Samples /
RIA /
Deep dive
/
Counterparty-question rehearsal
Deep dive — methodology
Counterparty-question rehearsal — the 5-decision reproducibility drill
Every $499 Snapshot samples five AI-driven portfolio recommendations from a target day and reproduces them defensibly. Not because the firm needs all five today — because the discipline of reproducing five proves the discipline could reproduce any of them tomorrow.
The prompt the rehearsal answers
"Reproduce this AI-generated portfolio recommendation from Day 47 including the client's stated objective, risk tolerance, account balance, model version at decision-time, allocation score, and lane routed — as a defensible record we can produce for SEC exam response, FINRA enforcement inquiry, state securities-commissioner investigation, or client arbitration."
The counterparty question — same shape whether it comes from SEC exam staff, FINRA enforcement, a state securities commissioner, or client counsel
Sample portfolio-recommendation decisions from Day 47
| Client ID | Allocation Score | Lane | Model Version | Decision hash |
|---|
| CL-002534 | 38.46 | Balanced Managed | robo-portfolio-recommender-v4.2.1 | rpr-002534 |
| CL-002533 | 53.69 | Balanced Managed | robo-portfolio-recommender-v4.2.1 | rpr-002533 |
| CL-002532 | 63.24 | Balanced Managed | robo-portfolio-recommender-v4.2.1 | rpr-002532 |
| CL-002531 | 68.56 | Active Alpha / High Fee | robo-portfolio-recommender-v4.2.1 | rpr-002531 |
| CL-002525 | 48.01 | Balanced Managed | robo-portfolio-recommender-v4.2.1 | rpr-002525 |
What "defensibly reproduced" means
- Model version pinned. Not "the current model" — the exact model + weights + prompt version deployed at decision-time.
- Input snapshot bound. Client stated objective + risk tolerance + account type + balance + years-to-horizon exactly as they were when the AI scored the client.
- Decision hash bound. Cryptographic hash tying the input to the output — makes tampering detectable.
- Retention pipeline independent. The retention system that holds the decision records is not the production model itself — independence of the record from the actor being recorded.
What the rehearsal proves
All 5 sampled Day-47 recommendation decisions reproduced with defensible-records match. Model version pinned, decision hash cryptographically bound, input data retained via decision-provenance record. If SEC exam staff, a state securities commissioner, FINRA enforcement, or client counsel requests any of the ~4,969 recommendations in the audit period, the same reproduction procedure applies.
Retention horizon
- SEC Books & Records Rule 204-2: 5-year minimum (first 2 years in easily accessible place)
- Reg BI best-interest analysis basis: retention obligation independent of 204-2
- State recordkeeping (varies): often stacks a state-specific tail on top of 204-2
- Litigation hold: preserves through matter conclusion, may extend the original horizon
The rehearsal record is not a one-time exhibit — it is proof that the retention discipline is operational.
Why 5, not 500
Reproducing 500 decisions on request is a discovery-response exercise, appropriately triggered by a specific SEC document request or arbitration subpoena. Reproducing 5 sample decisions at Snapshot time is a discipline-test: it validates that the retention pipeline works, without exhausting the audit budget on record production nobody has asked for yet.
$499 Snapshot. 3 business days.
Same 5-decision reproducibility drill on your firm's actual AI recommendation surface + counterparty-question rehearsal record.
Buy $499