Why Reading Comprehension Scores Plateau | When More Practice Papers Stop Working

51. Stop adding papers when the next paper cannot answer a new question

A full paper is worth running when it can reveal something the current evidence does not yet tell you: whether a repair survives integration, whether stamina holds, whether score variance narrows or whether a new paper family creates transfer. If the next paper is expected merely to produce another version of the same error pattern, use the time for targeted repair first.

52. Ask what the next intervention is supposed to change

“Do more comprehension” is not an intervention. “Reduce unsupported inference by training evidence sufficiency on fresh mixed questions” is. “Improve timing” is broad. “Reduce repeated broad rereads by strengthening passage indexing” is specific. Plateau-breaking plans should name the operation expected to change and the evidence that would show it changed.

53. One intervention should have one primary hypothesis

If a learner simultaneously changes annotation, reading order, inference method, vocabulary drills and checking routine, a later improvement cannot be attributed. Choose the highest-leverage hypothesis first. Keep other effective routines stable. Controlled change makes the next decision easier.

54. Preserve what is already working

A plateau can provoke wholesale reinvention. That is risky. If passage mapping is accurate and efficient, keep it. If comparison structure is stable, keep it. Repair the current bottleneck without dismantling systems that already convert marks reliably. Plateau-breaking should be surgical before it becomes architectural.

55. Use the smallest intervention that could plausibly restart progress

If one repeated scope error is costing marks, do not redesign the whole comprehension programme. Train quantifier and condition fidelity. If late-paper source blending is the issue, move source-control work into the late phase of sessions. Small interventions are easier to test, reverse and integrate.

56. Plateau-breaking is a sequence of experiments

Each intervention is a hypothesis: this mechanism limits current performance. Training changes that mechanism. Fresh retest asks whether the predicted improvement appears. If it does, keep and transfer. If it does not, revise the diagnosis. Progress becomes a sequence of evidence-based adjustments rather than a search for one permanent technique.

57. Set a test window before judging the intervention

Some changes can be evaluated quickly. A blank-scan routine may show value in one or two simulations. Vocabulary growth or syntax automaticity requires longer. Decide what kind of evidence is reasonable for the mechanism before declaring success or failure. Do not abandon a language-base intervention because one paper did not jump immediately.

58. Use fresh near-transfer before full far-transfer

After repair, first test the same mechanism under changed wording or a new passage of similar difficulty. If it survives, widen topic, genre, source structure or time pressure. Jumping directly from a narrow repair drill to the hardest unfamiliar mock can obscure whether the mechanism itself improved.

59. Delayed testing distinguishes learning from temporary performance

A learner can look excellent immediately after feedback because the explanation remains active. Retest after a meaningful delay on fresh material. If the repair persists without prompts, the system has changed more deeply. Plateau-breaking requires changes that survive time, not just a strong lesson.

60. An intervention should create a leading-indicator change before a score claim

If the target is comparison symmetry, the early evidence is fewer one-sided comparisons. If the target is passage mapping, the early evidence is fewer full rereads and more accurate evidence-region prediction. Total score may move later. First verify the mechanism closest to the training.

61. Hidden progress can accumulate before marks rise

A learner may improve passage mapping, reduce rereads and stabilise timing while still losing marks from one stubborn inference family. These gains create capacity that later interventions can use. Plateau analysis should not discard them simply because the score line is flat. They may be prerequisites for the next increase.

62. Hidden progress can lower the floor before raising the ceiling

Recovery and stamina work often reduce bad-day collapse first. The strongest score may not change, but the weakest scores improve. This matters because the learner is becoming more reliable. Once the floor rises, remaining skill deficits become easier to see and target.

63. Hidden progress can appear as lower cognitive cost

The learner may produce the same score with fewer rereads, shorter answers and less teacher prompting. That is meaningful. The task now consumes less working memory. The freed capacity can later support more difficult texts or higher-order reasoning. Efficiency gains can precede score gains.

64. Hidden progress can appear as stronger transfer

A score may remain similar while the learner now achieves it across more varied topics and genres. This broadening is important because narrow success has become robust success. The next plateau intervention should build on the wider base rather than dismiss it as “no improvement”.

65. Hidden progress can appear as less support

If the same score is achieved with fewer teacher prompts, fewer visible notes or less correction guidance, independence has improved. That is a transfer gain even before marks rise. Support fading should therefore be included in plateau evidence.

66. A plateau can be the price of increasing difficulty appropriately

When training becomes more varied or more representative, scores may temporarily stop rising because the task has become harder. Compare like with like. If the learner maintains the old score while handling greater paraphrase distance, longer passages or more varied genres, performance has improved even though the number appears flat.

67. Compare process at matched difficulty where possible

Use comparable passage families or question loads occasionally so a change can be interpreted. Exact matching is impossible, but broad comparability helps. If evidence-search time falls on similarly difficult passages without accuracy loss, the improvement is easier to trust.

68. Do not confuse exam-format adaptation with language growth

Early score gains can come from learning the format: command words, answer length, question order, summary procedure. Once that adaptation is complete, further progress may require deeper language and reading development. A plateau after quick early gains can therefore signal that the learner has reached the end of easy format improvements.

69. Do not confuse language growth with exam execution

A learner may read more widely and understand harder texts but still lose marks because examination answers are incomplete or timing is poor. Language growth has occurred; performance conversion has not. The next intervention should connect the stronger language base to exam execution rather than add even more general reading alone.

70. The plateau can sit between competence and performance

Competence is what the learner can do under favourable conditions. Performance is what survives time, fatigue, uncertainty and the actual response medium. If untimed fresh-text work improves but mocks do not, the gap lies in conversion. Train the conditions that are preventing competence from reaching the score.

71. The plateau can sit inside comprehension before the exam layer

If untimed fresh-text work itself remains flat, execution is not the main explanation. Return to language, text modelling, evidence, inference or question understanding. It is inefficient to intensify mock conditions when the component skill has not yet improved.

72. A plateau can reflect unstable marking interpretation

Open-ended comprehension sometimes permits more than one defensible phrasing. If marking standards shift between scripts, score trends become noisy. Use consistent criteria where possible and record the reasoning when an answer is borderline. Plateau analysis needs sufficiently stable measurement to avoid chasing marking noise.

73. Borderline answers deserve mechanism notes

Rather than simply recording right or wrong, note why an answer sits near the boundary: evidence is strong but wording too broad; inference is plausible but under-explained; comparison axis is correct but Source B support weak. These notes preserve learning value even when the scoring judgement itself is less binary.

74. Score bands can hide different error compositions

Two papers can produce the same total mark with entirely different losses. One may contain several small scope errors; another one catastrophic summary failure. Treating the equal totals as the same plateau would miss the changing system. Decompose the mark into mechanisms and question families.

75. Error composition should change as the learner improves

Early scripts may show basic question misreads and blanks. Later scripts should increasingly show narrower, more advanced errors. If the total score remains flat but the error composition moves from foundational to subtle, progress has occurred. The training system should evolve with the new error profile.

76. Persistent identical errors are stronger plateau evidence

If the same mechanism recurs across fresh topics, different teachers, delayed tests and varied wording, the evidence for a true unresolved bottleneck becomes stronger. That mechanism deserves direct intervention even if the learner is doing many other things well.

77. Recurrence under only one condition reveals a boundary, not a universal weakness

A source-attribution error that appears only late in long sessions suggests fatigue sensitivity. An inference problem that appears only in technical passages suggests domain or vocabulary interaction. The mechanism is conditional. Repair the boundary condition instead of reteaching the skill universally.

