Bob Resolve / Book of business
Resolve identity questions with source evidence
Compare the source records before deciding whether a pair is the same person. A false merge is worse than a missed match.
As of 2026-10-01.
Needs attention
Unresolved review pairs
19
Compare each pair side by side. Read the source evidence and guard rails before making a decision.
Compare review pairsRead-only. Apply review decisions separately to a new run; leave uncertain pairs unresolved.
Review breakdown and recorded status
Suggestions
High0Medium19
- Suggests same person
- 0
- Suggests different people
- 10
- Suggests unsure
- 9
Recorded review status
Browser review labels are unauthenticated declarations, not validated resolution evidence or approval.
- Confirmed by a reviewer
- 0
Merges a person approved.
Items no one has decided yet.
Unidentifiable rows
9
No name, birth date, or MBI. Set aside from identity grouping.
Ask the source owner to restore identity fields before rerunning these rows.
Records in
3,887
CRM 2,040, Enrollment 1,847, including 9 unidentifiable
Candidate identity groups
2,000
Engine output, not a count of human-confirmed identities.
Households
1,400
Linked by the agency's own household ids
Inspect grouped source records to see where each field came from and why records were combined.
Benchmark and engine details
These demo fixtures were used while building the rules. They are not held out and do not measure accuracy on real agency files.
MBI used for matching, then masked for display. Masking is not a no-shared-ids evaluation.
Real agency data and production access remain separate gates.
Run demo-run, engine 0.1.0. 0 identities left split, measured on synthetic data.
How sure are we? Measured on synthetic data
Precision of auto-merges
Meets target100.0%
Target at least 99.0%. Enrollment side derived from the answer key, shared ids on. Measured on synthetic data.
Automatic recall
99.6%
2,159 of 2,167 true pairs merged with no reviewer. Enrollment side derived from the answer key, shared ids on. Measured on synthetic data.
Blocking recall
Meets target100.0%
Target at least 98.0%. Enrollment side derived from the answer key, shared ids on. Measured on synthetic data.
Recall if every same-person suggestion were confirmed
99.6%
Hypothetical. Counts a same-person suggestion as found even with no reviewer. Enrollment side derived from the answer key, shared ids on. Measured on synthetic data.
Merges by tier
- Rules
- 2,159
- Jev
- Off
- LLM
- Off
- Human review
- 0
Auto-merged only when very sure.
Off: not built yet. Recording costs money and needs approval each time.
Off until an Anthropic key and a spend cap are approved.
No review decisions applied to this run yet.
Cost and run time
- Jev
- $0.00, 0 calls
- LLM
- $0.00, 0 calls
- Run time
- Not recorded
Off
Off
Fixed-clock demo run, so a rerun is byte for byte the same.