people — but the same path stringifies device/entity dicts via str(dict), so buyers see Device: {'id': UUID('…')}. Floors and seed residual stay green because they do not OCR still text. The prior is “residual_total=0 means chrome is honest.”

Wrong shape

# Prefer any FK dict for queue meta
raw = item["device_id"]  # {"id": UUID(...)}  id-only join
value = str(raw)         # "{'id': UUID('…')}"
# product_quality residual_total = 0  (bytes + seeds only)

while issue_triage_* still shows Python repr and Trust collapses.

Right shape

  1. Prefer entity dict only when it has display substance (name/code/email/…); else *_display / _ref_display_name — never str(dict).
  2. Route person through present(person, queue_meta) → Avatar; non-person refs through human display names.
  3. OBSERVE with:
  4. MCP product_quality(operation=score) — residual_total includes presentation
  5. MCP presentation(operation=residual|opportunities|cognition)
  6. CLI dazzle demo quality / dazzle qa hyperpart-opportunities
  7. Force /improve framework-ux hyperpart_presentation when presentation residual > 0.
  8. Recapture hero stills after emit fix; open the PNG (stills beat claims).

Why this matters here

Dazzle’s agent thesis is that tools form true beliefs about product surfaces. A green residual coexisting with dict/UUID chrome is a false belief — the same class as empty-desk false green, but for presentation honesty rather than seed hits. #1626 antagonist 01 Aug b: fieldtest −1.0 despite residual_total=0.