Predicting From a Picture
Task 4 returns the picture you have just described and asks what will most probably happen next. Describing it again is the single commonest failure, and it carries a hard ceiling.
The task and its one ceiling
Task 4 is Making Predictions: 30 seconds of preparation, 60 seconds of recording, and on the real test the same illustration you described in Task 3 comes straight back. The question is what will most probably happen next.
The failure this task is built around is re-description. An answer that says what is in the picture instead of what follows from it has answered the previous task twice, and it caps Task Fulfilment at 6. This is not a marginal deduction. It is a ceiling applied after everything else, so an eloquent, accurate, perfectly paced re-description still lands at 6 on that criterion.
The diagnostic is simple and worth applying to your own recordings. Count the verbs. If most of them are in the present continuous and describe things visible in the frame, you have described. If most of them are future or modal and describe things that are not yet visible, you have predicted. An answer that opens with in the picture I can see has usually already gone wrong in its first five words.
Tense discipline, and where the easy marks are
The temptation is to reach for will and use it for everything. He will drop the box. She will get wet. They will start arguing. That is grammatically fine and it flattens the entire answer into a single register of certainty, which is not how prediction works and not how the higher bands sound.
The available range is wider than most candidates use. Is going to for something already in motion. Will probably and almost certainly for the likely. Might and could for the possible. I doubt and it is unlikely that for the improbable. Would most likely when you are reasoning rather than asserting.
Grading your predictions by how confident you are is a top band marker under both Vocabulary and Content, because it shows you reasoning about evidence rather than reciting outcomes. The dog is almost certainly going to get to the sandwich before he does, though he might just catch it if he turns round now carries two different confidence levels in one sentence and reads as thought rather than recital.
Every prediction has to be traceable to the frame
A prediction that could have been made about any picture is not a prediction. A fire alarm going off, an ambulance arriving, somebody having a heart attack: these are inventions, and inventing content on the picture tasks caps Task Fulfilment at 8 on top of everything else.
The scenes are built with planted momentum. Something is balanced badly. Somebody is walking towards somebody else without looking. A queue is longer than the staff can handle. A drink is close to an elbow. Those are the arcs the picture is asking you to complete, and there are usually at least four of them.
The test of a good prediction is whether you can state the evidence for it in the same sentence. The duffel bag is going to slide off, because it is sitting right on the edge of an open boot and he has just put a second one on top of it. The because clause is not padding. It is what makes the sentence a prediction rather than a guess, and it also does real work under Coherence.
Likely, not dramatic
A well built scene carries mild momentum rather than catastrophe. Somebody gets soaked. A queue backs up. A bag of chips is lost to a gull. An ice cream lands on the pavement. If your answer turns the school pickup zone into a traffic collision with emergency services, you have stopped reading the frame and started writing fiction.
The word in the instruction is probably. It is doing real work. The task is a reasoning task disguised as an imagination task, and the reasoning is what is scored. The most likely outcome is usually mundane, and describing a mundane outcome precisely scores far better than describing a dramatic one vividly.
This also protects your answer from a second problem. Dramatic predictions tend to be one-shot: once the building is on fire there is nowhere else to go, and candidates who open with catastrophe often find themselves at second twenty five with the whole scene resolved. Mild predictions chain, because ordinary consequences have ordinary consequences.
Bank your predictions during Task 3, not during Task 4. On the real test it is the same illustration, so the thirty seconds of prep you get for Task 4 is better spent grading confidence levels than finding arcs you already saw a minute ago.
Predict motives, not just objects
Most candidates predict what will happen to things. The bag will fall. The door will close. The ice cream will melt. That is a complete answer and it stays flat, because physical outcomes are the easiest thing in the frame to see coming.
Predicting what somebody will do, and why, is a different order of answer. She is going to go down the slide, but I do not think she actually wants to. She keeps looking back at her father, and I suspect she is doing it because he is filming. That is a prediction about intention, and intention is abstract language sitting on a concrete observation.
It also opens up material when you are running short. Every person in the frame has a next action and a reason for it, and reasons are inexhaustible in a way that falling objects are not. If you have twenty seconds left, pick a person you have not mentioned and predict what they will do and why.
A worked answer, and the version that fails
The scene is a park on a hot afternoon: an ice cream van with a queue, a toddler carrying a cone at arm's length, a dog on a long lead, a man asleep on a bench with a newspaper over his face, and a group setting up a picnic near a sprinkler that has just come on.
The failing version. In this picture there is an ice cream van and many people are waiting in a line. A small child is holding an ice cream and a dog is nearby on a lead. A man is sleeping on the bench and some people are having a picnic. I think it is a nice sunny day and everyone is enjoying themselves. Every word of that is true, and none of it is a prediction. Task Fulfilment caps at 6.
The working version. That cone is not going to survive. The little girl is holding it out at arm's length and the dog has already noticed it, so I would say she loses it within about ten seconds, probably before her mother gets to her. The picnic group are going to have to move, because the sprinkler has only just started and it is swinging back towards them, though they might not realise for another minute. The man on the bench will almost certainly sleep through all of it. And I suspect the queue is going to get considerably longer, because the two families who have just arrived are heading straight for it.
Four predictions, four different confidence levels, every one traceable to something visible, and one of them about what somebody has not noticed rather than what will physically fall.
Attach a 'because' to at least two predictions. The evidence clause is what turns a guess into a prediction, it defends you against the invention cap, and it does work under Coherence at the same time.
Practise it
Five judgement questions on predicting forward from a picture you have already described.