Keep this light. Hold up one group's first-trial reading from earlier in this project (or invite a volunteer to name theirs) and ask the class whether they would trust a big decision on a single number. Do not re-stage the full apparatus yet.

Nature of STEM weave: scientists repeat and share results so one lucky or unlucky run does not become the story. Met Éireann checks rainfall and temperature readings more than once before a forecast is trusted; Irish food producers and engineers do the same with cold-chain and material checks.

Have groups sit with the same partners as earlier in this project and open their FairTestPlanner and first DataTable page so the plan is in front of them. The measure or fixed observation window for today's repeats should already be written on each planner from your before-lesson prep.

Keep evaluate off the board for now. Pupils meet it in Step 4 when they judge whether the prediction held. You can still use the word aloud in the worked cycle below.

Worked cycle to model aloud (one volunteer group's first-trial story): "I wonder if our first reading was a one-off. We predicted foil would keep the ice longest. We keep the same amount of ice, the same starting place and the same timer. We run Trial 2 and Trial 3, write each time, then take the middle value from trials that used the same measure. I observed the middle value still favours foil, so I think our prediction held — but next time I would start every cube from the freezer at the same moment so the starts match more tightly."

Misconception to head off: children may treat the first trial as the 'real' answer and the repeats as optional. Stress that the middle value (or a clear pattern across matching trials) is what we trust, even when it disagrees with Trial 1.

Before the step

Return each group's kit from earlier in this project: the same cups or wrappings, the same timers, and fresh ice cubes from the same tray or warm water poured by you only. Groups keep their own question (keep ice frozen longest, or keep water warm longest).

Endpoint checks are finished before this lesson (see Before the Lesson). Every group's FairTestPlanner should already name today's measure or fixed observation window. Do not open a round of plan conferences now. Start straight on set-up and Trial 2. If a planner was missed, set that one group to the class default short window in under a minute and write it on their planner, then move on.

Same-measure rule (say aloud once before they start): only trials that used the same measure and the same window go into the middle value or pattern later. If Trial 1 was a full melt or a different window, leave Trial 1 as a note on the DataTable and use Trial 2 and Trial 3 only for the trustworthy number. Prefer, whenever the project sequence allowed, that Trial 1 already used today's short window so three matching trials are available.

During the step

• Remind groups to change only the variable they named on their FairTestPlanner and to keep amount, starting temperature, place and timing the same.
• Raw trial numbers live on the DataTable page only in this step. Middle value, pattern and conclusion come later on the Results and Conclusion page.
• Circulate and ask: "Is this trial the same as the other repeats in every way that matters?"
• Pupils who should not handle ice or water can be the timer or the recorder.
• If a group's first trial used hand-feel instead of a thermometer, keep the same judge and the same finger for every repeat so the comparison stays fair.

Look-fors

• Groups reset the set-up fully between trials rather than rushing.
• Readings are written at once on the DataTable, not remembered later.
• Disagreement between matching trials is treated as useful, not as failure.

Safety: you pour any warm water — warm to the hand, never hot. Wipe spills at once. No tasting of ice or water from the test.

This is the paper Investigation Journal beat. Pupils use the ResultsAndConclusion page only. Nothing is typed into the platform. Raw trials should already be on the DataTable from Step 3; this page is for the tidy matching-trial list, middle value or pattern, conclusion and one improvement.

Board model before they write (about 2 minutes): order three sample numbers that share one measure, circle the middle one, and say aloud: "The middle value is what we trust." Keep middle value for three ordered matching trials only.

Two-trial aside (only if a group has two clean matching trials, e.g. window changed so Trial 1 does not count): do not call the result a middle value. Say: use both numbers and say if they are close or far; if the class chart needs one number, share the halfway number (add them, then halve — 10 and 14 make 12). Do not invent a third trial.

Same-measure rule (repeat once): if the measure or window changed for the repeats, use only Trial 2 and Trial 3 and follow the two-trial path. Trial 1 must not be mixed with later trials unless the endpoint matches.

Evaluate (today's third word, now it earns its place): one plain sentence tied to the checklist: "Evaluate means decide from the evidence whether the prediction held, and what to improve."

Sentence stems you can offer aloud: "Our middle value shows… so our prediction did / did not hold because…" or "Our two matching trials show… so…" and "Next time we would keep ___ the same more carefully."

What good looks like: only matching-trial numbers copied from the DataTable, a middle value or stated two-trial pattern, a conclusion tied to those numbers, and one concrete method improvement (not "try harder").

Drive the data-recorder on the IWB in explore mode. Rows are groups (or named wrappings if several groups tested the same kind of question). The last numeric column is each group's trustworthy number (middle value from three matching trials, or the halfway number / pattern figure from two matching trials) — that is what the bar chart will draw. There are 10 rows so a full class of groups can appear on one chart; leave unused rows blank if you have fewer groups.

Exactly what appears on screen: columns Group, What we measured, and Middle value; ten rows; a bar chart. Class average is off by default. Only switch the average on if every bar shares the same unit and the same observation window. If groups measured different things (minutes to melt versus minutes still warm, or different fixed windows), keep units clear in the "What we measured" cell and only compare bars that share a unit — chart ice-wrap groups as one run, then warm-cup groups as a second run. Announce which run is on screen before typing. Never leave a single class average mixing unlike quantities on screen.

Ask the watchers: "Are they reading the trustworthy number from matching trials, or the first trial? Which bar is tallest in this run, and what does that tell us?" Do not invent a separate desk task for the rest of the class.

Prompt when the chart is up: evidence over opinion — "What does the chart show, not what did we hope?"