78. Plateaus can be created by excessive strategy load

A learner may accumulate so many rules—underline this, label that, check five things, use three templates—that working memory is consumed by strategy management. Scores stop improving because the support system itself is heavy. Consolidate rules into a smaller operating architecture and keep only controls with demonstrated value.

79. New strategies have a switching cost

When a method changes, performance can temporarily dip because the learner is consciously managing unfamiliar steps. Do not reject a promising intervention after one awkward attempt if the underlying decisions are improving. At the same time, novelty should not be used to excuse repeated decline. Set a test window and watch both process and score.

80. Strategy churn can create a permanent plateau

Changing methods after every paper prevents automaticity. The learner remains a beginner in the strategy itself. Once a rational intervention is selected, keep it stable long enough to evaluate unless clear harm appears. Optimisation requires a baseline.

81. The best next intervention is usually upstream of the repeated loss

If an answer is wrong because the evidence region was incorrect, fix search or passage representation before answer wording. If a summary loses meaning because the point was wrong before writing, fix selection before paraphrase. Plateau-breaking follows causal order. Downstream polishing cannot rescue an upstream failure consistently.

82. But do not chase infinitely upstream

Every reading difficulty can eventually be traced to broad language exposure, but that does not mean every intervention should become years of general reading. Choose the nearest upstream cause that is both plausible and trainable. The aim is practical leverage, not philosophical completeness.

83. Use error cost as well as error frequency

A rare error that destroys five marks or triggers a timing cascade may deserve more attention than a frequent one-mark vocabulary miss. Rank plateau mechanisms by expected scoring cost, recurrence and repairability. Training time is also scarce.

84. Use opportunity cost when choosing the next skill to train

Every hour spent on one weakness is an hour not spent elsewhere. If a mechanism is already strong enough for the target paper, further polishing may have low return. A weaker but highly consequential skill may offer more. Plateau-breaking is a prioritisation problem, not a quest for perfection in every subskill.

85. Easy gains should be taken when they are real

Not every intervention must be intellectually deep. If a learner repeatedly leaves one subpart blank, a completion scan may recover more marks than an advanced literary-analysis programme. Plateau analysis should not ignore simple high-return process defects merely because they seem less sophisticated.

86. Deep gains should be accepted when easy gains are exhausted

Eventually, simple exam-technique improvements may be complete. Further progress can require vocabulary growth, syntactic sophistication, broad reading or richer discourse knowledge. These gains are slower but real. A mature plateau plan recognises when the remaining ceiling cannot be solved by another trick.

87. Motivation can fall when effort and score decouple

When the learner works more and the mark stays flat, practice can feel pointless. Make leading indicators visible. Show that rereads fell, late-paper accuracy stabilised or source attribution improved. This does not replace the score goal; it restores a credible causal story about why the work matters and what remains to change.

88. Do not use motivational pressure as a substitute for diagnosis

Telling a learner to try harder can increase volume while leaving the same bottleneck untouched. If effort is already high, the more useful question is whether the effort is aimed at the limiting mechanism. Plateau-breaking is often a problem of direction before intensity.

89. Rest can matter when fatigue is contaminating the evidence

A heavily loaded learner may produce poorer practice quality, making it difficult to tell whether a strategy is working. Sustainable training matters. This does not mean avoiding effort. It means that a tired script may be evidence about fatigue, not an accurate test of a freshly repaired inference skill.

90. The schedule should reflect the type of plateau

A language-base plateau needs regular distributed exposure. A recovery plateau needs full-section conditions. A question-recognition plateau needs mixed interleaving. A scope plateau may need short high-frequency contrast cases. One universal weekly schedule is unlikely to fit every limiting mechanism.

91. Do not let full mocks consume all repair time

Mature students often increase mock frequency as examinations approach. If every available session becomes a full paper, the active plateau may receive no targeted repair. Maintain enough space between simulations to change the system. Otherwise mocks simply keep reporting the same limit.

92. Do not let component drills eliminate integration

The opposite error is to remain forever in isolated repair. Once the mechanism becomes reliable on varied mini-sets, return it to mixed passages and full sections. A component improvement that cannot survive integration has not broken the performance plateau yet.

93. The intervention should be retired when it is no longer limiting

Focused practice has an opportunity cost. When a mechanism survives variation, delay, reduced support and realistic pressure, move it to maintenance. Continuing intensive work after the bottleneck has shifted can itself create the next plateau by starving more important skills.

94. Maintenance should be lighter than active repair

Retired skills can reappear in cumulative mixed sets and mocks. They do not need the same volume they received during repair. Maintenance confirms stability while leaving most focused time available for the current limit.

95. Reopen a retired skill only for repeated evidence

One isolated mistake does not always justify reopening an old programme. Ask whether the old mechanism has truly returned across fresh work or whether a new condition caused a superficially similar error. Reopen narrowly at the boundary that failed.

96. Plateau analysis should simplify the plan, not endlessly expand it

The output of diagnosis should be a small active list: one primary mechanism, perhaps one supporting condition, one transfer test. If the plan grows into ten simultaneous goals, prioritisation has failed. Complexity belongs in the analysis; execution should remain compact.

97. The learner should be able to state the current bottleneck

A useful student explanation sounds like: “I usually find the right region now, but my inference still becomes too strong,” or “My early accuracy is stable, but source labels fail late.” This is more actionable than “I am bad at comprehension.” The learner becomes a participant in the repair system rather than a recipient of marks.

98. Teachers should be able to state what would falsify the diagnosis

If the hypothesis is “vocabulary is the ceiling”, performance should improve when difficult words are controlled or pre-taught. If it does not, vocabulary may not be the primary cause. Good plateau diagnosis remains open to being wrong. That is how the system avoids weeks of training the wrong mechanism.

99. Parents should see a repair path, not only a flat score line

A useful progress report can state the current score band, what has improved underneath it, the current limiting factor, the active repair and the evidence that will show the repair has transferred. This makes a plateau intelligible without pretending that marks have risen when they have not.

100. A mature plateau system keeps changing the question

At first the question may be, “Why are answers wrong?” Later it becomes, “Why does the repair disappear on unseen text?” Then, “Why does it fail only under time?” Then, “Why does the score remain flat now that timing is stable?” Progress continues because the diagnostic question changes as the learner changes. More practice is useful only when it is attached to the right current question.

101. Practice Lab: Maren’s score is flat, but the blanks are gone

Maren’s average mark barely changes across four mocks. Yet her latest scripts show complete coverage, a usable reserve and no abandoned high-value questions. Earlier papers lost several marks to non-completion. The execution bottleneck has improved. The remaining losses are now concentrated in overstrong inference and one comparison family. Calling the entire period “no progress” would hide the fact that the limiting factor has moved.

102. Practice Lab: Iona’s score rises only on familiar worksheets

Iona’s school-workbook scores improve steadily, but unseen papers from a different source remain unchanged. Review shows she has learned the workbook’s recurring command phrasing and passage shapes. The plateau is transfer, not effort. Training switches to fresh wording, unfamiliar topics and mixed question families. Progress is judged on new material rather than the training series she has learned to predict.

103. Practice Lab: Leonie keeps practising inference, but inference is no longer the limit

Months ago, Leonie’s biggest losses came from inference. Those questions are now mostly stable. Her current scripts show late-paper source attribution failures and summary compression errors. Continuing the old inference programme feels productive because it once worked, but the return has collapsed. The intervention is retired to maintenance and focused time moves to the new limiting mechanisms.

