In two years, W3whiz's exception log has produced more than 300 written rules. Content makes up the biggest share, about 40%. The main lesson: reminders don't stop repeat mistakes, written rules do.
The log is the part of our delivery system that keeps improving it. Every mistake a senior catches is recorded, and anything that repeats becomes a rule, so it doesn't need catching again.
What the exception log is
The exception log is where every slip gets written down, so the same mistake doesn't need a senior to catch it twice.
It works in three steps:
- Log it. Every override, and every issue a senior catches that the template checks missed, gets an entry.
- Watch for repeats. One-off mistakes stay in the log. Anything that keeps coming back is a missing rule.
- Promote it. A recurring exception becomes a written rule in the definition of done, and where possible a check in the template.
The log sits next to our definition of done, the written standard for each deliverable type. The spec says what "done" means. The log is how the spec keeps getting better.
The numbers: 300+ rules in two years
We've written more than 300 rules in two years, and content accounts for the biggest share.
| Category | Share of rules | About how many |
|---|---|---|
| Content | ~40% | ~120 |
| Technical SEO | ~25% | ~75 |
| Templates & formatting | ~20% | ~60 |
| Links | ~15% | ~45 |
The pace wasn't steady. The first batch came from writing the original definition of done. Many more came in the early months, as the log surfaced gaps the spec had missed. Today it has settled at about five or six new rules a month.
Shares and counts are rough estimates from the log, not an exact audit.
Real rules from the log
Most rules are small and specific. That's why they work: a vague rule can't be checked.
Content (~40%)
- Follow the brief: target keyword, headings and length must match it.
- Answer the main question in the first paragraph.
- No unsupported claims: every stat or claim needs a source.
- AI first drafts must be edited by a person for facts, tone and the client's voice.
Technical SEO (~25%)
- Re-check after fixes go live. A fix isn't done because the ticket says so.
Templates & formatting (~20%)
- From our original review log: wrong template version, a metric labelled differently than last month, a partner's brand colour off-spec.
Links (~15%)
- No broken internal links in any deliverable.
- Outreach and citations only on the partner-approved site list.
What the log taught us
Rules beat reminders. Telling someone again didn't stop a mistake from coming back. Writing it down as a rule, or building it into a template check, did. If a mistake shows up a third time, the problem is the system, not the person.
Fewer new rules means the system is maturing. Early on, the log produced new rules fast because the spec had obvious gaps. Dropping to five or six a month isn't the log slowing down. It means most mistakes are now caught before anyone has to log them.
Start your own exception log
You don't need software. A shared sheet with four columns is enough:
| Date | Deliverable | What went wrong | Repeat? |
|---|---|---|---|
| e.g. Nov 3 | Monthly report | Wrong date range | Yes, third time |
Once a month, look at the repeats. Turn each one into a written rule, and add it to the template as a check if you can. Then watch the repeats drop.
If you want help setting up your log and your first definition of done, book 20 minutes.
The full story of how the log fits our system: the case study 30 to 37 Accounts With the Same Seven People, or We Said No to Good Revenue on LinkedIn.
