Why did I lose the mark?

The answer that scoresOne kilobyte (kB) is 1000 bytes and one kibibyte (KiB) is 1024 bytes. A card sold as 4 GB holds 4 x 10^9 bytes, which is 4 000 000 000 bytes, and an operating system working in binary prefixes will report that as about 3.73 GiB, because 4 000 000 000 / 2^30 = 3.73. If the question had said 4 GiB, the capacity would be 4 x 2^30 = 4 294 967 296 bytes. The rule is to read the prefix in the question and to keep to one system throughout the calculation.

The answer that scoresIn 8-bit two's complement the range is -128 to +127. The sign bit has a place value of -128, so 1000 0000 is -128 and 0111 1111 is +127. Zero has one representation only, 0000 0000, because the negative of zero by invert-and-add-1 produces a carry out that is discarded. One's complement, by contrast, covers -127 to +127 and has both 0000 0000 (+0) and 1111 1111 (-0), which is the weakness that two's complement removes.

The answer that scoresIf the resolution of a bitmap is doubled, both the width and the height are doubled, so the number of pixels becomes 2 x 2 = 4 times as large and the file size is four times as large, assuming the colour depth is unchanged. Doubling the colour depth, on the other hand, doubles the file size, because each pixel simply uses twice as many bits. So resolution changes the size by the square of the scale factor and colour depth changes it in direct proportion, and reducing either one saves space at a cost in detail or in the number of colours available.

The answer that scoresRun-length encoding is lossless: a run of identical values is replaced by a count and the value, so 12 white pixels become 12W, and decoding reproduces the original pixel values exactly. It reduces the file size greatly when there are long runs of the same value, as in icons, scanned line drawings and fax pages, which are mostly white. It performs badly on photographs, because neighbouring pixels rarely share a value for long and the count-value pairs add data instead of removing it, so RLE must be chosen to suit the content.

The answer that scoresRaising the sampling rate records the wave more often, so higher frequencies are captured and the recording is more accurate, but the file size rises in direct proportion and storage and bandwidth are finite, so the rate is chosen to match the content - 44 100 Hz for music, around 8 000 Hz for speech. A vector graphic does not lose quality when enlarged, because it is stored as a drawing list of objects with properties and is redrawn at the resolution of the output device, so the edges stay sharp. It is a bitmap that pixelates when enlarged beyond its own resolution, because the extra pixels have to be interpolated and contain no new detail.

The answer that scoresA switch connects the devices within one LAN. It learns which MAC address is on each port and forwards each frame out of the port where the destination is, so traffic does not reach every device. It does not change the IP packet and cannot pass data between two different networks. A router connects different networks, and each of its interfaces has an IP address. It reads the destination IP address of each packet, looks it up in its routing table and forwards the packet to the best next hop, which is how a LAN reaches the internet through the default gateway. A home router usually contains a switch and an access point as well, which is why the two are confused.

The answer that scoresThe internet is the global infrastructure: the interconnected networks, submarine cables, routers and exchange points that carry data using TCP/IP. The World Wide Web is one service that uses that infrastructure - the collection of web pages and resources, addressed by URLs and fetched with HTTP or HTTPS by a browser. Email, file transfer, voice calls and streaming are other services on the same internet. A device can be connected to the internet without anyone opening a web page, but a web page cannot be delivered without an internet connection.

The answer that scoresAn IPv4 address is 32 bits, written as four decimal numbers from 0 to 255 separated by dots, giving about 4.3 billion addresses. An IPv6 address is 128 bits, written as eight groups of four hexadecimal digits separated by colons, giving an address space large enough for every device to have a globally unique address, which removes the need for NAT. IPv6 also has a simpler fixed header, which can make routing slightly more efficient, but the reason for the change is the exhaustion of IPv4 addresses, not speed.

The answer that scoresEach host senses the medium before transmitting and only starts when it is idle, but two hosts can still begin at the same moment. While transmitting, each host compares the signal on the cable with what it is sending; when the two differ, it knows a collision has occurred. It stops transmitting, sends a short jam signal so that all stations detect the collision, and then waits for a random period chosen from a range of time slots before it senses the medium and tries again. If it collides again the range doubles, which spreads the competing hosts out and makes repeated collisions unlikely.

The answer that scoresA private address comes from a reserved range such as 192.168.0.0/16 and is not routed on the internet, so no host outside can address the device directly; the router's NAT translates the private address to the public one and matches the replies. That removes a large class of unsolicited inbound attacks and hides the internal layout of the network. It is not, however, a security control in its own right: outbound connections are still allowed, a port forward deliberately exposes an internal server, and anyone who gains access to the LAN can reach the device. A firewall and up-to-date software are still needed.

The answer that scoresROM holds the bootstrap program: the first instructions that run when the power is applied, which test the hardware, find the boot device and load the operating system from secondary storage into RAM. It is used because it is non-volatile, so the instructions are there every time the machine is switched on, and read-only, so they cannot be altered by a program or by a virus. The operating system itself, and every application and file, lives in secondary storage and is copied into RAM to run, because RAM is fast and can be written to but loses its contents when the power goes.

