STPM ICT 958 common mistakes
STPM ICT 958 · Understand the traps and how to avoid them.
When asked to distinguish RAM from ROM or primary from secondary storage, the candidate writes only one side of the comparison (e.g. 'RAM is volatile') and leaves the counterpart implied, so the examiner cannot award the contrast marks.
The candidate treats the question as a definition task rather than a comparison task and writes what first comes to mind about the more familiar term, running out of points before covering the second item.
What to do instead
Before writing, draw a two-column table with the two terms as headings and force at least two rows (volatility, writability, typical use, cost per bit, speed). Answer in paired sentences beginning 'Whereas X..., Y...' so every sentence scores on both sides.
In database questions the candidate draws an ER diagram using attributes as entities (e.g. an 'Address' box) and omits cardinality labels on relationship lines, then converts it to tables without resolving many-to-many relationships.
Candidates memorise the shapes of ER symbols but not the modelling rules, so they classify nouns by intuition and skip the step where an M:N relationship must become its own table with a composite key.
What to do instead
Apply a fixed checklist on every ER question: circle nouns and test each against 'does it have its own identity and more than one instance?'; mark 1 and M on both ends of every relationship; then convert by writing one table per entity, one foreign key on the many side for 1:M, and a new junction table for each M:N before writing any SQL.
The candidate writes pseudocode that is logically correct but uses inconsistent indentation and undeclared variables, or traces a loop by guessing the output instead of maintaining a variable table line by line.
Under time pressure candidates write code the way they think, not the way the mark scheme reads it, and they believe they can 'see' the answer without executing the loop on paper.
What to do instead
Adopt one house style for all answers (declare all variables at the top, indent every block by two spaces, close every loop explicitly). For trace questions, draw a table with one column per variable plus an output column and fill one row per iteration, including the final failed test that exits the loop.
On subnetting and IP questions the candidate calculates the number of hosts but forgets to subtract two for the network and broadcast addresses, or applies the subnet mask to the wrong octet when the prefix is not a multiple of eight.
The candidate has memorised '2^n minus 2' as a slogan without linking it to which addresses are reserved, and practises mostly with /24 masks where the arithmetic is easy.
What to do instead
For every question write the mask in binary under the address, draw the boundary line, and state explicitly 'usable hosts = 2^h - 2 because the all-zeros and all-ones host values are reserved'. Practise at least one /26, one /20 and one /27 example until the octet boundary is automatic.
When asked about normalisation, the candidate states 'remove redundancy' but cannot name which normal form a given table violates or show the decomposition, and often claims a table is in 3NF when a transitive dependency remains.
Normal forms are learned as definitions rather than as a diagnostic procedure, so the candidate recognises the vocabulary but has no method for testing a specific table.
What to do instead
Use a three-question test in order: are all attribute values atomic (1NF)? does every non-key attribute depend on the whole composite key (2NF)? does any non-key attribute depend on another non-key attribute (3NF)? Then write the decomposed tables with keys underlined and name the dependency removed at each step.
In security questions the candidate lists controls ('use a firewall', 'use encryption') without linking each to the specific threat in the scenario, and confuses authentication with authorisation or symmetric with asymmetric encryption.
Answers are recalled as a checklist of security words rather than selected against the scenario's assets, actors and vulnerabilities, and the paired terms sound similar enough that the distinction is never tested during revision.
What to do instead
Structure every answer as threat, mechanism, effect: name the threat from the scenario, name the control, then state what it prevents and its residual limitation. Keep a one-line glossary contrasting authentication (who you are) with authorisation (what you may do), and symmetric (one shared key, fast) with asymmetric (key pair, solves distribution).
The candidate answers an ethics or ICT-and-society question with opinion only ('it is wrong to hack') and cites no legislation, professional code or stakeholder impact, so it earns no marks beyond the first.
These questions look like general knowledge, so candidates do not prepare named examples of acts, codes or case types, and they misjudge how much structure the mark scheme expects.
What to do instead
Prepare a short bank of named references relevant to your syllabus (computer misuse legislation, data protection principles, copyright, professional codes of conduct) and answer in three moves: identify the stakeholders affected, apply the specific principle or law, then state a justified judgement with a counter-argument acknowledged.
On multimedia and web questions the candidate confuses lossy with lossless compression, gives file formats without properties, or writes HTML that omits closing tags and uses deprecated presentational tags for layout.
Format knowledge is absorbed passively from using software rather than from specification, and HTML is written from memory of what renders rather than from structural rules.
What to do instead
Learn each format as a triple (compression type, typical use, one trade-off), e.g. JPEG is lossy and suits photographs but degrades on repeated re-saving. For markup, write semantic structure first (headings, lists, tables with headers) and validate by checking every opened tag is closed and every image has alternate text.
When explaining an emerging technology the candidate gives a definition and marketing benefits but cannot describe how it works at a mechanism level or give a concrete limitation and an application context.
Coverage of new technologies in revision is often headline-level, so the candidate has vocabulary without mechanism, and exam questions asking 'describe how' or 'evaluate' expose the gap.
What to do instead
For each emerging topic prepare a four-part card: mechanism in two or three steps, one named real application, one technical limitation, one ethical or social concern. Rehearse explaining the mechanism to someone with no background, which forces precision about what is computed, stored or transmitted.
The candidate misreads the command word, writing an evaluation where a description was asked, or lists points where 'discuss' requires development, and leaves no time to answer the final high-mark structured question.
Command words are treated as interchangeable and timing is planned by question count rather than by marks, so a long final question is attempted in the last few minutes.
What to do instead
Before writing, underline the command word and the mark allocation, and match length to marks (roughly one developed point per mark). Allocate minutes in proportion to marks and rehearse past-paper timing so the last structured question always receives its full share.