Welcome to chapter twenty-one. A visually impressive model is useful only if it supports the educator's learning objective. That sentence is easy to agree with and hard to act on, because the instinct of a skilled maker is to make the thing better: more detailed, more realistic, more finished. For a teaching aid, better frequently means less. Fewer features, a more visible mechanism, a cruder appearance, and clearer relationships. This chapter is about designing the learning, and treating the object as the thing that serves it.
"Understand the elbow" cannot guide a build, cannot be assessed, and cannot fail. It is not an objective; it is a topic. An objective names an observable behaviour, performed by a specified learner, under specified conditions, to a specified standard. So "understand the elbow" becomes: after using the model, the learner will identify elbow flexion and explain that a tendon transmits force produced by muscle. "Learn about joints" becomes: given the model, the learner will predict the direction of movement before the cord is pulled, and justify the prediction by the attachment point. Each of those can be watched happening.
The test of an objective is its verb. Use verbs you can observe: identify, name, state, explain, describe, demonstrate, predict, compare, distinguish, justify, perform. Avoid verbs you cannot observe: understand, appreciate, know, be aware of, be familiar with, grasp. Those describe an internal state you have no access to. If you cannot see it happen, you cannot tell whether your teaching aid worked, and you cannot design an assessment for it.
Bloom's taxonomy is a standard hierarchy of cognitive demand. Remember: recall facts, like naming the three bones at the elbow. Understand: explain what a tendon does. Apply: predict the motion when the attachment point is moved. Analyse: explain why a shorter moment arm needs more force. Evaluate: identify what the model misrepresents and why it matters. And create: design a modification that teaches pronation. It is useful because it stops you writing a recall-level objective and then testing at the analysis level. If the objective says explain, a yes or no recall question does not assess it.
Miller's pyramid makes a different distinction: knowing is not doing. At the bottom, the learner knows the facts. Then knows how to apply them. Then shows how, in a simulated setting. And at the top, does it, in real practice. Knowing that a tendon transmits force is level one or two. Demonstrating the relationship on a model is level three at best. Doing the right thing with a real patient is level four, and no bench model establishes that. Be clear with yourself and the educator about which level your aid addresses.
Now fidelity, the word simulation people use for realism, and why it is not one thing.
Fidelity comes in at least six dimensions, and they are not interchangeable. Visual fidelity means it looks right, and it matters when the objective is recognition. Anatomical or spatial fidelity means the right structures in the right relationships, for locating things. Tactile fidelity means it feels right, for procedural skills guided by touch. Mechanical or functional fidelity means it behaves right: the motion, the force path, cause and effect. That is the elbow model's dimension. Environmental fidelity covers noise, time pressure and the team, and emotional fidelity covers whether the learner behaves as they would for real.
Choose your fidelity, and then deliberately sacrifice the rest. For a force-path objective, skin-coloured paint adds nothing. A transparent or open construction, which looks less realistic, teaches the relationship far better because the mechanism is visible. Reduced visual fidelity is not a compromise there; it is the right design. Conversely, a trainer for locating a structure by touch needs correct spatial relationships and plausible tactile behaviour, and a generic hinge would be useless however well made. So the question is never how realistic it should be. It is realistic in which dimension, and why that one.
Once the dimension is chosen, the making method usually follows, and it is rarely one method. Visual realism comes from moulage, pigmented silicone, vacuum-formed surfaces and dyed fabrics. Anatomical relationships come from parts cast or printed from imaging, or located by machined and laser-cut features. Tactile realism comes from layered silicones and gels, fabric and mesh membranes, foams and embedded tubing. Mechanical behaviour comes from turned pins and bushings, linkages, springs and cord. Instrumented feedback means sensors with a microcontroller, like the CPR manikin that reports compression depth. And notice how often the machined or printed part is a tool, a mould or a jig, rather than the thing the learner touches.
Apply that to the elbow model. A visible pivot: yes. A clearly labelled force input, distinct from the cord: yes, because that is the objective. A visible relationship between pulling and moving: yes. Exact bone contours and skin colour: no, and skin colour actively obscures the mechanism. Motorisation: no, because a motor hides the force source and would defeat the objective. An electronic angle readout is optional, only if the objective needs a number, and only if it adds feedback without supplying force or hiding the mechanism. Forearm rotation and the true tendon insertion are deliberately omitted, and must be stated. That table is the start of your limitations statement.
The object does not teach. The interaction teaches, and you can design it.
The pattern is predict, observe, explain. First, ask for a prediction: what will happen if I pull this, and why? That step surfaces what the learner already believes, including the wrong belief, and it is the step people skip. If they predict wrongly and then see the right thing happen, the mismatch is memorable in a way that watching a correct demonstration never is. Then they observe or operate it. Then they explain: what happened, what supplied the force, what did the cord do? The explain step is the assessment, and it should include naming something the model does not show, which checks whether they understood the model as a model.