104. Practice Lab: more full papers keep reproducing the same mistake

Maren completes six full papers in two weeks. Each contains a version of the same error: she finds relevant evidence but overstates what it proves. The seventh paper is unlikely to teach anything new. Full-paper volume pauses. Short contrast drills train evidence strength, assumption cost and certainty language. Only after fresh transfer succeeds does another mock return.

105. Practice Lab: the model answer feels clear but nothing changes

Iona reads the official explanation after every mistake and understands it immediately. A week later, the same failure returns. Her plateau comes from recognition without retrieval. New correction routine: close the answer key, state the decision rule, solve a fresh item, then retest after delay. Understanding the explanation is no longer treated as proof of learning.

106. Practice Lab: an easy practice set creates false confidence

Leonie scores highly on short literal and simple inference passages. Harder unseen texts remain unchanged. The easy work has become maintenance rather than growth. Practice difficulty rises one dimension at a time: greater paraphrase distance first, then distributed evidence, then mixed genres. The aim is to expose the next boundary without overwhelming every subsystem at once.

107. Practice Lab: an overly hard set hides the real problem

Maren tackles a dense multi-source article full of unfamiliar vocabulary, graphs and subtle evaluation. She loses marks everywhere. The script is too noisy to diagnose. A simpler paired-source task reveals that the central recurring problem is comparison-axis formation. The hard set was measuring many failures at once; the reduced task makes the repairable one visible.

108. Practice Lab: vocabulary breadth is the new ceiling

Iona’s exam technique is strong. She maps passages well, recognises commands and writes concise answers. Yet unfamiliar expository texts contain enough unknown vocabulary that comprehension slows and inference weakens. Strategy practice has reached diminishing returns. Her programme shifts toward distributed vocabulary and broad reading while comprehension papers remain as periodic integration checks.

109. Practice Lab: vocabulary depth, not breadth, is the ceiling

Leonie knows almost every word in the passage but repeatedly misreads subtle terms such as “qualified”, “marginal”, “tentative” and “substantial”. Her problem is not unknown vocabulary; it is degree and contextual sense. Training uses contrast cases, collocations and contextual substitution. The plateau breaks only when common words become more precise.

110. Practice Lab: syntax is hiding underneath “careless” mistakes

Maren often loses conditions in answers. At first this is labelled carelessness. Sentence-level review shows she sometimes attaches the condition to the wrong clause before answering. The first failure is parsing. Short syntax work on conditional and concessive structures produces more improvement than repeated reminders to “be careful”.

111. Practice Lab: discourse is the ceiling after sentence accuracy improves

Iona can paraphrase individual sentences accurately but struggles with overall purpose and structure. Her local reading is strong; paragraph relations are not. Training shifts from sentence meaning to passage movement: claim, evidence, concession, reframing, conclusion. Global-question performance becomes the leading indicator.

112. Practice Lab: familiar topics hide a weak fallback system

Leonie scores well on school, technology and social-media passages but drops sharply on ecology and economics. She relies heavily on prior knowledge to predict meaning. Relation-first reading on unfamiliar domains becomes the repair. The target is not encyclopedic knowledge; it is preserving cause, contrast, condition and evidence when topic familiarity disappears.

113. Practice Lab: evidence selection, not inference, is the limit

Maren practises inference language for weeks. Scores barely move. Script reconstruction shows she often begins from a merely related sentence rather than the strongest evidence corridor. Once evidence selection is isolated, her reasoning becomes much better without changing inference vocabulary. The plateau was one stage earlier than expected.

114. Practice Lab: the inference bridge is the limit

Iona consistently finds the right clues but adds motives that require too many assumptions. Evidence selection is already stable. Training now compares minimal, strong and overextended inferences from the same corridor. Her answer becomes smaller rather than more elaborate. Progress restarts because the repair targets the correct stage.

115. Practice Lab: scope is the hidden plateau

Leonie understands nearly every passage yet loses one or two marks repeatedly through “all”, “proves”, missing conditions and broadened time frames. These errors look small but now account for a large share of remaining loss. Short, frequent scope contrasts replace broad comprehension practice as the active intervention.

116. Practice Lab: paired-source work creates a new ceiling

Maren’s single-source comprehension is strong. When two voices appear, evidence ownership degrades. Her overall score plateaus because the examination now contains more advanced source work. The programme adds source identity, shared-axis comparison and late-session source control rather than more single-text inference practice.

117. Practice Lab: comparison evidence is relevant but not comparable

Iona selects good evidence from both texts, but Source A evidence concerns cost and Source B evidence concerns fairness. The plateau is not retrieval; it is axis formation. New drills require naming the shared dimension before any evidence is collected. Total comparison accuracy begins to rise without increasing reading volume.

118. Practice Lab: evaluation is stuck in a formula

Leonie writes that experts are reliable and eyewitnesses are useful regardless of the claim. Her answers sound polished but remain generic. Practice switches to claim-specific criteria. A national-prevalence claim triggers representativeness; a technical-process claim triggers expertise and method. Evaluation becomes less formulaic and more discriminating.

119. Practice Lab: summary selection is the real ceiling

Maren practises paraphrasing but summary marks stay flat. Review shows the language is good; the selected ideas include examples and consequences outside the requested category. Selection is now isolated from writing. Once the point set becomes accurate, paraphrase practice returns.

120. Practice Lab: summary compression is the real ceiling

Iona selects excellent points but distorts them when shortening. Actors swap, conditions disappear and causes become effects. Selection drills are retired. Training moves to proposition fidelity under word limits. The same summary score had been produced by a completely different mechanism from Maren’s.

121. Practice Lab: untimed performance improves but mocks do not

Leonie’s fresh untimed work is clearly better: stronger evidence, tighter scope, better comparison. Full mocks remain flat. The plateau sits between competence and performance. Timing, fatigue and recovery become the active variables. More untimed skill drills alone will not convert the gains.

122. Practice Lab: timing improves but marks remain flat

Maren now finishes comfortably with reserve, but her score stays in the same band. This is not evidence that the timing intervention failed. It removed a completion bottleneck. Remaining losses are now mostly inference scope and evaluation. The next repair moves downstream into accuracy.

123. Practice Lab: late-paper accuracy is the only stubborn area

Iona’s first two thirds improve, but the final third remains unchanged. Error types are not conceptually harder; they are fatigue-sensitive versions of skills she can already perform. Training moves source attribution and scope work later in realistic-duration sessions. The plateau is endurance under load.

124. Practice Lab: one hard question creates the whole plateau

Leonie’s ordinary accuracy is high, but scores swing because one ambiguous item occasionally creates six minutes of overrun and several later rushed mistakes. Average score barely changes. Recovery training reduces the catastrophic tail. Her best score stays similar, but bad days improve and the average eventually follows.

125. Practice Lab: variance falls before the mean rises

Maren’s scores used to range widely. After recovery and pacing work, they cluster tightly around the same average. A parent may see “still the same mark”. The teacher sees stronger reliability. With variance lower, the remaining language and inference bottlenecks can be diagnosed without being obscured by execution collapse.

126. Practice Lab: a useful intervention is hidden by harder material

Iona maintains the same score after training shifts to longer, denser passages. Process data shows fewer rereads and better source ownership despite the greater difficulty. The flat score does not mean no gain. On matched-difficulty retests, the improvement becomes visible.

127. Practice Lab: a score rise is caused by easier material

Leonie celebrates a large jump on one paper, but the passage mix is unusually familiar and direct. Fresh matched work returns to the old band. Plateau analysis resists overreacting to the high. The training plan stays unchanged until repeated evidence shows the mechanism has shifted.

