ELECTRIC ICARUS PROJECT / NOVA ACADEMY / FOUNDATIONS OF INQUIRY
LESSON 010 · UNCERTAINTY
Thinking Clearly When We Don’t Know Enough
Yesterday’s strategy worked. Tomorrow’s conditions are unknown. How should we decide what to do when evidence is incomplete—and how do we distinguish a careful estimate from a confident guess?
FREE · BEGINNER · 45–55 MINUTES · NO ENROLLMENT · OCTOBER 29, 2026
What you need: paper or a notes app, a pencil, and about fifteen minutes for the mini-lab. No software or previous statistics course required.
Previously: Lesson 009 — Adaptation: Learning to Change Without Losing Continuity.
The question after adaptation
Lesson 009 showed that a route which worked yesterday may fail when conditions change. But the traveler does not get tomorrow’s map in advance. The next task is to reason honestly from what is known, what is estimated, and what remains unknown. Uncertainty is not a defect to hide; it is information about the limits of a claim.
Learning objectives
- Distinguish uncertainty, variability, error, and missing information.
- Interpret a simple probability without treating it as a promise.
- Compare predictions with observations using a small, invented data set.
- Calculate a basic expected cost and explain its assumptions.
- Write a conclusion that separates evidence from interpretation.
1. Four different reasons we may not know
- Missing information: a fact has not been measured or recorded. We may reduce this by gathering data.
- Measurement uncertainty: a reported quantity has a range of values reasonably attributable to it because measurement is imperfect. This is not simply a synonym for a bad instrument.
- Variability: outcomes differ across occasions or individuals even when the process is understood.
- Model uncertainty: the assumptions or structure of a model may fail to represent the situation adequately.
These categories can overlap. The National Institute of Standards and Technology (NIST) explains why measured values need uncertainty statements. A weather forecast, by contrast, expresses uncertainty about a future event. Do not treat a measurement interval and a forecast probability as the same thing.
2. Probability needs a clearly defined event
In a simple mathematical model, probability assigns a number from 0 to 1 to an event. Zero means impossible within that model; one means certain within that model. A probability of 0.30 means 30%, or three chances in ten—not that the event must happen exactly three times in every ten attempts.
For example, the U.S. National Weather Service defines probability of precipitation as the likelihood that measurable precipitation (at least 0.01 inch of liquid equivalent) occurs at a particular point during a specified forecast period. A 30% chance does not mean rain will cover 30% of the area, fall for 30% of the day, or be 30% intense.
A useful probability statement answers four questions: What event? Where? Over what period? According to what evidence or model? For introductory probability language and sample spaces, see OpenStax Contemporary Mathematics, section 7.5.
3. A prediction is not a verdict
Suppose a forecast assigns a 70% chance of rain, and the day stays dry. That outcome alone does not show the forecast was dishonest or wrong. If many comparable 70% forecasts are well calibrated, rain should occur on roughly 70% of those occasions over the long run. A single event cannot establish whether that relationship holds.
Calibration asks whether outcomes occur at about the frequencies assigned by forecasts across many comparable cases. Accuracy is broader and depends on how performance is defined. A forecast that always says 50% may be reasonably calibrated in a perfectly balanced sample while giving little useful discrimination between different days.
4. Mini-lab · The forecast notebook
15 MINUTES · INVENTED DATA · PAPER ONLY
The following twenty hypothetical days are a classroom exercise, not actual weather observations or a forecast for any location. Each letter is an outcome: R = measurable rain; D = dry. We have ten days with a 20% forecast and ten days with an 80% forecast.
| Forecast group | Ten invented outcomes | Rainy days | Observed frequency |
|---|---|---|---|
| 20% chance | D, D, R, D, D, D, R, D, D, D | 2 of 10 | 20% |
| 80% chance | R, R, D, R, R, D, R, R, D, R | 7 of 10 | 70% |
Your investigation
- Count the rainy outcomes in each row without looking at the final two columns.
- Divide each count by ten and convert to a percentage.
- Compare the observed frequency with the forecast percentage for each group.
- Does the second row prove the 80% forecast was poor? Explain why ten cases are too few for a confident verdict.
- Imagine the next ten 80% forecasts have nine rainy outcomes. What is the combined frequency across twenty such forecasts?
Answer key
The 20% group has 2 rainy days out of 10, or 20%. The 80% group has 7 rainy days out of 10, or 70%, which is ten percentage points below the forecast. With nine rainy days in a second group of ten, the combined 80% group has (7 + 9) / 20 = 16 / 20 = 80%. These invented results illustrate why small samples fluctuate; they do not establish that a real forecasting system is calibrated.
5. A decision can be rational without being certain
Consider another invented example. Carrying an umbrella has a convenience cost of 1 point whether it rains or not. Going without one costs 0 points if dry and 6 points if it rains. Assume, unrealistically but simply, that the umbrella eliminates all rain-related cost.
With a 30% rain forecast, the expected cost of going without is (0.30 × 6) + (0.70 × 0) = 1.8 points. Carrying the umbrella costs 1 point, so it has the lower expected cost in this toy model. With a 10% forecast, going without costs (0.10 × 6) = 0.6 points, lower than carrying it.
These are model-based comparisons, not instructions for safety-critical choices. Real decisions may involve severe consequences, accessibility, forecast errors, changing conditions, and preferences that are not represented by points. Expected cost is one tool, not a substitute for judgment.
6. Uncertainty does not mean every claim is equal
Evidence can support one explanation more strongly than another without making either absolutely certain. Saying ‘we are not sure’ should lead to better questions: What would change our mind? What data would discriminate between competing explanations? How reliable is the source? How costly would an error be?
Be equally careful with the opposite mistake: a precise-looking percentage can create false confidence if the model, data, or event definition is weak. Good inquiry makes its assumptions visible.
7. Where philosophy enters
Nova Academy’s practical framing is that intellectual humility can be a form of strength: we can act on the best available evidence while remaining willing to revise. That is an educational and philosophical interpretation, not a scientific law establishing that all uncertain decisions have one correct response. The mathematical examples above are illustrative models; the values were chosen for teaching.
8. Knowledge check
- A forecast says 40% chance of rain. Does that mean rain for 40% of the day?
- What is the difference between missing data and measurement uncertainty?
- Why is a single dry day not enough to reject a 70% rain forecast?
- What is the observed rainy-day frequency for seven rainy days in ten?
- Name one assumption in the umbrella decision model that might fail in real life.
Check your answers: (1) No; the percentage concerns the chance of a defined precipitation event at a location during a period. (2) Missing data have not been collected; measurement uncertainty describes dispersion associated with a measured result. (3) A 70% forecast still allows dry outcomes. (4) 70%. (5) For example, that the umbrella prevents all rain-related cost, that the forecast is reliable, or that costs can be reduced to simple points.
9. Field note · What would change your mind?
Choose one low-stakes claim you currently believe. Write what you know, what you are estimating, and what remains unknown. Name one observation that would strengthen the claim and one that would weaken it. Finish with a sentence beginning: ‘Given the evidence available today, I think … but I would revise this if …’
The goal is not to eliminate uncertainty. It is to stop uncertainty from disguising itself as certainty.
Sources and further reading
- NIST — Measurement Uncertainty (measurement concepts).
- National Weather Service — What Does Probability of Precipitation Mean? (rain probability at a point over a defined period).
- OpenStax — Basic Concepts of Probability (events and probability).
- Nova Academy Lesson 009 — Adaptation (previous lesson).
Continue learning: Nova Academy homepage and full lesson index. Free, self-directed, non-accredited education.
