01 MATH40005 OVERANCHORING DIAGNOSIS AND CORRECTION
English review edition prepared on 4 October 2026 from a preserved source copy. It is a later presentation, not the historical study interface. Source checks and difficulty judgements describe the original material author's own process; they do not indicate Imperial College London endorsement.
Source SHA-256: cee63b6bdae9cca1b0e66dcbc1776bbb1ea55bc9164e55234d634c41dca5bb82Source date: 2026-08-06
Source page 1
Mathematical typesetting and diagrams are retained below. Chinese prose is replaced by the English passages that follow, in source reading order.
English passage 1 (6% down the source page)
MATH40005 Diagnosis of over-anchoring and correction of previous conclusions
English passage 2 (10% down the source page)
1. Review scope and evidence limits
English passage 3 (13% down the source page)
The course is MATH40005 Probability and Statistics. This review reopened and checked the 2020–2026 official papers and solutions, available examiner comments, previous analysis reports, old Practice Papers 1–12, their solutions, coverage reports and quality-check files. No separately uploaded lecture notes, problem sheets, tutorials or assignments were found in the current files. Official solutions and examiner comments therefore provide the strongest evidence for the course-method audit; no page-by-page comparison with unavailable notes is claimed.
English passage 4 (20% down the source page)
2. Core problems in the previous version
English passage 5 (23% down the source page)
Old Papers 1 and 12 almost copied the 2026 main examination question by question, retaining the entry point, first key step and method sequence of all four questions while changing only numbers or context.
English passage 6 (25% down the source page)
The old coverage report primarily used 55 knowledge-point labels. It conflated the occurrence of a matching topic name with practice of the historical question entry point and complete method sequence.
English passage 7 (27% down the source page)
Although there were 12 old papers, method families such as regression, Bayes, joint densities and t-tests were repeated frequently. Some proof, measurability, double-counting, random-mixture and special-estimator skills from 2020–2025 were compressed into a small number of questions.
English passage 8 (31% down the source page)
The old claim of 55/55 coverage describes labels only; it does not establish effective coverage of seven years of ability units.
English passage 9 (34% down the source page)
3. Question-by-question treatment statistics
English passage 10 (37% down the source page)
The 48 whole questions across the 12 old papers were reassessed: retain A=19, revise B=18, replace C=7 and retire D=4. All complete PDFs of the 12 old papers were marked retired. Reusable abilities were rearranged into six revised papers.
English passage 11 (41% down the source page)
4. Correction principles
English passage 12 (44% down the source page)
The 2026 examination calibrates only the overall difficulty: 2 hours, 80 marks, 4 questions of 20 marks, and A/B/C/D=32/20/12/16.
English passage 13 (46% down the source page)
The 2020–2026 papers determine ability scope, recognition signals, first key steps, method sequences, combinations and rotation.
English passage 14 (48% down the source page)
The six roles are: 2026 structural reinforcement; a 2025 bridge; 2023–2024 abilities; 2020–2022 abilities; seven-year gaps; and seven-year mixed-transfer pressure.
English passage 15 (51% down the source page)
The new matrix counts only A (direct question-type coverage) and B (complete method-transfer coverage) as effective coverage. C (similar labels) does not count.