128. Practice Lab: teacher prompts are carrying the skill

Maren answers well during lessons whenever the teacher asks, “Where is your evidence?” Independent papers remain flat. The plateau is self-initiation. Prompts are faded from spoken question to written cue to no cue. Progress is judged by unseen independent performance, not guided accuracy.

129. Practice Lab: the wording cue is carrying the skill

Iona is excellent at “What does this suggest?” but slows on “What can be inferred?” and “What is implied?” She has learned the phrase more than the operation. Training groups varied command wording under the same underlying job. The plateau breaks when operation recognition transfers across surface language.

130. Practice Lab: answer-key dependence stalls self-correction

Leonie checks every response immediately and rarely predicts why an answer is wrong. She can recognise corrections but not generate them. New routine requires first-failure diagnosis before the model answer is opened. Over time, her independent error hypotheses become more accurate and fresh-test transfer improves.

131. Practice Lab: blocked drills create a false breakthrough

Maren completes twenty inference questions accurately because she knows every next question is inference. In mixed papers, results remain flat. The plateau is operation selection. Practice interleaves literal, inference, tone and purpose so command recognition returns to the task.

132. Practice Lab: components are strong but integration fails

Iona performs well on isolated comparison, summary and inference drills. Full sections remain flat because time, switching and working-memory demands interact. Recombination begins with two mechanisms, then mixed passages, then full simulation. The parts did not need more isolated practice; coordination did.

133. Practice Lab: full papers hide one weak component

Leonie keeps sitting mocks. Scores fluctuate enough that no one sees that vocabulary-in-context is consistently weak across every paper. Decomposition reveals the pattern. A focused context-meaning module produces better return than another full simulation.

134. Practice Lab: the learner is training below the target difficulty

Maren is fluent on short single-source passages but examination materials contain longer multi-source texts. The practice environment no longer reaches the target demands. Difficulty is raised systematically rather than all at once: length first, then source count, then subtle stance. Transfer becomes measurable.

135. Practice Lab: strategy overload becomes the plateau

Iona has accumulated colour codes, acronyms, templates, confidence labels and checking lists. She spends attention managing the method. The plateau breaks when the system is compressed to passage map, question job, evidence, answer contract and one personal risk. Fewer explicit rules free working memory for the text.

136. Practice Lab: strategy churn prevents automaticity

Leonie changes methods every week after watching new advice. No strategy becomes familiar enough to evaluate. A sensible baseline is frozen for several fresh attempts. Only one variable changes at a time. The score initially stays flat, but process data finally becomes interpretable.

137. Practice Lab: the wrong downstream repair wastes weeks

Maren’s answers are vague, so she practises more formal phrasing. The real first failure is that she never found sufficient evidence. Language becomes prettier but scores stay flat. Once evidence search is repaired, the existing language is already adequate. The plateau was caused by repairing the visible output instead of the upstream mechanism.

138. Practice Lab: error frequency and error cost disagree

Iona misses several one-mark vocabulary items but only occasionally loses a high-value summary through category confusion. Frequency suggests vocabulary; cost suggests summary. Training allocates some maintenance to vocabulary but prioritises the rarer, larger loss. Plateau-breaking uses scoring impact as well as count.

139. Practice Lab: an easy mechanical gain restarts the score

Leonie’s comprehension is sophisticated, yet she repeatedly overlooks one subpart in long questions. A simple completion scan recovers marks immediately. The intervention is not intellectually glamorous, but the plateau does not care. High-return defects deserve repair regardless of prestige.

140. Practice Lab: there are no easy gains left

Maren’s timing, checking, command reading and answer completeness are stable. Remaining errors arise in dense syntax and nuanced vocabulary. The next gains will be slower. The programme shifts toward language growth rather than endlessly searching for another exam trick. Plateau analysis prevents unrealistic expectations about the speed of future improvement.

141. Practice Lab: effort rises while quality falls

Iona doubles paper volume and begins completing tired, superficial reviews. Scores remain flat. Reducing quantity allows deeper error reconstruction, targeted repair and delayed retest. Total practice hours fall slightly while the causal quality of each session rises.

142. Practice Lab: motivation falls because no progress is visible

Leonie sees the same total mark for weeks. Her training log, however, shows fewer rereads, improved late accuracy and stronger unseen transfer. Making these leading indicators visible restores a plausible path: the system is changing, but one inference-scope ceiling still holds the total score. Motivation is supported by evidence, not empty reassurance.

143. Practice Lab: fatigue contaminates the intervention test

Maren tests a new comparison routine after a long school day and performs poorly. The result is inconclusive because fatigue is unusually high. A fresh rested mini-set shows the routine itself is sound. Later it is deliberately retested under fatigue. Plateau experiments need conditions clear enough that one weak session is not overinterpreted.

144. Practice Lab: mocks crowd out repair

Iona completes three full papers each week and has little time to work on the recurring comparison-axis problem they reveal. Mock frequency is reduced temporarily. Focused repair, transfer and one later confirmation mock replace the cycle of measure → measure → measure. The next full paper finally asks a new question.

145. Practice Lab: isolated drills crowd out integration

Leonie becomes excellent at scope drills but the improvement disappears in full passages where inference and timing compete. The repair has stayed isolated too long. Training recombines scope with fresh inference, then mixed passages, then a mock. The plateau breaks only when the skill survives competition.

146. Practice Lab: an old skill is being overtrained

Maren still spends two sessions each week on question-command drills even though those errors have nearly vanished. Meanwhile, syntax errors recur. The old skill moves to monthly mixed maintenance. The released time targets syntax. Progress restarts because training resources follow the current bottleneck.

147. Practice Lab: one old error returns once

Iona makes a pronoun-reference mistake after months of stability. Instead of reopening the entire old programme, she records it and watches the next few fresh texts. The error does not recur. The skill remains in maintenance. Plateau systems avoid reacting to isolated noise as if every past weakness has returned.

148. Practice Lab: the learner can finally explain the bottleneck precisely

Leonie once said, “I am just bad at comprehension.” After several cycles she can say, “I understand the passage and usually find the right evidence, but late in long papers my comparison answers lose source labels.” That sentence points directly to the next practice design. Diagnostic language has become part of learner independence.

149. Practice Lab: the diagnosis is falsified

Maren’s teacher suspects vocabulary is causing technical-passage losses. A controlled version replaces difficult terms with familiar words. Performance remains almost unchanged. The hypothesis is rejected. Further analysis shows the real problem is following long causal chains. Good plateau work becomes more accurate by allowing diagnoses to fail.

150. Practice Lab: the score finally moves after the bottleneck moves twice

Iona first repairs broad evidence search. Scores rise, then flatten. New diagnosis identifies overextended inference. That repair improves fresh mini-sets, but mocks remain flat until late-paper pacing is stabilised. Only then does the total score rise again. There was no single magic intervention. Progress came from repeatedly identifying the current limiter rather than repeating the intervention that worked first.

Advanced Plateau Lab A: the same mark after a harder transfer set

Maren scores 68% on a familiar single-source set and 68% again two weeks later on a longer unfamiliar multi-source set. A score-only interpretation says nothing changed. The process evidence says otherwise: source ownership is intact, rereads are lower and every question is completed. The later task places greater demands on the same system. The correct plateau decision is not to intensify basic comprehension practice. It is to preserve the new transfer gain and identify the narrow remaining losses inside the harder environment. A flat number can represent higher competence when the task itself has moved.

Advanced Plateau Lab B: a higher mark produced by lower challenge

Iona’s score jumps on a short narrative paper with direct evidence and familiar vocabulary. Her previous work involved denser arguments and paired sources. The rise feels encouraging, but the task has changed enough that the result cannot demonstrate a broken plateau. She records the high score without changing the training plan. A later matched argument provides more useful evidence. Plateau analysis protects both against pessimism and against premature celebration. Scores become meaningful only in relation to what the learner was asked to do.

