Module 5: Conducting an Assessment

5.1 The Assessment Engagement Shape Core

A standard assessment is 2 sessions over 1-2 weeks, plus async artifact review.

• Session 1 (90 min): Structured interview with the client's data-aware stakeholder(s). Usually owner + ops lead or CTO + lead engineer.
• Async (3-5 days): Artifact review. Schema dumps, sample exports, audit log samples. Optionally a live introspection run via SDC Agents SMB.
• Session 2 (60 min): Findings presentation and scoring validation.

Total practitioner time: 8-12 hours including prep, review, and writeup.

5.2 Pre-Engagement Checklist Core

Before Session 1:

• Signed engagement letter with NDA
• List of systems the client considers in scope (CRM, accounting, ops, EHR, etc.)
• Identification of the right interviewees (owner + technical lead minimum)
• Read-only credentials or sample exports for at least the primary datasource
• Time blocked: 90 min uninterrupted, no Slack, no email

5.3 The Interview Core

The interview is structured around the six dimensions but does not feel like a survey. Open with business context, then drill down.

Opening (10 min)

• "Tell me about the business and where data fits in."
• "What is the most painful data problem you have right now?"
• "If you could wave a wand, what would change?"

The pain question is critical. The map will eventually show a lot of red. You want the client to see their own pain reflected in the map, not feel ambushed by your scoring.

Per-dimension drilldown (10-12 min each)

For each dimension, ask 3-4 questions from the bank, listen, take notes, and assign a working score in the margin.

Schema Integrity question bank

• "If I asked your database administrator for a data dictionary, would I get one? When was it last updated?"
• "When you add a new field to a form, what happens in the database? Who decides? Who tests it?"
• "Have you ever had a report break because someone renamed a column? What did you do?"

Constraint Enforcement question bank

• "What stops a user from entering a customer with no email?"
• "If a value is supposed to be one of five categories, where is that list enforced?"
• "When the rules change, how do you update old records?"

Semantic Identity question bank

• "If I gave you a customer ID from your CRM and asked your accounting system to find the same customer, what would happen?"
• "Have you ever had two records for the same person? How did you find out?"
• "What happens to an ID after a record is deleted?"

Provenance question bank

• "If a number on your dashboard is wrong, can you trace back to who entered it?"
• "How long do you keep audit logs?"
• "Can you tell me the value of [field X] on [date Y]?"

Interoperability question bank

• "How do you share data with [partner / vendor / regulator]?"
• "How long did your last integration take?"
• "If a new partner asked for your data tomorrow, what format would you send it in?"

Governance question bank

• "Who is responsible for data quality? What is their title?"
• "When was the last data quality issue and how was it resolved?"
• "Do you have any compliance reporting obligations?"

5.4 Artifact Review Intermediate

During the async period, gather and review:

• A copy of the data dictionary (or absence thereof)
• A schema dump from the primary database (pg_dump -s, mysqldump --no-data, or similar)
• A sample of 100 rows from each of 2-3 representative tables
• A sample of the audit log (last 7 days)
• Any existing data quality reports

Use SDC Agents SMB for rapid introspection

sdc-agents introspect run \
  --datasource client_main \
  --output-format json

The 13-field column metadata returned by introspection answers half the assessment questions automatically:

• Nullable counts → Constraint Enforcement evidence
• Sample value variance → Schema Integrity evidence
• Identifier column detection → Semantic Identity evidence
• Detected anomalies → Direct evidence for the report

5.5 Documenting Findings Core

For each dimension, capture:

Dimension: [name]
Working score: [1-5]
Evidence:
  - [Quote or observation 1]
  - [Quote or observation 2]
  - [Artifact reference]
Confidence: [High / Medium / Low]
Notes:
  - [Anything that might shift the score during validation]

5.6 Handling Scope Ambiguity Intermediate

You will encounter clients with 14 systems and no clear boundary. Pick a scope and stick to it. The map is for a defined data estate, not for "the company." If the client has 3 distinct lines of business, do 3 separate maps.

If you hit those, pause, and re-scope before continuing.