Mira improved her checking.
She checked every line twice.
Her accuracy on completed questions improved.
Her total examination score fell.
She had solved one part of the system and damaged another.
A part can become better while the whole becomes worse.
High Performance Is a System Property
Students often improve performance one component at a time.
Write more detail.
Check more carefully.
Move faster.
Use more advanced vocabulary.
Spend longer planning.
Any one of these can help.
Any one can also damage the whole task when its costs are ignored.
In this eduKatePunggol series, local optimisation means improving one component according to its own measure while failing to account for its effect on the wider performance system.
The Local Metric Can Mislead
If a student measures only checking accuracy, more checking may appear better.
If the same checking consumes the time needed to finish the paper, the global result changes.
If a writer measures only vocabulary sophistication, more difficult words may appear better.
If precision and fluency fall, the whole composition can become worse.
Always ask what the improvement costs elsewhere.
Local Optimisation in Mathematics
A Mathematics learner may optimise one stage excessively.
- Beautiful algebra that takes too long.
- Detailed checking on low-risk steps while high-value questions remain unanswered.
- Choosing an elegant method that is harder to execute under exam conditions.
- Using a universal method even when a shorter specialised route is available.
The best local method is not automatically the best total-paper method.
Local Optimisation in English Reading
A reader can spend too long perfecting one difficult inference answer.
The answer improves.
The final two questions remain incomplete.
Locally, the first answer improved.
Globally, the paper worsened.
Local Optimisation in Writing
Writing offers many tempting local targets.
- perfect opening;
- more descriptive detail;
- more examples;
- more sophisticated vocabulary;
- more editing.
Each should be evaluated against the whole piece.
An opening that takes twelve minutes may damage the ending.
A paragraph rich in detail may slow the narrative or obscure the argument.
A sophisticated word that introduces imprecision is a local gain with a global cost.
Local Optimisation in Science
A Science student can optimise the wrong component too.
Memorising more keywords can improve recall while leaving causal reasoning weak.
Adding more detail can make an explanation longer while reducing clarity.
Checking every factual term can consume attention that should be protecting the model-evidence relationship.
Local Optimisation and Verification Economy
Verification Economy is one defence against local optimisation.
Checking should be allocated according to risk and leverage, not maximised everywhere.
The objective is not “most checking.”
It is “best total reliability for the available time.”
Local Optimisation and Adaptive Pacing
Adaptive Pacing protects the whole by redistributing time according to task value and risk.
A learner may deliberately accept a merely good local answer because the next question offers greater expected return.
Local Optimisation and Sensitivity Analysis
The article on Sensitivity Analysis asks which variables have the greatest leverage.
Local optimisation warns against improving low-sensitivity variables while neglecting the ones that dominate the final outcome.
The Whole-System Scorecard
When evaluating a proposed improvement, inspect several dimensions.
- accuracy;
- time;
- attention cost;
- transfer;
- reliability;
- recoverability;
- downstream effects.
An intervention is stronger when it improves the whole profile rather than one visible metric.
The Bottleneck Can Move
Fixing one component may expose another limit.
This connects to Bottleneck Migration.
A learner improves calculation speed.
Now method selection becomes the main delay.
Continuing to optimise calculation speed may produce diminishing global returns.
Local Improvement Needs a Global Retest
After improving one component, test the complete performance again.
- Repair the target component.
- Measure whether it improved.
- Recombine it with the full task.
- Measure total performance.
- Check whether a new bottleneck or cost appeared.
A component is not truly improved until the whole system benefits.
The Local Optimisation Trap in Study
Students can optimise study metrics that feel productive but do not improve learning.
Neater notes.
Longer hours.
More completed questions.
More highlighted pages.
The relevant question is whether retrieval, understanding, transfer and performance improve.
The broader learning architecture remains in How Studying Works.
Do Not Optimise What Is Easy to Measure
Easy metrics attract attention.
Number of questions.
Minutes studied.
Pages written.
These can be useful signals.
They are dangerous when they become the objective instead of a proxy for the objective.
The Local-to-Global Ladder
- Name the component being improved.
- Name the local metric.
- Identify the resources the improvement consumes.
- Identify downstream dependencies.
- Predict possible global costs.
- Implement the change.
- Retest the whole task.
- Keep the improvement only if total performance benefits.
The Parent Version
When a child improves one visible habit, ask whether the whole learning system improved too.
“Your notes are much neater. Can you retrieve the material more accurately now?”
“You are checking more. Are you also finishing the paper?”
The goal is not to dismiss the local improvement.
It is to connect it to the true objective.
The Tutor Version
After every focused repair, return the skill to representative mixed performance.
A technique that looks excellent in isolation may create costs once time, selection, transfer and fatigue return.
Mira Stops Maximising the Wrong Thing
Mira did not stop checking.
She changed its allocation.
High-risk branch points received deliberate verification.
Routine lines received lighter checking.
She completed more of the paper while preserving most of the accuracy gain.
The local skill remained.
The whole system finally improved.
The Local Optimisation Test
- What component is being improved?
- What local metric says it is better?
- What does the improvement cost in time or attention?
- Does it damage another component?
- Does it create a new bottleneck?
- Does the complete task improve after recombination?
- Is the learner optimising an easy-to-measure proxy rather than the true objective?
- Does the improvement survive representative performance?
- Can the learner stop investing when marginal global benefit becomes small?
- Is total performance, not local perfection, driving the decision?
Next: Meet All the Conditions, Not Just the Most Obvious One
Whole-system performance depends on more than one requirement at a time.
The next article asks how learners satisfy multiple simultaneous constraints without letting the most visible condition dominate the rest.
Next: How High Performance Learning Works | Constraint Satisfaction — Meet All the Conditions, Not Just the Most Obvious One.
Research Note
Local optimisation is a standard systems idea: improving one subsystem according to a local objective does not guarantee improvement of the whole system. This article applies that logic educationally to study and examination performance, where accuracy, time, attention, transfer and reliability interact and can trade off against one another.
Series Note
“High performance learning” is used descriptively throughout this eduKatePunggol series. The series does not claim affiliation with or reproduce any third-party branded educational framework using similar terminology.