Advanced Plateau Lab C: the average is flat because the floor is rising

Leonie’s best scores were already strong, but occasional collapses dragged the average down. Recovery training removes several disastrous papers. Her highest mark does not improve, so the ceiling looks unchanged, but the lowest mark rises substantially. For examination reliability, this is major progress. Once the floor stabilises, the average can rise without the learner needing a new peak performance. The next intervention can then target ordinary remaining errors rather than catastrophic variance.

Advanced Plateau Lab D: the average is flat because the ceiling rises while the floor falls

Maren produces a new personal best but also several unusually weak papers after adopting a complicated strategy. The average remains flat. This pattern is not a stable improvement. The method may create occasional upside while increasing variance and cognitive load. Review asks which behaviours differ between strong and weak attempts. The goal is not to keep the spectacular high at any cost. A reliable system should preserve gains without making ordinary execution more fragile.

Advanced Plateau Lab E: one question family improves but another silently replaces it

Iona’s inference marks rise after targeted repair. The total score remains unchanged because comparison and evaluation now account for more of the lost marks. The training has not failed; the composition of the score has changed. The inference programme moves to maintenance, and the new high-cost family becomes active. Plateau work must look at where marks are now disappearing, not where they disappeared when the programme began.

Advanced Plateau Lab F: better reading creates harder self-selected practice

Leonie starts choosing more demanding essays and source sets because easier material no longer challenges her. Her percentage remains similar. A raw trend line hides that she has raised the training environment herself. The teacher keeps one matched-difficulty anchor set in the programme to separate genuine stagnation from challenge escalation. Plateau analysis needs some stable reference points when the learner’s practice is intentionally becoming harder.

Advanced Plateau Lab G: correction quality improves before independent accuracy

Maren still makes several inference errors on first attempt, but her post-task diagnosis is now precise. She can identify whether the problem was evidence, assumption or certainty without teacher prompting. This is not yet the final score gain, but it is a stronger learning loop. Because self-correction quality predicts what can change between attempts, the programme preserves the diagnostic routine while continuing fresh independent testing. Plateau breaking can begin with better learning behaviour before it appears in first-pass marks.

Advanced Plateau Lab H: first-pass accuracy rises while checking adds less

Iona used to recover several marks during checking. Now the final score is similar, but fewer corrections are needed because answer contracts and scope control have moved upstream into the first pass. This is a structural improvement. The checking phase can now focus on genuine residual risk instead of repairing basic omissions. A flat final mark can therefore conceal a stronger first-pass system with more available reserve.

Advanced Plateau Lab I: teacher feedback is accurate but no longer high leverage

Leonie’s teacher keeps writing “support your inference with evidence”. The comment is true, but the scripts show that she usually has evidence now; the remaining problem is that her conclusion is too strong. Accurate feedback can still be stale if it targets a stage that is no longer limiting. The feedback changes to “reduce the assumption gap and calibrate certainty”. Plateau-breaking depends on diagnostic resolution, not merely correctness of the general advice.

Advanced Plateau Lab J: strong guided reasoning fails after support is removed

Maren can produce excellent evaluation when a teacher asks which criterion matters. Independent scripts remain generic. The plateau is not evaluation knowledge itself; it is criterion selection without an external cue. Prompt fading becomes the intervention: full question, partial cue, self-cue, no cue. Progress is measured on fresh unseen work after the support disappears. Guided excellence is treated as a component of learning, not as proof that independent performance is solved.

Advanced Plateau Lab K: more difficult vocabulary is not the actual limit

Iona struggles on technical passages, so vocabulary seems the obvious cause. A controlled version supplies definitions for every difficult term. Her performance improves only slightly. The main errors remain in tracking multi-step causation. The vocabulary hypothesis is downgraded. Practice shifts to causal-chain representation, while vocabulary remains a supporting condition rather than the primary bottleneck. Plateau diagnosis should distinguish what makes a task feel difficult from what actually causes the marks to disappear.

Advanced Plateau Lab L: stronger vocabulary does not convert because syntax remains weak

Leonie learns the key academic words in a passage family, yet dense sentences still produce wrong interpretations. She recognises every noun but attaches conditions and qualifiers to the wrong propositions. Vocabulary intervention has reached its limit. Syntax becomes the new limiter. The case shows why language growth itself has layers: knowing more words cannot compensate indefinitely for weak sentence architecture.

Advanced Plateau Lab M: syntax improves but global purpose stays flat

Maren becomes much more accurate on complex sentences, but whole-text purpose and structure remain weak. The local language bottleneck has improved. The remaining ceiling sits at discourse integration: how claims, examples, concessions and conclusions form one movement. Sentence-level work moves to maintenance. Passage-architecture training becomes active. A plateau can move upward in scale as lower-level processing improves.

Advanced Plateau Lab N: a stronger passage map does not fix answer scope

Iona can now state the passage movement accurately and find evidence quickly. Her scores still leak through broad quantifiers and overconfident wording. The map repair succeeded but cannot solve a proposition-fidelity problem downstream. Training moves to scope contrast cases. Plateau analysis respects causal boundaries: a useful upstream improvement should be preserved even when it does not automatically repair every later stage.

Advanced Plateau Lab O: scope work succeeds but full papers remain unchanged

Leonie performs scope drills accurately, including after delay. In full papers, the same errors return only in the last third. The skill exists; fatigue breaks its preservation. More scope explanation is unlikely to help. The boundary condition becomes the training target: scope-sensitive questions are moved late in sessions, full-duration work is added and low-load routines are protected under fatigue.

Advanced Plateau Lab P: stamina training fails because the early paper is inefficient

Maren’s late accuracy is weak, so stamina is suspected. Timeline review shows she spends too much time and cognitive effort over-annotating the first passages. By the final third she is carrying unnecessary time debt and mental load. The first repair is early efficiency, not more endurance sessions. Once early cost falls, late accuracy improves without any special stamina programme. Fatigue symptoms can be downstream of inefficient execution.

Advanced Plateau Lab Q: recovery training lowers variance but exposes a language ceiling

Iona once lost many marks after one hard question. Recovery training stabilises the rest of the paper. Her results become consistent, yet the mean still sits below target. The remaining errors cluster around nuanced vocabulary and writer stance. The plateau has become cleaner. Reducing execution noise did not solve every problem; it made the deeper language limit easier to see.

Advanced Plateau Lab R: timing repair creates reserve but the reserve is misused

Leonie becomes faster and reaches first completion with eight minutes. Scores remain flat because she spends the reserve rereading secure answers and sometimes changes correct responses. The timing intervention succeeded. The new limiter is checking allocation. 110 becomes the next owner. Plateau analysis prevents the team from pushing speed further when the new spare capacity is simply being used poorly.

Advanced Plateau Lab S: checking improves but one blank remains a recurring mechanical loss

Maren’s high-level checking is sophisticated, but she repeatedly misses a small subpart hidden under a multi-part prompt. The simplest intervention—a visual component scan—has higher expected return than deeper interpretation work. Once the omission disappears, the advanced checking routine remains unchanged. Plateau work prioritises actual scoring leverage over intellectual complexity.

Advanced Plateau Lab T: full mocks show no score gain because the repair has not had time to consolidate

Iona learns a new source-comparison method and sits a full mock the next day. The result is unchanged. Process review shows she is still consciously recalling the steps and loses time doing so. The method has not yet automatised. Instead of rejecting it, the programme returns to varied mini-sets and delayed practice. A later mock is scheduled only after the operation can run with lower conscious cost.

