See the hole, not the donut: finding what matters in an information flood

1. Information is not yet understanding

Imagine a large report: dozens of tables, hundreds of metrics, many conclusions. You can retell its contents and still not understand the situation.

It helps to separate levels:

1. Data — individual observations, numbers, events, and facts.
2. Information — data organized so that they communicate something.
3. Understanding — an explanation of how facts relate and why they matter.
4. Judgment — an assessment of what to do given goals, risks, and uncertainty.

“Support tickets rose 30%” is information. Understanding starts with questions: why did they rise, which tickets count, did the measurement change, and did growth translate into more customers?

Judgment needs the next step: what does this mean for the business, and which action is justified?

The more data we get, the more important it is not to confuse these levels. Extra facts can deepen understanding — or hide the absence of a clear question.

2. The donut metaphor: noticing the shape of absence

“See the hole, not the donut” invites attention not only to the object, but to the space around it: gaps, boundaries, exceptions, and assumptions.

We usually notice what is present: a number in a table, an argument in a deck, an experimental result, an expert claim. Sometimes the key sits in what is missing:

• The report shows revenue, but not profit.
• The study reports an average, but not spread or sample size.
• The product page lists benefits, but not limits.
• The plan has deadlines, but not dependencies.
• The meeting reaches agreement, but no one checked the premises.
• The stats show successes, but not how many attempts failed.

Absence is not always a problem. Sometimes data truly do not exist, sometimes they are unnecessary, sometimes they were simply never collected. The point is not to jump to conclusions, but to turn a noticed gap into a testable question.

A practical move: after any important material, ask three questions:

1. What is shown here?
2. What is not shown that could change the conclusion?
3. Under what conditions would this conclusion stop being true?

The third question is especially useful: it reveals the bounds of an idea, not only its strengths. In ThinkLens, a similar move is to separate claims, evidence, and gaps before arguing with the author.

3. Start with the question, not the information

One of the most common mistakes is gathering material before deciding what it is for.

Someone opens dozens of tabs, reads articles, compares opinions, and saves links. An hour later they have more information — and not necessarily more clarity.

State the task in one sentence:

• What decision do I need to make?
• What hypothesis am I testing?
• What uncertainty do I want to reduce?
• What fact could change my current conclusion?

The last question matters most. If no possible answer would change the decision, more research may be unnecessary — or the question is too abstract.

Instead of “study the AI tools market,” try: “Choose a tool for daily work on a large TypeScript project within a fixed monthly budget.” Now you know what you need: codebase quality, cost, limits, privacy, IDE integration, and reliability.

A good question acts as a filter. It stops every interesting detail from posing as important.

4. Signal and noise: why noticeable is not always important

Information grabs attention for many reasons. It can be new, emotional, unusual, easy to grasp, or confirmatory of what we already believe. None of those traits alone makes it significant.

Signal is an observation that improves understanding or decision quality. Noise distracts from the task or creates a feeling of understanding without improving it.

Checks that help:

• Link to the goal. Does the fact help answer the question you set?
• Impact on the decision. Would your action change if this fact were different?
• Reliability. Where did it come from, how fresh is it, and how was it obtained?
• Repeatability. Is this a stable pattern or a one-off?
• Cost of error. What happens if you misread this signal?

These checks are not perfect, but they allocate attention more wisely.

Remember the reverse too: a quiet, rare, or subtle signal can be critical. A single anomaly in an industrial safety system may not prove a serious fault — but you should not ignore it only because it is rare. Significance comes from context and consequences, not volume.

5. Look for disconfirmation, not only confirmation

Thinking tends to protect opinions already formed. We notice supporting arguments faster, remember confirming examples more easily, and judge contradictory facts more harshly.

So deliberately reverse the search.

Instead of “Why is my idea right?” ask:

• What observation would show I am wrong?
• How else could the same facts be explained?
• What would a competent skeptic say?
• What fact would convince me if I held the opposite view?

This does not mean doubting everything equally. It means testing important assumptions in proportion to the cost of being wrong — in the spirit of a scout mindset: map the territory, do not defend your map.

A practical habit: write next to each key hypothesis what would falsify it. For example: “I think conversion dropped because of the new UI. I will check whether the drop timed with the release and whether it appears for users still on the old version.”

The assumption becomes a testable hypothesis.

6. Look at relationships between facts

A single fact can be true and still mislead without context. Meaning often lives in how several measures relate.

A company reports 20% revenue growth. That sounds good. If costs rose 35%, customer count barely moved, and growth came from one large one-off deal, the picture changes.

Look for:

• Comparison: against what is the metric set?
• Dynamics: how did it change over time?
• Proportions: what grows faster or slower?
• Distribution: is the result the same across groups?
• Causal links: what might have caused the change?
• Limits: which factors are left out?

Correlation does not prove causation. Two measures can move together because of a third factor, chance, or how data were collected. Causal claims need more: experiment, natural comparison, timing, and a plausible mechanism. More in our note on correlation and causation.

7. Do not confuse the average with the typical

Averages are handy and can hide important differences.

If five people take 10, 11, 12, 13, and 54 minutes on a task, the average is 20 minutes. Almost nobody takes exactly that long. One unusual case pulled the number.

Ask:

• What do the median and range look like?
• Are there outliers?
• Do groups differ?
• How many observations sit beyond the average?
• What is being averaged, and is it measured the same way?

