[GRADE D -- Analysis of FOIA exemption patterns across the corpus (GOY-14 mining G20-G24)]
A critical finding: the structured exemptions property on Document nodes is not populated in the graph. All queries for b3, b7D, b1, b6+b7C, and b7E using property-based filtering returned 0 results. The exemption census query returned 0 rows. Exemption data exists only within:
This means the 258-page deletion's exemption profile comes from the ghost entity's constraint metadata, not from document-level FOIA annotations. The graph lacks systematic exemption tracking across the 1.5M document corpus -- a major unmined frontier.
Despite the empty exemption fields:
The FBI's "Special Redaction Project" consumed 4,737 overtime hours to redact Epstein files. Bloomberg's Jason Leopold uncovered this through FOIA.
The top documents by total_redactions property and by RedactedEntity count each returned 50 results (capped), indicating substantial redaction infrastructure exists in the graph's document metadata.
WHAT THIS SHOWS AND DOES NOT SHOW: The absence of structured exemption data at the document level is itself a finding -- it means FOIA exemption codes were not systematically extracted during document import. The ghost entity metadata preserves exemption codes only for the 10 FBI entities. The presence of "confidential source" (50 hits) and "cooperating witness" (20 RP hits) in fulltext confirms the FBI was managing confidential human sources in the Epstein investigation, but does not identify who those sources were.