
Governance & Compliance
https://hmmnm.com/iso-27001-from-scratch-part-4-evidence-documents-and-the-trail/
TL;DR: ISO 27001 audits are won or lost on evidence, not intent. If you maintain a short list of mandatory documents, separate policies from evidence, and use consistent naming and versioning, an auditor can trace any control to proof in minutes — and you can simulate that entire trail in your own home lab before certification ever matters.
Authorization note: everything in this article is performed on your own lab machines or VMs that you own and control. Do not run evidence-collection scripts, account dumps, or log queries against systems you do not have explicit permission to audit.
An ISO 27001 auditor does not grade your security on vibes. They grade it on whether you can show, on demand, that a control exists, is documented, and is actually operating. The phrase auditors use is “objective evidence” — artefacts that independently demonstrate the control works: logs, tickets, review records, signed policies.
The good news: the standard’s mandatory documentation set is small. Everything else you generate is supporting evidence you control.
ISO 27001:2022 explicitly requires the following. Build these first, because everything else hangs off them.
In a lab, keep them in a single repository so the whole trail is one folder tree:
mkdir -p isms/{01-policies,02-procedures,03-evidence,04-internal-audit,05-management-review,06-records}
tree isms -L 1
# Expected output:
# isms
# ├── 01-policies
# ├── 02-procedures
# ├── 03-evidence
# ├── 04-internal-audit
# ├── 05-management-review
# └── 06-records
Beginners conflate two different things:
A policy with no evidence is a nonconformity waiting to happen. Evidence with no policy is uncontrolled activity. The audit trail always runs: policy → procedure → evidence → record of review. If any link is missing, the auditor writes a finding.
When an auditor asks for “your last access review,” you have 30 seconds to produce it. That only works with a rigid convention. Use this pattern:
[TYPE]-[CONTROL-REF]-[DESCRIPTION]-[YYYYMMDD]-v[X.Y].[ext]
Examples:
POL-A.5.1-information-security-policy-20240601-v3.0.pdf
SOP-A.5.18-access-review-procedure-20240315-v1.2.pdf
EVD-A.5.18-q2-access-review-export-20240630-v1.0.csv
EVD-A.5.24-incident-log-april-20240430-v1.0.xlsx
AUD-internal-audit-report-access-controls-20240520-v1.0.pdf
Rules worth enforcing:
Maintain a document register that doubles as your audit index:
# document-register.md (excerpt)
| Doc ID | Name | Owner | Approved | Next review |
|--------|-------------------------------------------|--------|------------|-------------|
| POL-001| Information security policy v3.0 | CISO | 2024-06-01 | 2025-06-01 |
| SOP-014| Access review procedure v1.2 | IT Ops | 2024-03-15 | 2025-03-15 |
| EVD-201| Q2 access review export v1.0 | IT Ops | 2024-06-30 | n/a |
Control A.5.18 (access rights) requires you to demonstrate provisioning, modification, and review of access. In your lab, build the pack in four steps.
On your own Linux lab host:
awk -F: '$3 >= 1000 {print $1","$3","$6","$7}' /etc/passwd > accounts-20240630.csv
cat accounts-20240630.csv
# Expected output (example):
# devuser,1001,/home/devuser,/bin/bash
# audituser,1002,/home/audituser,/bin/bash
getent group sudo
# Expected output:
# sudo:x:27:devuser
groups devuser
# Expected output:
# devuser : devuser sudo docker
Save the output as EVD-A.5.18-group-membership-20240630-v1.0.txt.