The same logic applies to ratings, studies, performance metrics, and quality scores. A summary number is useful for overview; it does not always explain what is going on.

8. Notice base rates and the denominator

Percentages and vivid examples mislead when the denominator is unknown.

“Errors doubled” sounds alarming. From one error to two per million operations is different from one thousand to two thousand at the same volume.

“90% of users are satisfied” is incomplete until you know how many answered, how they were selected, and who did not take part.

Before interpreting any percentage, clarify:

1. What is the absolute count?
2. What is the comparison base?
3. How were observations selected?
4. Did the measurement method change?
5. What is the base rate in comparable conditions?

That helps you neither exaggerate rare cases nor underweight mass effects.

9. Separate fact, interpretation, and assumption

In discussions these three layers often blur.

For example:

• Fact: after the app update, support tickets rose 15%.
• Interpretation: the update may have worsened the experience.
• Assumption: navigation changes caused it.

The fact may be data-backed, the interpretation reasonable, and the assumption still untested. If you do not separate them, a working hypothesis becomes “the known cause.” The same move appears in our series on claims and evidence layers.

Label confidence explicitly:

• confirmed by data;
• likely;
• possible;
• unknown;
• needs checking.

That does not weaken analysis. It shows where knowledge ends and guesswork begins.

10. Use “what would change my mind?”

Not every uncertainty needs more research. Sometimes new information is costly, slow, or barely affects the choice.

You are choosing between two technologies. One is a bit faster; the other is easier to maintain. If the speed gap does not affect users and you will maintain the system for years, another hour of performance testing may help less than estimating support cost.

Ask before you keep searching:

• What decision would I take under each possible result?
• How likely is a new fact to change the choice?
• What does obtaining that fact cost?
• What are the consequences of a wrong decision?

This is close to the value of information: data are especially useful when they can change an action and improve expected outcomes.

Sometimes it is better to decide with stated uncertainty than to gather forever for the illusion of certainty.

11. Read in several passes

You do not need equal attention on every page. For large documents, studies, and reports, a staged approach works well.

Pass one — map. Structure, headings, conclusions, charts, sources. Does the material answer your question?

Pass two — arguments. Find key claims and what data they rest on. Separate evidence from opinion.

Pass three — gaps. Look for limits, alternative explanations, excluded cases, and conditions under which conclusions fail.

Pass four — application. Note what changed in your understanding, which decision it supports, and what still needs checking.

This spends deep attention where it matters instead of spreading it evenly. For daily reading habits, see the series note.

12. Turn reading into notes that help you think

Saving a link or highlighting a paragraph is not the same as extracting useful knowledge. A good note should help you recover meaning and apply it later.

For important material, try five points:

1. Main claim: what does the author assert?
2. Grounds: what data or arguments support it?
3. Limits: what does the material not prove?
4. Link: how does this relate to what you already know?
5. Implication: what should you do, check, or revise now?

If a note answers no practical question, it may only archive information instead of turning it into understanding.

Do not try to write everything down. Note value often comes from preserving the logic of an idea, not from length. See also the note-taking guide.

13. Apply the method to everyday situations

When reading news

Do not stop at the headline. Find the primary source, date, raw data, and what is actually confirmed. Ask what facts are missing and whether the headline swaps probability for certainty.

When analyzing business metrics

Do not treat one KPI in isolation. Compare it with costs, quality, retention, and long-term effects. Check whether one metric improved at another’s expense.

When reading a scientific study

Look at sample size and composition, measurement method, control group, uncertainty, and authors’ limitations. One effect in one study is not always a universal law.

When discussing a problem

Separate observed facts, explanations, and emotional judgments. Clarify what the dispute is really about: data, causal model, goals, or preferences.

When working with AI

Do not treat a smooth, confident answer as proven. Ask for assumptions, alternatives, sources, and unknowns. Verify facts that the decision depends on. AI can help find links and frame hypotheses; rhetorical polish does not guarantee truth — see also how to read AI summaries.

14. A short checklist for finding what matters

Before you conclude or decide, walk through this list:

• I know which question I am answering.
• I separated facts from interpretations and assumptions.
• I checked the source, date, and how the data were obtained.
• I looked at context, comparison base, and dynamics.
• I looked for missing information and alternative explanations.
• I considered what would falsify my conclusion.
• I assessed whether new information would change the decision.
• I marked what remains unknown.

You do not need every item for every everyday question. Depth of checking should match the cost of error. Choosing a film needs one level of analysis; medical, financial, engineering, or organizational decisions need much stricter grounds.

Closing: what matters often sits at the edge of the visible

Working with information is not a contest of pages read or a drive to know everything. It is a discipline of attention.

A strong analyst notices not only what confirms the picture, but what does not fit. They separate fact from explanation, do not confuse loudness with importance, check alternatives, and know when there is enough data to act.

The donut metaphor points to a simple principle: you cannot fully understand an object by looking only at its solid part. Sometimes shape is defined by emptiness — a missing element, an unasked question, a hidden premise, or an uncounted constraint.

So when you study any complex topic, look in two directions: at what the information says, and at what it leaves outside attention.

What matters is not always the biggest number, the brightest fact, or the most confident conclusion. Sometimes what matters is the question nobody has asked yet.