Advanced Plateau Lab U: a method looks effective because the teacher selected friendly examples

Leonie’s new inference routine works beautifully on lesson examples designed to illustrate it. On ordinary mixed texts, the gain is smaller. This is not deception; instructional examples often make the target pattern visible. Plateau evidence must eventually leave the teaching distribution. Fresh material with no guarantee of the relevant pattern shows whether the learner can recognise when the method applies and when it does not.

Advanced Plateau Lab V: interleaving initially lowers accuracy

Maren performs strongly on blocked question families. When practice becomes mixed, accuracy drops. The initial decline is informative because operation selection has returned to the task. The teacher does not immediately return to blocked work or declare failure. After several mixed sets, classification speed improves and accuracy recovers. A productive intervention can temporarily make practice look worse by restoring a difficulty the old drill had removed.

Advanced Plateau Lab W: a fresh-text dip is the price of removing cue dependence

Iona’s guided scores are high, but independent unseen texts are lower. The programme deliberately reduces prompts and familiar wording. Scores dip at first. The goal is not to preserve the supported number; it is to build autonomous performance. Plateau analysis judges the intervention by whether independent fresh-text accuracy rises over the test window, not whether support removal is painless on day one.

Advanced Plateau Lab X: broad reading improves comprehension without changing weekly paper scores yet

Leonie spends several weeks reading varied non-fiction and literary prose while doing fewer full papers. Her weekly comprehension score remains similar. Yet unknown-word rate falls, dense sentences require fewer rereads and unfamiliar topics feel less disruptive. These leading changes are exactly what the language-base intervention was expected to produce. The programme continues long enough to test whether the stronger base later converts under examination conditions.

Advanced Plateau Lab Y: exam-format fluency rises quickly and then stops

Maren gains marks rapidly after learning command words, answer contracts and summary procedure. The improvement then stalls. There is no evidence that the format training stopped working; it has largely finished its job. The next score band requires deeper evidence discrimination and language control. Early fast gains and later slow gains can belong to different layers of the task. Expecting the first rate of progress to continue indefinitely creates false plateau anxiety.

Advanced Plateau Lab Z: a parent sees no gain because only total marks are discussed

Iona’s family hears the same headline mark for several weeks and assumes the programme is ineffective. A one-page plateau memo shows that blanks are gone, variance is lower, source accuracy has improved and the current bottleneck is now inference certainty. The report does not claim the score has risen. It explains what changed and what remains. Clear diagnostic reporting protects decision quality when lagging outcomes move more slowly than the underlying system.

Advanced Plateau Lab AA: the learner chases every low-frequency error

Leonie reacts to each new mistake by adding a new drill. After a month, the programme contains many tiny targets and no sustained focus. Plateau analysis ranks errors by recurrence, cost and causal position. One-off anomalies enter the ledger but not the active plan. Focus returns to two mechanisms responsible for most lost marks. The learning system becomes simpler and more powerful because novelty no longer controls priority.

Advanced Plateau Lab AB: a rare high-cost failure deserves priority

Maren rarely loses source identity, but when she does it contaminates a multi-part comparison and several marks. Meanwhile, she frequently misses one low-value vocabulary item. Pure frequency would prioritise vocabulary. Expected cost points to source ownership. The programme trains the rare but catastrophic mechanism while maintaining vocabulary lightly. Plateau-breaking allocates effort by consequences as well as counts.

Advanced Plateau Lab AC: the learner reaches the current efficient floor

Iona has compressed first reading, answer length and checking safely. Further speed attempts repeatedly damage scope. The correct decision is to stop trying to make those operations faster. The plateau is not a refusal to improve; it is evidence that the current process is near its efficient floor. Future gains must come from another dimension—language depth, reasoning quality or stamina—not from forcing an already-efficient operation below its safe boundary.

Advanced Plateau Lab AD: the score moves only after the programme becomes simpler

Leonie’s plateau plan once contains eight active goals. Each appears sensible, but none receives enough focused repetitions to stabilise. The plan is reduced to one primary limiter, one supporting condition and one transfer test. Old strengths remain in mixed maintenance. Within several weeks, the target mechanism finally changes reliably and the score begins to move. The lesson is not that simple problems need simple analysis. It is that complex analysis should produce a simple executable training decision.

Technical Appendix A: The plateau dashboard

A useful plateau dashboard should make the system visible without turning every comprehension lesson into data administration. Track only variables that can change a training decision. The first layer is outcome: total score or broad score band across comparable work. The second layer is stability: lowest recent result, highest recent result and how wide the spread has been. The third layer is process: the few indicators closest to the active repair.

Possible process indicators include first-completion coverage, full-reread count, evidence-region accuracy, inference overreach, source-attribution accuracy, comparison symmetry, summary point selection, quantifier fidelity, reserve at first completion and harmful answer changes. Do not track all of these at once. Select two or three that match the current hypothesis. If the active repair is source ownership, tracking vocabulary uncertainty adds noise.

The dashboard should also record task context in a compact way: genre, approximate difficulty, single-source or multi-source, timed or untimed, paper or digital, fresh or previously seen. This prevents obvious apples-to-oranges comparisons. A score on a familiar short narrative should not be treated as directly equivalent to a dense unfamiliar multi-source argument.

Finally, add a short interpretation field: what changed underneath the score? Examples: “same score, but no blanks”; “same score on harder passages”; “variance narrower”; “inference stable, summary now main loss”; “support faded successfully”. This field prevents the learner, teacher or parent from reading a flat number as proof that the entire system is unchanged.

Technical Appendix B: Plateau confirmation protocol

Before declaring a plateau, ask five questions. One: is the pattern present across multiple sufficiently comparable fresh tasks? Two: has difficulty or format changed enough to explain the flat score? Three: are process indicators still improving? Four: has score variance narrowed or the performance floor risen? Five: is the same mechanism actually recurring, or are different errors merely producing a similar total?

If the answer to several questions is unclear, the correct conclusion may be “not enough evidence yet.” This is better than redesigning training on weak evidence. A learner can lose time when adults react to every short-term score movement. Plateau confirmation creates a small delay before intervention, allowing repeated evidence to separate signal from noise.

Once a genuine plateau is confirmed, the next task is not immediately to choose a new strategy. First establish what has already improved and should be preserved. Then identify the largest remaining recurring loss. This preserves gains while narrowing the intervention space.

Technical Appendix C: Limiting-factor decision tree

1. Is untimed fresh-text accuracy improving? If no, stay below the performance layer and diagnose language, passage modelling, question interpretation, evidence, reasoning or answer construction.
2. If untimed accuracy improves, do timed fresh passages improve? If no, investigate speed–accuracy conversion, pacing, switching cost or time-sensitive support dependence.
3. If timed passages improve, do full sections improve? If no, investigate integration, stamina, recovery, question order, interface and working-memory competition.
4. If full sections improve but total scores remain flat, has task difficulty risen? If yes, compare matched difficulty and process quality before declaring failure.
5. If task difficulty is stable, has score variance changed? A rising floor or narrower spread may indicate meaningful stability gains.
6. If neither mean nor variance changes, what error family now accounts for the largest repeatable cost? Route to that mechanism.
7. Does the mechanism survive reduced support and delay? If not, the plateau may be transfer or retention rather than initial learning.
8. Does the mechanism fail only under one condition? Train the boundary condition rather than the whole skill again.