Here is a teaching script built on that pattern. Before movement: point to the parts representing the upper arm and forearm, then predict what happens when I pull the cord, and why. During movement: watch where the cord pulls relative to the pivot, and what changes. After movement: which movement did you see, what supplied the force, what does the cord represent, and name one thing a real elbow has that this does not. And if the learner says the tendon contracts, do not correct them and move on. Record it. Then change the labels, the demonstration or the model, and try again with a different learner. Their error is data about your design.
This warning matters. It is almost irresistible to help. You built the thing, you want it to work, and a hesitating learner is uncomfortable. So you say, "and what's pulling it?", they say "the muscle", and you record a success. You have just recorded your own explanation as evidence of the object's effectiveness. If you prompt, record the prompt. Completed with two prompts is an honest and useful result; completed is not. It is the same distinction as chapter nineteen: assisted performance is a weaker claim than independent performance, and the record must say which you saw.
Keep two checklists, because a model can pass one and fail the other. The learning checklist asks whether the learner achieved the objective: did they predict, did their explanation match, did they identify a limitation, what misconception appeared, and what prompts were given. Its failure is a robust, beautiful object that teaches the wrong thing. The object checklist asks whether the thing survives use: setup time, reset time and method, damage, loose parts, cleaning, storage, visibility across a room, and who maintains it. Its failure is an excellent demonstration that takes six minutes to reset, and so is never used.
Reset time is the requirement educators care about most and mention least. A teaching session has a fixed length and a queue of learners. An aid that takes ninety seconds to reset between attempts, in a class of twelve, consumes eighteen minutes. That aid will be used once and then left in the cupboard, regardless of how good it is. So ask explicitly: how long have you got between learners? The answer is a first-class requirement.
The most likely harm from a teaching aid is teaching something false, so every simplification should be checked for what it might quietly teach. A single pivot might teach that the elbow is one hinge; state the omission of forearm rotation every time. A cord pulled by hand might teach that the tendon generates force; label the force input and the transmission separately. A guide loop placed for convenience might be taken as the real tendon route; label it as a model convenience. Rigid strips might suggest bone is uniform. And because the misconceptions you are most likely to build in are the ones you share, an anatomically qualified educator must review the anatomy, labels and limitations. If you have none yet, record it as pending.
The engineering bridge: an orientation to the materials and methods simulation uses.
If you move further into simulation, tissue simulants become a topic in their own right. Platinum-cure silicones in a range of hardnesses make skin and soft tissue. Gelatin, ballistic gel and agar make needle and ultrasound phantoms, cheap and perishable. Foams give bulk, but open-cell foam absorbs fluid. Textiles make membranes, wearable trainers and covers, and sewing gives strength and easy replacement. Glass and clear plastics make vessels and chambers, but acrylic crazes with some alcohol cleaners. Machined and laser-cut parts make frames and joints, with acetal and brass for low-friction pivots. Printed polymers suit complex shapes and moulds. And sensors with a microcontroller can measure what the learner does.
A caution. Realistic feel is an educator's judgement, not a durometer reading. The correct process is that the educator tells you what the learner must be able to feel and discriminate, you make two or three candidates, they compare them, and you record which they chose and why. Matching a published tissue stiffness is not the objective and can actively mislead, because the property that matters for a procedural skill is usually the contrast between layers, not the absolute stiffness of any one of them.
A worked example. The educator's objective: after using the model, a first-year learner will identify elbow flexion and explain that a tendon transmits force produced by muscle. That objective requires the model to make visible the motion, the force input, the transmission element as a transmitter rather than a generator, and the relationship between pulling and moving. It permits the model to omit everything else: appearance, exact anatomy, pronation, extension mechanics, and scale. The learner task follows: predict which way the forearm part will move and why, watch, then say what supplied the force, what the cord did, and one thing a real elbow has that this does not.
The likely result, honestly recorded: the learner understands flexion but thinks the cord supplies its own energy. That is a partial result, not a failed project. The diagnosis is that the force source and the transmission element are still visually merged. The response is to separate them further: a visible handle on the pull end, a distinct muscle element where the force originates, perhaps a short sewn fabric sleeve or a cast silicone segment in a contrasting colour, and a different colour, texture and label for the cord. Then check again with a different learner, because the first one now knows the answer.
For the exercises: write one observable objective and an assessment at the same cognitive level, and check that explicitly. Audit every feature of your model for the fidelity dimension it serves, and delete one that serves none. For each feature you keep, name two ways it could be made and say which better serves its purpose. Write and run a predict, observe, explain script with one willing learner, recording their prediction, explanation, every prompt and any confusion. Build both checklists, measure your reset time, and get the educator review, or record it as pending.
The limits. Anatomical accuracy is certified by a qualified educator, every time, not by you. Anything teaching a clinical procedure must follow the host educator's approved method, and the method comes first. One learner is a useful observation, not evidence of effectiveness. Learner satisfaction is not learning; "that was great" tells you about a reaction, not knowledge. And a model that demonstrates is not a model that teaches: whether the learner's explanation changed is the question, and it has to be asked. Next, building and reviewing your first project.