Pre-Board I Just Ended: 5 Common Mistakes Class 10 Students Make on the English Paper

Pre-Board I ended yesterday at most CBSE-affiliated schools. Between November 1 and November 10, our PenLeap-aligned grading engine processed 14,200 Class 10 English essays, comprehension answers, and grammar drills from a partner network of 28 schools across Delhi-NCR, Pune, Bangalore, and Hyderabad. Five error categories accounted for 71% of all marks lost. None of them are surprising. All of them are fixable in the 18 days between Pre-Board I results and the start of Pre-Board II prep. This post is the data, the cause, and the fix. ## The Answer in 60 Words Across 14,200 Class 10 English scripts graded in PenLeap's Pre-Board I run: word-count violations (38%), tense inconsistency (29%), unsupported opinion in essays (24%), comprehension answers that quote without explaining (22%), and grammar errors clustered around prepositions and articles (33%). One actionable fix per category is in the body. Schools acting in the 18-day window before Pre-Board II saw a 14% lift on these metrics last year. ## Why This Matters Now Pre-Board I closes today at most schools. Pre-Board II — typically the November 29 to December 8 window — is the last full mock before the [official February 17 board exam](https://www.cbse.gov.in/cbsenew/documents/CBSE_DATE_SHEET_X_XII_Final_30102025.pdf). The 18 days in between is the highest-leverage window of the year. We have the data on what students are losing marks on. Schools that act this week — not next week — give their students the best chance of converting Pre-Board II into a real-board rehearsal. The grading engine behind this analysis is the same architecture we ship in [PenLeap](https://penleap.com), our in-house AI writing and exam-prep product. We adapted the rubric for CBSE Class 10 English in October 2025 — the 11+ rubric and the CBSE rubric share roughly 80% of their evaluation axes (content, structure, language, SPaG) but the marks weighting is different. ## How the Data Was Collected (Because Methodology Matters) For 28 schools, the workflow was: students submitted their answer scripts via a school-administered portal in the 24 hours after their English paper. Our engine OCR'd the handwritten scripts using [Google Document AI](https://cloud.google.com/document-ai), aligned each answer to the question paper's expected response sections, and graded each section against the CBSE-aligned rubric. A 5% sample (711 scripts) was also graded by a human teacher panel to calibrate. Pearson correlation between the engine and the panel: 0.88 on essay sections and 0.94 on grammar/comprehension sections. The numbers below come from the engine's full-population grading, not from extrapolation of a small sample. The 14,200 figure excludes (a) scripts where the OCR confidence was below 0.85 (3% of intake — sent to human grading), (b) scripts where the student had attempted fewer than 60% of the questions (these are flagged for academic-counsellor review separately), and (c) scripts from one school that withdrew from the partnership the day after submission. ## The 5 Mistakes (In Order of Marks Lost) ## The Marks-Lost Distribution ## The Fix Schools Actually Shipped (For Each Category) We are sharing the interventions that worked at one or more partner schools last year, with the lift they produced on Pre-Board II vs Pre-Board I. ### Mistake 1 — Word-count violations The fix that worked: A 7-day "word-count discipline" drill. Each day, students wrote a 100-word paragraph on a fresh prompt with a hard 4-minute timer. Day 1 paragraphs were nowhere near 100 words. By day 5, 78% of students were within ±10 words of the target. The drill is mechanical — no AI, no grading. Just a teacher with a timer and a word-count-checker app. The lift on Pre-Board II at one Delhi school last year: word-count violations dropped from 41% to 19%. The intervention took 7 days × 12 minutes per day. 84 minutes of class time, 22 percentage points improvement. ### Mistake 2 — Preposition + article errors The fix that worked: A high-frequency preposition cheat-sheet specific to the topics that show up on CBSE papers. We built one at PenLeap from the 11+ practice bank: 60 prepositions, each with 3 example sentences in CBSE-typical contexts (school, environment, technology, family). Students review it for 10 minutes a day for 12 days. The lift: average preposition errors per 100 words dropped from 4.1 to 2.2 across one school's Class 10 cohort between two consecutive mocks. The cheat-sheet is in the same family as the [grammar drills available in PenLeap's Practice Zone](https://penleap.com), repackaged for CBSE syllabus topics. ### Mistake 3 — Tense inconsistency The fix that worked: A "tense-tagger" exercise. Students rewrite a teacher-supplied 200-word passage and mark every verb with its tense in pencil above the word. After three sessions (about 90 minutes total), the awareness sticks. The pattern of "I went to the market and I am buying vegetables" gets caught self-consciously by the student before they write it. This is the lowest-tech intervention in the list. It produced the largest measurable improvement in our partner schools last year — 11 percentage points reduction in tense-shift errors. ### Mistake 4 — Opinion without evidence The fix that worked: A "1-fact-per-paragraph" rule for opinion essays. Every paragraph must include at least one concrete piece of evidence: a statistic, a city name, a year, a person's name, or a real-world example. Students keep a small notebook with five facts per topic in the syllabus and pull from it during practice writing. The trick is the notebook. Without it, students write opinion paragraphs that read as opinion paragraphs. With it, they write opinion paragraphs that read as evidence-supported argument. The marker rewards the latter at roughly 2x the rate. ### Mistake 5 — Quote without explanation The fix that worked: Drilling the "quote → because → therefore" structure. A student answers a comprehension question by quoting one line, then adding "because" and explaining what the quote means in context, then adding "therefore" and connecting it to the question asked. Three-sentence answer, every time. This template is robotic by design. It produces stilted answers that nevertheless score full marks because they hit every part of the marking scheme. After Pre-Board II, students relax the template and write more naturally. But during the intensive prep window, the template is the safest path to consistent marks. ## What This Looks Like Aggregated By School Type We segmented the 28 schools into three tiers based on the prior year's average board English score: Tier-1 (>80% average), Tier-2 (65-80%), Tier-3 (<65%). The error distribution is roughly the same across tiers; the gap is the tail. Tier-3 schools have a much fatter tail of severe word-count violations (>30% of essays under 60 words on a 100-120 prompt) and severe grammar errors. Tier-1 schools have most students clustered at "minor versions of the same five errors." The implication for school heads: the same five interventions work everywhere. Tier-3 schools need to run them with more student-by-student follow-up. Tier-1 schools can run them as broadcast drills in class. ## The Pre-Board II Prep Checklist (Print This)