The tree is deliberately hierarchical. It asks whether the learner can perform before asking whether the learner can perform fast; whether the learner can perform fast before asking whether the learner can coordinate a whole section. This reduces the chance of treating an upstream competence problem as an exam-strategy problem or a performance problem as a knowledge problem.

Technical Appendix D: Error-composition matrix

A flat total score becomes more informative when losses are grouped by mechanism. Useful broad families include representation—passage map, source identity, chronology; question interpretation—command, scope, mark demand; evidence—wrong region, insufficient evidence, keyword matching; reasoning—inference, cause, comparison, evaluation; expression—paraphrase distortion, incompleteness, answer inflation; execution—timing, recovery, fatigue, checking, mechanics.

Compare composition over time. Suppose early losses are 40% question interpretation, 30% evidence and 30% everything else. Later the total mark is unchanged but question-interpretation loss falls almost to zero while evaluation and scope rise. The learner has not remained the same. The error distribution has become more advanced. The plateau system should stop training basic command reading and focus on the new composition.

Also note cost concentration. If one mechanism causes a small number of high-mark losses, it may deserve priority even when its count is low. Conversely, a frequent low-cost error may be suitable for maintenance rather than dominant training. This is why the matrix records mark consequence alongside frequency.

Technical Appendix E: When to stop assigning more full papers

Pause full-paper volume temporarily when several of the following are true: the same error mechanism has appeared in multiple fresh papers; the learner cannot explain what changed between one paper and the next; review time is being squeezed out by completion volume; model-answer reading is replacing active repair; the learner is too fatigued to produce representative practice; targeted mini-sets have not yet been used on the identified bottleneck; or the next paper is unlikely to test a different hypothesis.

Stopping full papers does not mean abandoning examination realism. It means moving from measurement to engineering. Use the time to isolate the limiting mechanism, vary its surface, retest after delay and recombine it gradually. Then return to a full paper with a clear question: does the repaired process survive the integrated environment?

Resume higher mock frequency when component repairs are stable enough that integration and readiness have again become the main uncertainties. At that point, a paper can reveal coordination, stamina, recovery and score stability. The same tool becomes useful again because its diagnostic job has changed.

Technical Appendix F: Intervention selection matrix

• Repeated wrong evidence region: passage modelling and evidence-corridor prediction; not answer phrasing.
• Correct evidence, overextended conclusion: inference bridge and certainty calibration; not more evidence search.
• Correct idea, wrong source: source identity and attribution; not generic reading practice.
• Correct evidence from both sources, weak comparison: shared-axis formation and symmetry.
• Summary point set wrong: category and selection before paraphrase.
• Summary points right but meaning distorted: proposition fidelity and compression.
• Untimed strong, timed weak: speed–accuracy and pacing.
• Early strong, late weak: stamina and preservation under fatigue.
• Single hard item creates later errors: recovery and state preservation.
• Guided strong, independent weak: prompt fading and self-initiation.
• Familiar topics strong, unfamiliar weak: vocabulary, relational reading and uncertainty triage.
• Blocked drills strong, mixed tasks weak: interleaving and operation recognition.
• Full papers flat, components strong: recombination and simulation.
• Full papers flat, one component repeatedly weak: isolate that component.

The matrix is a routing device, not a rigid prescription. A learner may have two interacting mechanisms. Choose the more upstream or more costly one first, then test whether fixing it removes the apparent downstream problem. Keep the active plan small enough that results remain interpretable.

Technical Appendix G: Teacher plateau-review protocol

1. Collect several fresh samples rather than relying on one paper.
2. Separate outcome, stability and process indicators.
3. Compare task difficulty and format before reading the score trend.
4. Reconstruct the first meaningful failure in representative errors.
5. Group errors by mechanism and estimate both frequency and mark cost.
6. Identify what has improved and freeze those successful routines.
7. Name the current limiting hypothesis in one sentence.
8. State what evidence would falsify that hypothesis.
9. Select the smallest intervention capable of changing the mechanism.
10. Define a leading indicator and an accuracy control.
11. Test near transfer, then far transfer, then integration.
12. Retire the intervention when the bottleneck moves.

The teacher’s language matters. “Still careless” is weak because it hides mechanism and implies continuity even when the error type may have changed. Better: “You now locate the evidence reliably; the remaining loss is that your conclusion becomes stronger than the evidence.” This both acknowledges real improvement and identifies the current repair.

A plateau review should finish with a smaller plan than it began with. The teacher may have considered ten possibilities during diagnosis, but the learner should leave with one active focus, one supporting condition and one transfer test. If the plan cannot be stated compactly, prioritisation is incomplete.

Technical Appendix H: Self-study plateau protocol

A self-studying learner can begin with a three-paper or three-passage review. Do not start by calculating a complex average. For each attempt, write the total result, the largest lost-mark mechanism and one process note. Then look for recurrence. The same inference overreach across different topics is stronger evidence than three different mistakes producing three similar scores.

Next answer five prompts: What is better than one month ago? What still fails repeatedly? Does it fail untimed? Does it fail only when mixed, late or unfamiliar? What one change would test my explanation? These questions produce a working hypothesis without requiring a teacher.

After choosing an intervention, avoid opening the answer key too early. Attempt a fresh item, record the reasoning, then mark. If the repair works immediately, retest after several days. If it does not work, do not simply intensify it. Ask whether the diagnosis was wrong or whether the support needs to be reduced more gradually.

A useful self-study rule is: no new full paper until the previous paper has produced one concrete repair and that repair has been tested at least once on fresh material. This keeps full papers connected to learning rather than collection.

Technical Appendix I: Parent support during a plateau

A flat score often creates understandable pressure to increase volume. Before doing that, ask what the current training is trying to change. Useful questions include: “What has improved even though the mark is flat?” “What is the repeated error now?” “Is it the same error as last month?” “What is being practised between papers?” “How will we know the repair has transferred?”

Parents do not need to diagnose fine-grained comprehension mechanisms themselves. Their useful role is to ensure that the training cycle exists: evidence → diagnosis → focused repair → fresh retest → integration. If every week consists only of new papers and new scores, the plateau may be receiving measurement without repair.

Avoid using the same score as proof that effort is failing. Ask whether the learner is handling harder material, relying on less support, completing more of the paper or producing more stable results. These gains should not be exaggerated into score improvement, but they should be recognised because they determine what the next intervention should be.

Also avoid demanding a new method after every disappointing paper. Strategy churn can prevent automaticity and make progress impossible to interpret. Once a rational plan is in place, allow enough fresh evidence to judge it unless it is clearly harmful.

Technical Appendix J: Six-week plateau restart programme

Week 1 — Confirm and decompose. Review several recent fresh scripts. Separate score level, variance and error composition. Identify what has genuinely improved and what still repeats. Choose one primary limiting hypothesis.

Week 2 — Isolate the mechanism. Remove unnecessary competing demands. If the target is comparison axis, use short source pairs. If it is scope, use near-miss answers. If it is syntax, work dense sentences. If it is recovery, use controlled traps. Establish an accuracy floor before pressure.

Week 3 — Vary the surface. Change topic, wording, genre or evidence location while preserving the target mechanism. Reduce support gradually. The goal is to make the repair portable rather than exercise-specific.

Week 4 — Delay and interleave. Retest after several days without reminders. Mix the target with other question families so the learner must recognise when to use the repaired operation. Track the leading indicator and its accuracy partner.

Week 5 — Recombine under performance conditions. Use timed passages, late-session tasks or mixed source sets according to the plateau type. If the skill survives, place it back inside a full section.

