Bolt Series · School–Teacher–Student Performance Calibration · Article 17
Wait, What? Better Performance While Help Is Present Can Hide a New Dependency
A teacher gives a cue every time a student hesitates. Accuracy rises. The lesson looks successful.
Then the cue disappears and the old errors return.
The supported performance was real. But the school and teacher were measuring a learner-plus-support system, not yet the learner’s independent performance.
Bolt’s question is therefore precise: did the support improve future capability, or did it only improve performance while the support was carrying part of the job?
Quick Answer
Owned Bolt calibration job: detect when teacher, coach or technology support makes current performance look stronger than the learner’s later independent performance, and measure whether dependence is reducing across repeated return tasks.
This is not an argument for withholding help. Novices often need strong support. The calibration error is treating supported success as if it proves the same performance will survive after the support is removed.
The Support Ledger
Before interpreting a strong performance, record what carried the task:
- teacher explanation;
- teacher error cue;
- worked example;
- formula or reference aid;
- peer contribution;
- parent checking;
- AI hint or generated step;
- immediate correctness feedback;
- reduced task difficulty or extra time.
Support can be educationally appropriate and still change what the performance means.
Observable Dependency Signatures
- Accuracy drops sharply when one familiar cue is removed.
- The learner waits for confirmation before every step despite previous success.
- The learner can execute after a method is named but cannot select the method independently.
- The learner reproduces a model answer but cannot handle a changed surface form.
- The same support is still required after many apparently successful practice attempts.
- Teacher prompts become more frequent as task difficulty rises, obscuring the true independent baseline.
Competing Explanations for Collapse After Support Is Removed
- The learner became dependent on the cue.
- The learner never acquired the underlying knowledge or procedure.
- The unsupported task added a new transfer or selection demand.
- The support was removed too quickly.
- The later task was harder or less familiar.
- The learner’s state changed.
- The supported performance had been misinterpreted from the beginning.
Bolt does not label dependence from one failed unsupported attempt. It uses staged return tasks to discover what the support had been carrying.
School–Teacher–Student Calibration
School
The school should avoid comparing heavily scaffolded classroom performance directly with independent examination performance without recording the condition difference. Support policies should state their educational purpose and the conditions under which independence will later be measured.
Teacher or Coach
The teacher should know what each cue is doing. If a prompt supplies method selection, error detection or the next procedural step, the learner’s successful answer contains teacher contribution. That is not a failure; it is a measurement fact.
Student
The student should not be shamed for using support. The useful distinction is between “I can do this with this support” and “I can do this independently under these conditions.” Both can be true and both can guide the next performance cycle.
Worked Example: “Check Your Sign”
Every time a Mathematics student makes a sign error, the teacher says, “Check your sign.” The student immediately fixes it.
After several lessons, the teacher gives a fresh question and stays silent. If the sign error returns, the earlier performance was supported by an external error detector.
A more informative sequence is:
- Direct cue: “The sign is wrong here.”
- Reduced cue: “One error remains in these lines.”
- Criterion cue: “Check against your usual error pattern.”
- No cue: learner checks independently.
- Delayed changed problem: test whether the monitoring returns later.
Bolt measures what survives across those conditions. MindOS owns the learner operation used to build the monitoring capability.
How Do We Know?
The Education Endowment Foundation’s Teacher Feedback to Improve Pupil Learning guidance emphasises that feedback should move learning forward and that teachers should plan how pupils will receive and use it. It also warns against feedback policies that over-specify methods or create workload without sufficient benefit.
The EEF Feedback evidence summary draws on 155 studies and rates the overall evidence as high, while noting that effects vary and professional judgement remains necessary.
The 2023 meta-analysis Effects of Regulated Learning Scaffolding on Regulation Strategies and Academic Performance found a moderate overall effect across 46 studies, with outcomes varying by scaffold, grade, subject and context. The result supports the usefulness of scaffolding while also undermining any claim that one support pattern works identically everywhere.
A 2025 study, Who Regulates? Effects of Scaffolding in System- and Self-Regulated Learning, found benefits of scaffolding in its sample but no simple main effect for who carried regulation. That is exactly why Bolt measures the return performance rather than treating “more independence” as an ideology.
Evidence Boundary
Dependence is not a diagnosis or a fixed learner trait. A support can be necessary now and unnecessary later. Some authentic adult tasks are legitimately tool-supported or collaborative, so independent performance is not always the intended construct.
The independence test is required only when the educational claim itself concerns what the learner can do without that support.
Common Misconceptions
- “Help causes dependence.” Not automatically.
- “Independent work is always better.” It depends on the construct and learning stage.
- “If the learner succeeds with a hint, they know it.” The hint may be carrying part of the performance.
- “Remove support quickly to test toughness.” Poorly timed removal can create a different problem rather than a clean measurement.
- “Technology assistance is neutral.” It can alter what the performance demonstrates, just as human support can.
What Should Change Next?
Identify the function the support is carrying, then design a fair return performance in which that function is reduced, removed or changed only if independence matters to the claim. Compare performance rather than assuming success or dependence in advance.
RFE: What part of the observed performance belonged to the learner, what part belonged to the support system, and does the learner carry more of the performance on the next fair return?
Bolt Direction Graph
Supported performance → declare support contribution → predict what should remain without it → fair reduced-support return → compare performance → delay/change conditions if relevant → update school, teacher and student judgement about independent capability.