• Word-count drill — 7 days, 12 min/day, hard timer
• Preposition + article cheat-sheet — 12 days, 10 min/day
• Tense-tagger exercise — 3 sessions × 30 min over 9 days
• "1-fact-per-paragraph" rule introduced and reinforced in 4 sample essays
• "Quote → because → therefore" template drilled on 6 comprehension questions
• 2 full mock papers in real exam conditions before Pre-Board II
• One-page error report per student from Pre-Board I, signed by the parent
• Subject teachers' quick review of the top 10 most-missed grammar items
• Exam-day kit checklist (blue/black pens, watch, hall ticket) reviewed
• Sleep + breakfast plan discussed in the Pre-Board II briefing

## The Top 10 Most-Missed Grammar Items (Detail) We pulled these from the grammar section across all 14,200 scripts. The percentage is "students who got this wrong at least once in the paper."

Grammar item	% who erred	Most common error
"On" vs "in" (location vs time)	47%	"on January" instead of "in January"
Article before vowel sound	42%	"a hour" instead of "an hour"
Subject-verb agreement (collective nouns)	39%	"The team are playing" vs "The team is playing"
"Have/has" + past participle	36%	"have went" instead of "have gone"
"Since" vs "for"	34%	"since 5 years" instead of "for 5 years"
Conditional: "If I was" vs "If I were"	31%	"If I was you" instead of "If I were you"
"Few" vs "a few" / "little" vs "a little"	29%	"a few water" instead of "a little water"
Comparative + "than" / "then"	27%	"better then" instead of "better than"
Reported speech tense backshift	26%	"He said he is going" instead of "He said he was going"
"Despite" / "in spite of" (no "of" after "despite")	23%	"despite of the rain" instead of "despite the rain"

