[GRADE D -- v3.0 Detective Mode Mining | Phases D38-D40]
Ghost score = redaction proximity / document mentions. High score = high redaction proximity relative to low visibility. The top ghost controllers:
| Name | Ghost Score | Redaction Proximity | Doc Mentions |
|---|---|---|---|
| Gaver | 50.00 | 5 | 1 |
| AUSA Barnes | 40.00 | 4 | 1 |
| antonia barnes | 30.00 | 3 | 1 |
| SA redacted | 20.00 | 2 | 1 |
| R. Alexander Acosta | 10.00 | 1 | 1 |
| Rey Epstein | 10.00 | 2 | 2 |
| JANE DOE NO 4 | 10.00 | 1 | 1 |
| Mark T Luttier | 10.00 | 1 | 1 |
| Phyllis A Dames | 10.00 | 1 | 1 |
| SHARON R BOCK | 10.00 | 1 | 1 |
| Goldberger | 0.51 | 4 | 79 |
| Alfredo Rodriguez | 0.24 | 6 | 255 |
"Gaver" (score 50): The highest ghost score in the corpus. 5 redaction-proximity connections from a single document mention. This entity name appears once, surrounded by 5 redactions, and nowhere else. Identity unknown.
"AUSA Barnes" / "antonia barnes" (scores 40/30): Likely the same person -- an Assistant US Attorney with 4 and 3 redaction-proximity connections respectively. May refer to SDNY prosecutor involved in the Epstein case. The pattern of AUSA names appearing near redactions but with minimal overall visibility suggests deliberate minimization of prosecutorial involvement records.
R. Alexander Acosta (score 10): The former US Attorney who signed the 2008 NPA. His ghost score of 10 -- moderate redaction proximity with minimal document visibility -- places him in the "deliberately reduced footprint" category. For the man who gave Epstein the plea deal, appearing in only 1 document while having redactions nearby suggests records were sanitized.
Alfredo Rodriguez (score 0.24, but 6 redaction proximity): Epstein's house manager who stole the "little black book" and attempted to sell it. His high absolute redaction proximity (6) despite 255 document mentions suggests that information NEAR his testimony was systematically redacted.
AuraDB memory error. The query designed to rank persons by appearance across flights + emails + documents + images + organizations exceeded the 278MB transaction memory limit. This confirms the query requires optimization for the AuraDB Professional tier.
Relationship type census (36 active types):
| Relationship | Count | Previously Traversed |
|---|---|---|
| MENTIONED_IN | 1,581,564 | v1.0 |
| ICIJ_LINKED_TO | 1,462,273 | Never |
| VOTED_IN | 919,191 | Never |
| LINKED_TO | 328,183 | Never |
| LINKED_TO_OFFSHORE | 278,555 | v2.0 |
| CO_APPEARED_WITH | 182,175 | v1.0 |
| SENT_EMAIL | 66,321 | Never |
| TRAVELED_TO | 20,792 | Never |
| IMAGE_FROM | 10,399 | Never |
| FLEW_ON | 7,354 | v3.0 (first) |
| TRAVELED_WITH | 219 | v3.0 (first) |
| ASSOCIATED_WITH | 160 | v3.0 (first) |
| COMMUNICATED_WITH | 9 | v3.0 (first) |
| PAID_BY | 1 | v3.0 (first) |
| RECRUITED_BY | 1 | v3.0 (first) |
| REPRESENTED_BY | 3 | v3.0 (first) |
| LIKELY_IDENTITY | 3 | v3.0 (first) |
| NEAR_REDACTION | 67 | v3.0 (first) |
v3.0 traversed 18 relationship types total -- 13 for the first time. The graph census reveals that 1,462,273 ICIJ_LINKED_TO edges and 919,191 VOTED_IN edges have NEVER been traversed by any mining operation. These represent the single largest unexplored data mass in the graph.
Node type census (27 types, 5.37M total):
| Node Type | Count | Mining Status |
|---|---|---|
| Document | 1,518,414 | Extensively mined |
| SanctionedEntity | 1,115,260 | v3.0 (false positives) |
| OffshoreEntity | 913,818 | Pipeline gap |
| Officer | 771,315 | Pipeline gap |
| Person | 367,413 | Extensively mined |
| VisitRecord | 202,828 | v2.0 |
| RedactedEntity | 107,480 | v1.0/v3.0 |
| JmailEmail | 70,039 | v1.0/v2.0/v3.0 |
| ReconstructedPage | 39,588 | v2.0/v3.0 |
| ImageEvidence | 38,955 | v2.0 |
| FinancialTransaction | 9,266 | v3.0 (first) |
| Flight | 4,482 | v1.0/v3.0 |
| Transcript | 1,530 | v2.0 (empty) |
| Engagement | 970 | v3.0 (first) |
| Investigation | 5 | v3.0 (first) |
| UnaccountedPeriod | 0 | v3.0 (confirmed empty) |
WHAT THIS SHOWS AND DOES NOT SHOW: The graph census confirms the v3.0 paradigm shift was overdue. The 2,096 curated relationship edges -- the highest-confidence data in the graph -- were invisible to v1.0 and v2.0. The ghost controller analysis reveals persons whose visibility in the record has been deliberately minimized relative to their proximity to redacted information. The remaining unmined frontier is enormous: 1.46M ICIJ edges, 913K offshore entities, and 771K officer nodes have never been traversed.