> ## Content Index
> Fetch the complete content index at: https://www.associateshandbook.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The three levels of errors, and the three layers that catch them
- URL: https://www.associateshandbook.com/the-three-levels-of-errors/
- Published: 2026-07-28T17:00:00.000Z
- Updated: 2026-08-07T12:02:47.000Z
- Author: Moritz
- Tags: ErrorFree, Concept

*A complete system for producing clean work. Three kinds of mistake, three lines of defence, and the habits that make both run without thinking about them.*

Errors creep into everything: PowerPoint, spreadsheets, meetings, the handover between case teams, a late-night version mix-up, or simply the hour when you were too tired to notice. You cannot defend against that with attention alone.

You need to know what you are defending against, and you need more than one net. This post covers both.

**Part one: the three levels of errors**

**First-level errors are careless mistakes.** Typos and spelling. Misalignment, inconsistent fonts, a colour that does not match. A duplicated footnote, a missing reference. The work stays correct, and it starts to look untrustworthy, which turns out to be almost as expensive. The aphorism does the work: if they cannot spell their words right, how can we trust the other insights?

These are easy to identify with a quick visual check, which is exactly why nobody forgives them.

**Second-level errors are process and calculation mistakes.** Subtracting where you should add. A decimal or a comma in the wrong place. A missing currency or unit conversion. A cell you forgot to fix, a hardcode buried in a formula, a key input you never updated. An email to the wrong recipient, or without the attachment, or with the wrong document.

These are more dangerous because they hide. Nobody spots them on a visual pass, and by the time they surface they have usually been built on for a week. Care will not catch them. Procedure will.

**Third-level errors are logic and judgement mistakes.** A market growth rate that is not physically possible. A supply and demand imbalance nobody questioned. An outdated fact from an expert everyone assumed knew better. These involve real thinking, industry knowledge and critical judgement, and no amount of proofreading will find them.

Two examples, both mine.

As an associate I was sizing market potential for an industrial rubber manufacturer, using industry reports as a baseline and validating selected countries through expert calls. One expert was enthusiastic about growth in India and convinced me to raise my estimates substantially, which made India a headline investment candidate. We were a small team and short on time, so we did not review the findings properly before the management update. The client’s CEO laughed and killed it with one question: “who is even supposed to produce that? There are not nearly enough facilities.” I had never sense-checked the expert from the supply side, and it made months of the team’s work look silly in one sentence.

Years later, as a manager, one of my consultants modelled the scrap copper market. The supply numbers looked reasonable on their own, so we built them into the model. Going deeper, predicted supply ran nearly four times predicted demand, which would mean more than 1.6 million tonnes a year quietly stacking up in American junkyards. Supply and demand have to roughly match, because you can only sell what someone buys. We had missed a large export flow, and it took several rounds of discussion to unpick.

**Part two: the three layers that catch them**

Three levels of defence catch the overwhelming majority of all three error types.

**Layer one: first time perfect.** Build quality in as you go. When you put a chart on a slide, set the axis title, fix the colour coding and sense-check the data right then. No “I will come back and clean this up later,” because later is when you are tired and the deadline has moved.

One caveat, because this can paralyse people. Do not let first-time-perfect block you. If you are stuck drafting bullets, write a rough version fast and iterate straight towards good. The goal is a habit of small continuous checks rather than perfection on the first line.

**Layer two: the internal loop.** When you finish something, review it yourself. Train your eye to screen your own output systematically across all three levels. After a year or two this becomes fast enough that you will spot most problems within seconds of opening a page.

Associates often skip this on the theory that the supervisor will check it anyway. That is true and it misses the point. Skipping your own review moves work from a less time-constrained person to a more time-constrained one, and it costs the case team more than it saves you.

**Layer three: the external loop.** Nothing leaves the team without a second pair of eyes. For anything client-facing this is non-negotiable: every slide, every client email, every model output gets reviewed by at least one other person.

The simplest way to organise it is to swap. When you finish your section, trade with a colleague who has finished theirs. Fresh eyes catch what yours cannot, particularly after four hours of staring at the same content. If time is short, narrow the scope of the review rather than skipping it. For important deliverables the loop escalates: you, your supervisor, the project leader, the partner, sometimes the client, and back to you.

**Which layer catches which error**

|                        | **First-level** | **Second-level**              | **Third-level**                 |
| ---------------------- | --------------- | ----------------------------- | ------------------------------- |
| **First time perfect** | Most of them    | Some                          | Rarely                          |
| **Internal loop**      | Nearly all      | Most, if you have a checklist | Some                            |
| **External loop**      | The remainder   | The remainder                 | This is the only reliable catch |

Read the bottom right cell carefully. Logic and judgement errors are almost never caught by the person who made them, because the flawed assumption is invisible from the inside. That is the entire argument for the external loop, and it is why “I checked it myself” is not an answer.

**Part three: the habits that make it automatic**

**Review in two passes.** The first pass is mechanical: spelling, formatting, alignment, references, sources, footnotes, number consistency. If the tagline mentions four steps, are there four steps on the page? The second pass is conceptual: does the logic hold, does the story make sense, would you be comfortable presenting this to the client? Mixing the two means missing both, because your brain switches context too often.

**Take a break before you review.** Work on something else, even for three minutes. If you have been inside a file for four hours you have gone blind to it, and distance is what lets you see the page rather than remember it.

**Know when your error rate spikes.** Stress, distraction, tiredness and a preoccupied mind all multiply mistakes, and stressful things outside work count. Plan extra review time in those windows. Do not press send just before lunch if you do not have to.

**Do not ship after midnight.** I have seen more mistakes in deliverables finished between 1 and 3 a.m. than at any other time. Concentration drops, pattern recognition weakens, and your tolerance for “good enough” quietly rises. If you finish something late, save it, sleep, and do a ten-minute pass in the morning.

**Keep an error log.** Note down, or screenshot, every error someone else caught. Your supervisor will usually point them out rather than correcting them, which is annoying at first and the fastest way to learn. Review the list regularly and think about how to stop each category from recurring. Reviewing and culling errors is a trainable skill, like public speaking.

**START THIS WEEK**

**1\.** Open a note called “Errors” and put the last three mistakes anyone caught in your work into it. Label each one first, second or third level. The distribution tells you which layer you are missing.

**2\.** Agree a swap partner. Find one person on your team at a similar level and agree that you review each other’s sections before anything goes up. This single habit removes most first- and second-level errors from your output.

**3\.** Write your own two-pass checklist for one artefact type, probably slides. Ten items maximum, in a fixed order. Use it on everything for two weeks and add whatever slipped through.

**Nothing should leave the team without a second pair of eyes.**