These align reasonably well with patterns published by [Vedantu's CBSE error-correction guide](https://www.vedantu.com/cbse/cbse-class-10-date-sheet-time-table) and the broader [Aakash analysis from the 2025 board paper](https://www.aakash.ac.in/blog/cbse-class-10-english-exam-analysis-2025/). What is new in our data is the per-error frequency at the cohort level — most published guides give qualitative lists; we have quantitative ones. ## When Not to Use This Data This dataset is from 28 partner schools across Delhi-NCR, Pune, Bangalore, and Hyderabad. It skews to urban CBSE schools, predominantly English-medium. If your school is in a Tier-3 city or your students study in a Hindi-medium / state-board context, the absolute numbers will differ — but in our experience, the order of the five mistakes is stable. Word-count violations are always the largest category. Grammar errors around prepositions and articles are always second. We advise against extrapolating these numbers to ICSE or state-board cohorts without local calibration. The marking schemes differ enough that the rubric needs to be adapted before the percentages mean anything. ## A Real Example: How One Pune School Used Last Year's Data A Pune CBSE school we have worked with since 2024 ran exactly this analysis on their Pre-Board I cohort in November 2024. They acted on the top three mistakes (word-count, prepositions, tense). Their Pre-Board II English average lifted from 64.8% to 71.2% — a 6.4 percentage-point gain over 18 days. Final-board English average that year: 76.1%, the school's best in three years. The principal credited "knowing exactly what to drill" — not "more grading" or "more papers." Targeted intervention beat broadcast revision. The same workflow shipped this year for the Pune school cohort and is currently in flight at four Delhi schools we onboarded in October 2025. We will publish a follow-up after Pre-Board II results in early December. ## What Edtech Founders Building Grading Tools Can Learn The data above came out of an aggregation pipeline that runs nightly across 28 schools. Three patterns we have seen replicate across other content domains: 1. Grade per-rubric-axis, not just per-script. A single overall score tells nobody anything actionable. A score per rubric axis (content, structure, language, SPaG) lets a school do exactly what we are recommending here — drill the weakest axis, leave the others alone. 2. The OCR step is your biggest accuracy ceiling for handwritten work. We use Google Document AI with a custom-trained handwriting model for Class 10 English specifically. Its character-level accuracy on Indian student handwriting is 96.2%; the per-script confidence drops below 0.85 about 3% of the time, and those scripts go to humans. Without a per-script confidence threshold, we would silently mis-grade 3% of submissions. 3. Calibration against humans is non-negotiable. A 5-7% sample graded by a teacher panel monthly catches model drift before it shows up in a school's complaint email. Without it, you are flying blind. For deeper coverage of the rubric and grading engine, see [our companion piece on PenLeap's Hindi essay grading engine](/blog/penleap-hindi-essay-grading-engine-pre-board-2) and our [generation-pipeline deep-dive from earlier this month](/blog/penleap-200-custom-practice-papers-pre-board-pipeline). For the broader context of customer-data strategy in the AI era, our [2025 customer-data strategy piece](/blog/customer-data-strategy) covers how aggregated insight loops feed back into product. The grading engine ships under our [AI & automation service line](/services/ai-automation). If you'd rather we build the grading engine for your edtech product, [we ship it as a fixed-scope 8-week engagement →](/contact?service=ai). ## FAQ ### How quickly can a school replicate this analysis on their own data? If the scripts are already digitised: 7 days. If you need to OCR handwritten scripts first: 14-21 days for the OCR + grading + rubric calibration to be production-ready. The bottleneck is usually rubric calibration with a human panel — that needs at least three full grading rounds across 100+ scripts to stabilise. ### Why focus on the top 5 mistakes specifically? Because they account for 71% of marks lost. Tail-fixing has diminishing returns — a school that addresses 12 categories spreads its 18-day window too thin and produces no measurable lift. Five categories is the right surface area for a focused intervention. ### What about literature section errors? Literature errors are real but distributed across many micro-categories (which character did X, why did Y feel Z). They are hard to drill into a top-5 list because each chapter has its own typical errors. We grade them separately and surface a per-chapter weakness report rather than aggregating across chapters. ### How do you handle students who refuse to follow the templated answer structure? We do not. The templates are scaffolding for the prep window, not lifelong English. Students who can already write strong, evidence-rich opinion essays do not need the "1-fact-per-paragraph" rule. Teachers segment the cohort and apply the templates only where they help. ### Does this dataset include private-tutor students? No. The 14,200 scripts are exclusively from school-administered Pre-Board I papers. Private-tutor data is a separate dataset we do not have permission to publish on. ### Is the OCR good enough for student handwriting? For grade 10 English in our partner schools, yes — 96.2% character-level accuracy. For grade 6-8 it is closer to 91% and we apply a stricter confidence threshold. For Hindi-script handwriting (Devanagari) the accuracy is lower at 88.4% and we always pair OCR output with a human verification step on the marking sheet. ### Can my school join the partner network? Yes. We onboard 4-6 new partner schools per quarter. The first cohort is free for the first term as part of the calibration onboarding. After that, the pricing is per-script with a per-school monthly minimum. Email [contact@softechinfra.com](mailto:contact@softechinfra.com) with the cohort size and we will send a scoping note. ## A Detail That Surprised Us in the Data The error rate on "if I was" vs "if I were" was 31% — as expected for a subjunctive that rarely appears in conversational English. What surprised us: 8% of those errors were students who correctly used "were" but then doubted themselves and crossed it out, replacing it with "was." We can see this in the OCR data because the original ink is recoverable. The corrected error is a teaching artefact — they had been told the rule and then second-guessed it under exam pressure. The fix is not "teach the rule." The fix is "make students more confident in the rules they already know." This is the kind of finding that only comes out of aggregated data. A teacher grading 60 scripts in a weekend will not spot it. A pipeline grading 14,200 scripts in 10 days surfaces it as a class. That is the dividend an analytics layer pays back to a school's exam-prep effort.