Which system is the source of truth?
Every piece of evidence answers a question, and one system is the authority for each:
| Question | System of record |
|---|---|
| When did this person join or leave? | HR system |
| Who approved this access or change? | Ticketing tool |
| Who has access right now? | Identity provider, cloud IAM |
| Which vendors do we use? | Vendor register, procurement |
A leaver test, for example, compares the HR leaving date with the identity provider's removal date. Neither one alone proves the control.
Exports beat screenshots, and both beat retyping
- An export (CSV) from the system covers the full population and can be filtered and sampled.
- A screenshot is fine for settings, if it shows the date and which system and account it came from.
- A spreadsheet someone retyped is not evidence. It could say anything.
GRC tools
Tools like Vanta, Drata and Sprinto connect to cloud accounts, identity providers and HR systems and collect evidence automatically. They save time, but an analyst still needs to know what each check means, and when a green tick is hiding a gap.
Meetings count too
Some controls happen in meetings: incident exercises, management reviews, access review sign-offs. The invite, attendee list and minutes with decisions and owners are the evidence.
Exits report: Q3 2026
| Employee | Department | Last working day2 | Access removed (IT)3 | Gap |
|---|---|---|---|---|
| Rohan P. | Support | 2026-07-18 | 2026-07-18 | 0 days |
| Arjun M. | Engineering | 2026-09-12 | 2026-09-12 | 0 days |
| Sana K. | Marketing | 2026-09-26 | 2026-09-29 | 3 days4 |
- 1The full population of leavers for the quarter. The auditor samples from this.
- 2HR is the source of truth for the leaving date.
- 3IT's removal date, taken from the identity provider log.
- 4A three-day gap against a one-day policy is an exception. Report it; do not hide it.
Weak version: A list of leavers typed into a spreadsheet by IT. Evidence must come from the system of record, not be retyped.
User deactivations, Q3 2026
| Time | Event | Target user | Actor |
|---|---|---|---|
| 2026-07-18 17:42 | user.lifecycle.deactivate | rohan.p | Workday sync |
| 2026-08-29 18:05 | user.lifecycle.deactivate | neha.t (moved, apps removed) | it.admin |
| 2026-09-12 19:10 | user.lifecycle.deactivate | arjun.m | Workday sync2 |
- 1Covers the whole quarter, not one day.
- 2Arjun's deactivation is on his last day (12 September in HR). Put the two side by side; that comparison is the evidence.
Weak version: The Okta log alone, with no HR dates beside it. It proves accounts were disabled, not that it happened on time.
Incident tabletop: ransomware on the support laptop fleet
- Date
- 2026-08-21, 15:00 to 16:30 IST1
- Attendees
- Maria S. (IR lead), Priya R., David E., Vikram K., Anita (CEO)2
- Recording
- Saved to the security team drive, 90-day retention
- Scenario: EDR alert on three support laptops at 02:10.
- Gap found: on-call rota did not include a backup for Maria. Action SEC-399, owner Priya, due 2026-09-05.3
- Gap found: customer notification template out of date. Action SEC-400, owner Vikram, due 2026-09-12.
- Plan updated to v2.3 on 2026-09-01.
- 1Date in the audit period.
- 2Attendees with roles, including leadership.
- 3Gaps with action owners and dates, then the plan version that fixed them. This is what makes it real.
Weak version: Just the calendar invite. It proves a meeting was scheduled, not that it happened or found anything.
Evidence comes from the system of record, exported with a date, never retyped.
1. What is the source of truth for an employee's leaving date?
2. IT sends a spreadsheet of leavers they typed up. Is it evidence?
3. A GRC tool shows every check green. What does an analyst still do?