One weak link still slows the bad news
This is one result from the free test Is Evidence Keeping Up With Your Output? It is the mixed reading, and it is the most common honest one. Nothing about your checking is broken. Some problems reach you while the work is still easy to change, and others reach you only in particular conditions or after somebody has started using it.
Mixed is a state, not a missing answer
Real checking is uneven. You verify the number and skim the reasoning. Your automated checks are excellent for the situations someone listed and blind to the ones nobody did. Your review is fast on weekdays and absent in August.
That unevenness is not carelessness. It is what a system looks like when different parts were built at different times for different reasons. The thing worth naming is not the unevenness. It is that the whole system performs at the level of its slowest part, and you have probably been judging it by its best part.
Why the chain does not average
The test reads five links between your work and useful bad news. Something outside the work must be able to disagree with it. The check must reach situations nobody listed. The warning must arrive while the work can still change. It must appear where the work gets edited. And it must start without someone choosing to start it.
These are sequential, not additive. A problem has to pass through all five to become a correction. An excellent independent oracle that only ever sees three listed cases stops at the second link. An outstanding automated check whose output lands in a dashboard nobody opens stops at the fourth.
This is why a strong link cannot compensate for a weak one, and why "we have good tests" and "we have great reviewers" are answers to the wrong question. The question is which link is narrowest.
Two teams that look identical from outside
The first team has a solid reference dataset, checks that run on every commit, and warnings that appear directly in the editor. Their gap is coverage: every scenario in the suite came from the original spec, so a customer using the product in an unanticipated way is still the first to find that failure.
The second team has broad, imaginative test cases including situations nobody planned for. Their gap is location: results land in a nightly report that one person reads on Monday, so a Wednesday failure spends five days between discovery and the person who could fix it.
Both teams would describe their checking as good. Both are limited by exactly one thing, and it is a different one.
Find the narrowest link before you improve anything
The principle: effort spent anywhere except the narrowest link buys nothing. Improving a strong link feels productive, is easier to justify, and moves the actual outcome not at all.
Take the last serious mistake that reached someone else, and walk it backwards through the five links. At which one did it stop?
- Could any source outside the work have disagreed with it? If not, the break is independence.
- Was the situation one nobody had listed in advance? If so, the break is coverage.
- Did the signal exist but arrive after use? The break is timing.
- Did the signal arrive on time but somewhere other than where the work gets edited? The break is location.
- Did the check exist and simply not run that time? The break is how the check starts.
One incident is a weak sample. Three give you a pattern, and the pattern is usually blunt: the same link, three times.
Then fix one link, and only one
Choose the narrowest link and make one change to it. Not a program. One change, on the next piece of work.
If it is independence, write down the correct value of the decisive fact, from a source that had no part in the work, before you look at the output. If it is coverage, add one believable situation that appears nowhere in the starting instructions. If it is timing, name the latest moment a check result can still change the work. If it is location, put the failure next to the part it affects, in the tool used to edit it. If it is how the check starts, attach it to a step that already happens.
Then retake the test on your next piece of work. If the narrowest link has moved to a different one, the change worked, and your next move is at that new link. If it has not moved, you fixed something that was not the constraint.
Where this profile ends
This is not the result to act on with a full quality program. Broad initiatives spread effort evenly across five links, which by definition puts most of it where it cannot help.
It also does not mean you are close to safe. A chain with no broken link and no strong link still delivers bad news slowly. The distance from here to Bad news can reach the work while it still helps is usually two or three specific changes, made one at a time, each at whichever link is narrowest when you make it.