Week 6 — Confirmation and re-diagnosis. Run one representative fresh simulation. Do not ask only whether the score rose. Ask whether the target mechanism changed, whether variance improved, and what now causes the largest repeatable loss. If the bottleneck moved, retire the old repair to maintenance and begin a new cycle.

Technical Appendix K: Twelve-week plateau rebuild

Weeks 1–2: confirm the plateau and stabilise measurement. Weeks 3–4: repair the largest current component bottleneck. Weeks 5–6: build near and far transfer with varied fresh texts. Weeks 7–8: integrate the repair with timing, source switching and mixed question demands. Weeks 9–10: test stamina, recovery and full-section stability. Weeks 11–12: simplify the final operating system, retire solved bottlenecks and focus on the new limiter.

For a language-base plateau, the twelve-week programme should allocate more time to vocabulary, syntax and broad reading throughout rather than treating them as short modules. Full papers become periodic measurement and integration tools. For an execution plateau, full-section work may enter earlier because the component skills are already present.

The rebuild should not promise a linear score rise. Some weeks may show stronger process without higher marks, especially when difficulty increases or support fades. The criterion is whether the training system keeps producing evidence of adaptation and whether the current limiting factor becomes smaller, more conditional or replaced by a different one.

Technical Appendix L: One-page plateau memo

A concise plateau memo can fit on one page:

• Current score band: broad recent range, not one isolated mark.
• Current stability: improving, unchanged or worsening variance.
• What has improved: two or three concrete process changes.
• Current repeated loss: one mechanism stated operationally.
• Condition: untimed, timed, late, unfamiliar topic, paired source or another boundary.
• Working hypothesis: why this mechanism is now limiting performance.
• Active repair: one focused training change.
• Leading indicator: what should change first.
• Accuracy control: what must not deteriorate.
• Transfer test: fresh task and delay.
• Decision date: when to keep, modify or reject the intervention.

The memo gives teacher, learner and parent a shared model without exposing the learner to a large technical dashboard. It also prevents forgotten rationale. If a strategy is changed later, the reason is documented: the old bottleneck was solved, the hypothesis was falsified, or the intervention failed to transfer.

Technical Appendix M: Full longitudinal walkthrough — Maren’s moving bottleneck

Maren begins with a score band that has been flat for two months. Her family believes she needs more full papers because she is “not used to the exam”. A review of six scripts shows something else. She completes every paper, stays within time and handles literal questions well. Most losses come from open-ended inference and evaluation. The first hypothesis is that evidence selection is weak.

Cycle 1 — evidence. In short unseen tasks, Maren is asked to mark the smallest sufficient corridor without writing an answer. Her selections are often broad and sometimes topically relevant but not question-specific. Two weeks of passage-map → corridor work reduce wrong-region choices substantially. Near-transfer and delayed tests confirm the gain. On the next full paper, inference improves and the total score rises modestly.

The score then flattens again. Continuing evidence drills produces little further change. New error reconstruction shows evidence is now usually correct. The remaining inference loss comes from statements such as “this proves” and invented motives. The bottleneck has moved from evidence to assumption control.

Cycle 2 — inference bridge. Practice uses the same evidence with several possible conclusions: direct, strongly inferred, plausible but unsupported. Maren learns to rank by assumption cost and calibrate certainty. Inference becomes more precise. Fresh mixed sets improve, but full-paper scores still vary widely.

Cycle 3 — recovery. Mock timelines reveal that low scores are produced when one difficult comparison creates a long stall and later rush. High scores occur when that stall does not happen. Recovery, not ordinary accuracy, now controls variance. SETC-SEC-COMP-070 routines are trained: minimum engagement, one repair, preserve state, move, return. The score floor rises even before the average rises.

Cycle 4 — language depth. With evidence, inference and recovery more stable, the remaining losses cluster around nuanced stance vocabulary and dense conditional prose. There is no quick exam-strategy fix. The programme shifts toward vocabulary depth, syntax and broad reading while mocks become less frequent. Progress is slower, but the diagnosis is now appropriate to the remaining ceiling.

Maren’s story illustrates why “find the technique that works” is the wrong model. The technique that worked first stopped being the limiting intervention precisely because it succeeded. Plateau-breaking required the training question to change four times while preserving the gains from earlier cycles.

Technical Appendix N: Full longitudinal walkthrough — Iona and Leonie reach the same score through different systems

Iona and Leonie both average the same comprehension mark across four papers. A score-only view suggests the same problem. Their process profiles are almost opposite.

Iona: untimed fresh-text accuracy is high. She finds evidence well and writes precise answers, but she is slow. Several questions are left incomplete near the end. Her first read is reasonable; the expensive stage is repeated checking after low-confidence inferences. Her plateau is performance conversion and confidence calibration. Training uses evidence-based stopping rules, 120 speed–accuracy work and 110 checking discipline. Her score rises when coverage improves without harming accuracy.

Leonie: finishes every paper with time. Her problem is not speed. She answers some inference and evaluation questions very quickly and confidently but uses familiar formulas rather than claim-specific reasoning. Her high-confidence wrong answers reveal a conceptual ceiling. Slowing her down indiscriminately would not solve it. Contrast cases and evaluation-criterion selection become the repair.

After six weeks, both learners may still briefly occupy the same score band. Iona’s coverage and reserve are improving; Leonie’s error composition is becoming more discriminating. Their next interventions remain different because the same outcome number does not define the same system.

This is the central discipline of plateau analysis: resist the gravitational pull of the total score long enough to identify the mechanism producing it. Only then does “more practice” become a meaningful phrase, because the practice has an object.

Technical Appendix O: Handoff to SETC-SEC-COMP-140

One plateau deserves a dedicated next spoke: the learner is accurate early but cannot keep that accuracy through long English examinations. SETC-SEC-COMP-140 — Reading Stamina for Long English Exams | Keep Accuracy After the Easy Questions — owns that endurance layer. It will examine decision fatigue, late-paper source and scope drift, working-memory preservation, recovery across passage boundaries, sustainable pacing and how to train accuracy to survive the final third.

The boundary is clean. 130 identifies stamina as the current limiter when plateau evidence points there. 140 owns the detailed mechanism and training architecture for fixing it. If the plateau is instead vocabulary, inference, transfer, speed or simulation, the learner should route to the relevant earlier owner rather than treating stamina as a universal explanation.

151. How SETC-SEC-COMP-130 connects to the lane

This page owns the comprehension-specific plateau layer: confirming that a plateau is real, separating score from hidden process change, recognising moving limiting factors, deciding when practice-paper volume has stopped helping, selecting the next intervention and knowing when an old repair should be retired. It routes individual errors to 080, transfer problems to 090, full-section validation to 100, checking to 110, and speed–accuracy ceilings to 120.

Return to SETC-SEC-COMP-000 for the complete comprehension performance system. The next reserved spoke is SETC-SEC-COMP-140, the reading-stamina problem.

152. Final principle: a plateau is a request for a new diagnosis

When progress stops, the easiest response is to repeat what worked before with greater intensity. Sometimes that is correct. Often it is not. A plateau may mean the old repair succeeded, the learner has reached a deeper language ceiling, the practice has become too familiar, the skill does not transfer, the performance environment is exposing fatigue, or the score is hiding improvement in stability and independence.

Maren protects the structure of the training problem. Iona asks whether evidence and reasoning have actually changed. Leonie asks whether the improved competence survives the conditions that create the score. Together they keep the diagnostic question current.

WHEN MORE PRACTICE STOPS WORKING, DO NOT AUTOMATICALLY ADD MORE PRACTICE. FIND THE LIMITING FACTOR THAT PRACTICE IS NO LONGER CHANGING—THEN CHANGE THE PRACTICE.