The brokerage promises the client a relationship with the firm, but every book is isolated (rightly), so when the roster changes, the firm forgets. This layer is that memory, built without breaking the isolation. The rule: identity and aggregates pool into a governed projection; content, history, and the mailbox never do. Each agent's book stays isolated (the moat). Across books, resolution links, it never merges: two agents who know the same client each keep their own relationship; the firm only learns “same human.”
The orphaned book routes to the agent who already has a link, not round-robin. The no-signal case is flagged honestly, not dressed up as a match. Each card below is the engine's own recommendation, carrying its own stated reason.
Governance is the spine, not the headline. Brokers buy the outcome; these rules are why the agents say yes, and why this is not the surveillance version anyone could build. They ship opinionated: one matrix, not a matrix builder.
| What pools | The agent sees | The broker sees | Another agent sees |
|---|---|---|---|
| Records, notes, message content, the mailboxThe moat. This is the row that never moves. | Everything, unchanged | Nothing | Nothing |
| Resolved identity links (“same human”)Link, never merge. A false link leaks a client, so name-only is review. | Own book, plus an unnamed co-ownership flag | Aggregate collision counts | Nothing until both agents confirm |
| Transaction facts: deals, sides, GCIThe firm already owns these. Pooling them breaks nothing. | Own production, plus influence they earned elsewhere | Firm rollups by agent and by cohort | Nothing |
| Individual activity: calls, opens, cadenceAggregates to the broker are fine. Individual activity is surveillance. | Their own | Never, at any granularity | Never |
| At the orphan momentThe one moment naming is on: the client needs an agent today. | Sees their book leave with them, in writing | A ranked reassignment with the reason shown | The named client, on assignment |
Agreed while relations are good, so offboarding stops being a fight. It cuts both directions: the same architecture is the arrival-side answer to a recruit asking what the firm will be able to see of what they bring.
| Agent | Production | Influence | |
|---|---|---|---|
| Tom Gallagher | active | $42K | +$10K |
| Maria Santos | active | $26K | – |
| Dale Foster | departed | $22K | – |
Tom Gallagher's $10K of influence is GCI driven inside Maria Santos's book, by handing over one buyer (Paul Briggs). A flat per-agent production ranking cannot see it: the hidden originator, at firm scale.
The same cohort lens as Layer 4, rolled up across every agent's book. The broker sees which sources drive firm production, not just which agents.
Honesty notes (carried up from Layers 2 to 4): co-owned GCI is counted once, in the
owning book (co-ownership is flagged, not double-billed). Cross-agent links are assisted and banded: a
human confirms, and a name-only match is review, never auto, because a false cross-agent link leaks one
agent's client to another. Reassignment is a ranked recommendation. Synthetic fixture; production
wires the same engine to the real per-agent Cloze books. Every number on this page renders from
brokerage/firm_graph.json, which the engine emits.