Return to the lesson

Educator guide: The Evidence Desk

This guide is for educators, carers and independent older learners. The lesson targets approximately age 15, with adjustments for prior experience. A short adult walkthrough and automated checks can establish that the software works; they do not establish learning gains or universal age suitability. No children were recruited or tested.

Purpose and prerequisites

Allow about 10–15 minutes for a first investigation, with optional deeper exploration. Useful prerequisites are reading a simple line chart and understanding an average. Percent change and correlation coefficients are explained through the case; formal algebra is not required to complete the core lesson.

Setting and provenance

Northbank Flight Fair, the newsletter and all headlines are fictional. Every record is synthetic and original to this lesson. There are no measured flight outcomes or real people, organisations, personal attributes, medical claims or political claims. Do not interpret the flight-time patterns as scientific findings or aerodynamic advice.

The source package contains a canonical datasets.json, four CSV tables and DICTIONARY.json. The browser bundle is generated from that JSON. Graphs, statistics and accessible tables all use those same records. The construction deliberately makes the teaching points visible; actual investigations are less tidy.

Suggested walkthrough

1. Frame the graph — about 3 minutes

Start with the two identical views. Use “Demonstrate a tighter view”, then change the limits yourself. Read the table: the first and last values are 10.2 and 11.0 seconds. The difference is 0.8 seconds, about 7.8%, regardless of the axis. Select the scale and apparent steepness as the things that changed, then record an observation.

A non-zero axis is not automatically deceptive. A labelled zoom in a line chart can clarify small differences. The choice should serve the question and match the claim. In the final activity, however, interpreting the heights of cropped bars as absolute ratios is misleading.

2. Question the sample — about 3 minutes

The complete population has 60 invented trials: 20 front-table long-wing trials and 40 side-rack short-wing trials. The front group repeats 11,12,13,14 seconds; the side group repeats 6,7,8,9. The population mean is 9.1667 seconds and its long-wing share is one third.

The front-table procedure samples only from those 20 records. Random tickets sample from all 60. Both shuffle with the same documented seeded procedure and select without replacement. Using the same method, size and seed reproduces a sample exactly. The seed is a reproducibility tool, not a hidden learner identifier. A lightweight finite pseudorandom generator simulates equal-chance selection; it is not a claim of true randomness or a validated random-number service.

Try 5 and then all 20 front-table records. The larger biased sample still excludes short-wing trials; all 20 have a mean of 12.5 seconds. Then try random tickets and successive seeds. Random sampling gives every record a chance, but one random sample need not exactly match the population. These controls demonstrate method and variation, not an assurance that random data are always balanced.

3. Look behind the link — about 3 minutes

The separate correlation dataset has 12 short-wing and 12 long-wing records. Within each design, symmetric patterns give zero linear correlation; pooling the groups produces a strong positive association. Toggle the groups while keeping both axes fixed. Some plotted points overlap, and all records remain visible in the accessible table.

The constructed data suggest that design could account for the pooled relationship between practice and flight time. They do not prove a specific causal explanation, rule out all other factors, or show that practice never matters. A fairer follow-up could hold design and launch conditions constant and randomly assign practice duration. A coefficient near zero rules out neither nonlinear relationships nor other dependence.

4. Write the fair claim — about 3 minutes

This is a new fictional case. Ten long-wing volunteers have paired before/after means of 12.0 and 12.6 seconds, an increase of 0.6 seconds or 5%. The headline’s axis begins at 11.4 seconds, making the visible bar heights appear to double. Switching to the full scale leaves the source values untouched. The other 40 short-wing registrations were not measured; there is no control or random assignment.

Ask for a cautious claim, actual supporting values and at least two limitations. The guided builder accepts any pair among axis framing, sample coverage and causal evidence. The alternate writing route uses an explicit self-review checklist. The app saves that writing but does not score, understand or certify it. The exemplar is one defensible response, not the only acceptable wording.

Differentiation and co-play

Feedback and privacy

Feedback evaluates only predefined selections. Incorrect choices receive a specific explanation and can be changed without limit or penalty. No leaderboard, pressure mechanic or child ranking is present. Notes and progress exist only in the current tab’s memory; refresh or reset clears them. Notebook export is a local text-file download. There are no accounts, forms sent to a server, analytics, advertisements, purchases, chat or generative-AI calls in the lesson. The hosted preview necessarily serves ordinary static files; the application does not collect learner information.

Educational references

These references were checked on 14 September 2026. They support the statistical ideas, not a claim that this specific lesson has been validated. No external reference is required to run the lesson, and outbound sources are kept in this educator-facing guide.

Offline follow-up

Invent six values and sketch them twice with different scales. Ask another person for one conclusion that remains true for both drawings and one question those values cannot answer. For a sampling extension, make 20 paper tickets in two categories, place one category at the front of a tray, and compare a front-only sample with a shuffled sample. No personal data need be collected.

Licence and attribution

The original lesson code, explanatory text, synthetic data and code-drawn graphics are released under the MIT licence in the source archive. AI assistance was used in authoring; no AI service runs in the learner experience. No third-party media, chart library, font files or educational examples are bundled. The interface uses system fonts available on the learner’s device. Reference authors do not endorse this lesson.