Worked example: Source Profiling Agent conversation

Demonstration only. Synthetic DRH source extracts. Not a live Trust agent.

What this example shows

This is a curated worked example from Run 2 of the Demo Rivers Health warehouse demo — not a verbatim chat transcript. The dialogue illustrates how the Source Profiling Agent interprets profiler outputs: cross-source volume trends, extract-change risk and handoff boundaries.

Evidence chain

Agent rule

• Source Profiling Agent rule

Evidence data

• Volume trends — IUCS contacts, cases opened, agency spend by month
• DQ register — data-quality issues flagged during profiling
• Linkage analysis — CareCall to CareCase match scenarios
• Extract change log — CareCase inclusion rule change (March 2026)

Checks and scripts

• Source profiling report — full profiler narrative output
• Profiler script — reproducible profiling run

Audit trail

• Runs 2–5 internal QA — cross-artifact consistency checks
• Demo run index — Run 2 artefact register

Turn 1 — Why did CareCase cases jump in March?

User:

CareCase cases opened went from 525 in February to 542 in March. Is urgent care demand increasing?

Source Profiling Agent:

I would not assume operational demand from the case count alone. Cross-source evidence does not support a front-door surge in March.

Volume evidence (volume_trends.csv):

IUCS contacts fell 13.9% month-on-month in March while cases opened rose slightly. The increase in cases without SourceContactId (123 → 322) accounts for most of the March movement.

Extract change evidence (localops_extract_change_log.xlsx):

• ChangeDate 2026-02-28, SystemName CareCase, ExtractName NightlyCaseExtract
• ChangeType: Inclusion rule change — PendingAdminClosure and AwaitingSignoff included from March run
• ExpectedImpact field notes higher case counts from March; not activity-led

Hypothesis (draft): March case increase may be extract-driven. CareCall IUCS does not corroborate operational demand.

Checks before using March cases in a pack:

1. Filter cases with empty SourceContactId and review CaseStatus / ExtractInclusionFlag.
2. Compare to CareCall IUCS trend for same month.
3. Confirm with data owner whether inclusion rule changed.

Status: Draft finding — pending human review.

Turn 2 — What about January and February?

User:

January and February also look busy. Is that the same issue?

Source Profiling Agent:

January–February look different from March — multiple sources move together.

CareCall (volume_trends.csv): IUCS 1,592 in Jan (+22% vs Dec), 1,477 in Feb — both above Sep–Nov baseline (~1,263).

CareCase: Cases opened 461 (Jan) and 525 (Feb); most rows retain SourceContactId (only 123 without in Feb vs 322 in Mar).

RosterFlow: Urgent-care bank shifts 219 in Feb vs ~190 in prior months.

LedgerWise: Agency nursing spend on CC-URG-401 ~£61k in Feb (volume_trends.csv, AgencyNursingSpendGBP).

LocalOps (localops_validation_notes.xlsx): Notes dated Feb 2026 reference bank-holiday pressure and recovery plan.

Hypothesis (draft): Jan–Feb pattern is multi-source operational uplift, not the same mechanism as March.

Turn 3 — Can we start building the warehouse?

User:

OK — propose the fact tables.

Source Profiling Agent:

That is outside this agent's brief. Profiling is complete; staging recommendations are in source_profiling_report.md section 8.

Next step: Hand off to Warehouse Design Agent with suggested_run3_agent_task.md. Do not skip design review.

Required sign-off before build: Information Lead / Performance & BI (demo roles).