Walk through each account against the joiner/mover/leaver record. Produce a review sheet where each row has a decision and a reviewer signature (a name and date is fine in a lab):
# access-review-q2-2024.md (excerpt)
| Account | Expected? | Decision | Reviewer | Date |
|-----------|-----------|----------------|----------|------------|
| devuser | Yes | Retain | A. Learner | 2024-06-29 |
| audituser | Yes | Retain | A. Learner | 2024-06-29 |
| oldintern | No | REMOVE (exited 2024-06-14) | A. Learner | 2024-06-29 |
The removal itself must be evidenced, not just noted:
sudo userdel -r oldintern
awk -F: '$3 >= 1000 {print $1}' /etc/passwd | grep oldintern
# Expected output: (empty — account gone)
echo "oldintern removed 2024-06-29 per Q2 access review" >> /var/log/access-review-actions.log
Your finished pack: the SOP, the raw export, the review sheet, and the removal log. Four artefacts, one control, fully traceable.
Control A.5.24 requires a documented incident process plus evidence it operates. Trigger a controlled incident in your lab: deliberately fail three sudo attempts and capture what your logging picks up.
# As a normal user, generate the event:
sudo -k
sudo cat /etc/shadow # enter wrong password three times
# Then, as admin, extract the evidence:
sudo grep "authentication failure" /var/log/auth.log | tail -n 3
# Expected output (example):
# Jun 30 10:14:02 labhost sudo: pam_unix(sudo:auth): authentication failure;
# logname=devuser uid=1001 euid=0 tty=/dev/pts/0 ruser=devuser rhost=
Now log it through your incident procedure. Save the extraction as EVD-A.5.24-failed-sudo-detection-20240630-v1.0.txt, open a ticket (even a simple markdown ticket file works in a lab) with an ID, classification, and closure decision, and record the timeline:
# INC-2024-007.md
- Detected: 2024-06-30 10:15 (auth log review)
- Classified: false positive / test trigger — no escalation
- Evidence: EVD-A.5.24-failed-sudo-detection-20240630-v1.0.txt
- Closed by: A. Learner, 2024-06-30 10:22
The auditor’s chain is complete: procedure document → detection log → ticket → closure record. Even the “we decided this was nothing” decision is evidence the process runs.
Clause 9.2 requires internal audits, and they must be independent — you audit an area you weren’t responsible for. In a small team or a solo lab, that means rotating scope. Plan at least one audit per major control area per year:
| Quarter | Scope | Controls sampled |
|---|---|---|
| Q1 | Governance and policies | A.5.1, A.5.36, SoA currency |
| Q2 | Access control | A.5.15–A.5.18 |
| Q3 | Operations and incidents | A.5.24–A.5.28, A.8.15 logging |
| Q4 | Supplier and physical | A.5.19–A.5.23 |
Audits follow the same trail you just built: pick a control from the SoA, pull the policy, pull the most recent evidence, check dates and completeness, and write findings. Log each finding as a nonconformity with an owner and a due date — that record itself is mandatory (clause 10.2).
A lab-sized audit checklist entry looks like:
# internal-audit-q2-2024.md (excerpt)
Control: A.5.18 Access rights
- Policy exists and current? PASS (SOP-014 v1.2, reviewed 2024-03-15)
- Quarterly review performed? PASS (EVD-201, 2024-06-30)
- Leaver removed within SLA? PASS (removed 15 days after exit)
- Finding: access review uses manual CSV export; recommend automation. (Minor NC, due 2024-09-30)
Clause 9.3 mandates management review, and the standard lists the required inputs. Your minutes must show each was considered:
The outputs matter too: decisions on improvement, changes to the ISMS, and resource needs. In your lab, write the minutes as a structured document with one paragraph per input and explicit decisions, dated and “signed” by the accountable person:
# MGT-review-2024H1-v1.0.md (excerpt)
Input 4 (risks): Risk R-03 (single lab host, no offsite backup) unchanged.
Decision: accept for now; revisit after budget review Q3.
Input 5 (objectives): MTTD objective 30 min — met (avg 22 min, see EVD-205).
Output action: automate access review export. Owner: IT Ops. Due: 2024-09-30.
Note how the outputs feed the next audit and the next review — that closing loop is exactly what auditors check for in clause 9 and 10.
POL/SOP/EVD-A.x.y-date-vN so the mapping is self-evident.userdel above) generates new evidence for the next audit.