The answer that scoresA buffer is a temporary area of memory, usually RAM, that holds data while it is being transferred between two parts of a system working at different speeds. The printer buffer holds the document because the processor can send it far faster than the printer can print it: the processor writes into the buffer and carries on with other work, and the printer takes data out as it needs it. Nothing in the buffer is kept permanently - when it has been printed it is overwritten, and when the power goes it is lost.

The answer that scoresThe sensor measures the condition - the thermistor's resistance changes with temperature - and an ADC converts the signal into a digital value. The microprocessor compares that value with the set value stored in memory. If the reading is below the set value the microprocessor sends a signal to the actuator, which switches the heater on; when the next reading reaches the set value it switches the heater off. The sensor is read again after each action and the result is fed back, so the system corrects itself and does not overshoot.

The answer that scoresXOR (EOR) gives an output of 1 when the two inputs are different and 0 when they are the same, so its truth table is 00 -> 0, 01 -> 1, 10 -> 1, 11 -> 0. OR gives 1 when at least one input is 1, so its table is 00 -> 0 and 1 for the other three rows. The difference is the last row, and it is the row an examiner will look at first.

The answer that scoresSRAM stores each bit in a flip-flop of four to six transistors, so the bit stays put with no refresh, which makes it very fast; but each cell is large and expensive, so it is used in small amounts for cache and high-speed buffers. DRAM stores each bit as charge on a capacitor beside a single transistor; the charge leaks, so every cell must be refreshed thousands of times a second, which makes it slower, but a cell is far smaller and cheaper, so main memory is built from DRAM.

The answer that scoresThe memory address register holds the address of the memory location that is currently being read from or written to, and it is connected to the address bus. The memory data register holds the data or instruction that has just been read from that location, or the data that is about to be written to it, and it is connected to the data bus. During the fetch stage the address in the PC is copied to the MAR, the instruction at that address is read into the MDR over the data bus, and it is then copied to the CIR.

The answer that scoresAn interrupt is detected at the end of the execute stage of the fetch-execute cycle, so the instruction being executed is completed first. The processor then saves the contents of its registers, including the program counter, on the stack, checks the priority of the interrupt, and loads the start address of the relevant interrupt service routine into the program counter so that the ISR runs. When the ISR finishes, the saved register values and the program counter are restored from the stack, so the interrupted program continues from the instruction it had not yet executed.

The answer that scoresLoad the status byte, then AND the accumulator with a mask that has a 1 only in bit 2, so that every other bit becomes 0 and bit 2 is unchanged. If the bit was clear the accumulator now holds 0, and if it was set the accumulator holds 4. Compare the accumulator with 0 and branch: JPE (the compare was true, so the results were equal) means the bit was clear, and JPN (the compare was false) means the bit was set. OR would be used to set a bit, and XOR to toggle one.

The answer that scoresThe leftmost bit is the sign bit, and here it is 1, so the number is negative and the shift must keep it negative. An arithmetic right shift copies the sign bit into the vacated positions on the left, so B10010110 shifted one place right becomes B11001011, and shifted two places right becomes B11100101. The value is halved by each place, with the result rounded down towards minus infinity, so the sign is preserved. A logical right shift of the same pattern would fill with zeros and give B01001011, which is a completely different, positive, number.

The answer that scoresLDX 200 uses indexed addressing: the processor adds the operand to the contents of the index register and reads the location at that address, so with IX holding 4 it reads address 204. LDI 300 uses indirect addressing: the processor reads address 300, finds an address stored there, and then reads the contents of that second address - two memory accesses rather than one. Direct addressing, LDD 200, is the instruction that simply loads the contents of address 200, and immediate addressing, LDM #200, would load the number 200 itself.

The answer that scoresA computer needs an operating system because the resources of the machine are limited and shared. The OS allocates processor time and memory to each program and keeps programs apart, so that one cannot overwrite another's data; it manages files and storage, decides who may access them, and controls the peripherals through device drivers so that a program can print without knowing how the printer works. It provides the user interface and the platform that application software is written against, and it handles errors so that a single faulty program does not stop the machine. A word processor is application software: it needs the operating system, not the other way round.

The answer that scoresA translator converts the program into machine code. An assembler translates an assembly language program, a compiler translates a whole high-level language program into an executable before it is run, and an interpreter translates and executes a high-level language program one statement at a time. The operating system's part comes afterwards: it loads the executable (or loads the interpreter, which then loads the source file), allocates it memory, and schedules the resulting process on the processor. Mixing the two up loses marks in both parts of the question.

The answer that scoresA compiler produces an executable file of machine code, and that file can be supplied to a user who has neither the compiler nor the source code; the code is therefore protected and the program runs quickly. An interpreter produces no separate translated file. It reads the source program and executes each statement as it translates it, so the user needs the interpreter installed and can read the source code. That makes an interpreter suitable for development and for teaching, where immediate feedback on the statement containing an error matters more than run-time speed, and a compiler suitable for software that is to be sold and run by people who must not see the code.

The answer that scoresDefragmentation is for magnetic hard disks. As files are created, resized and deleted, the blocks belonging to one file become scattered across the disk, so reading the file requires the read/write head to move to several places; defragmentation rewrites the blocks so that each file is contiguous and the free space is collected together, which reduces head movement and speeds up access. A solid-state drive stores data in flash memory with no moving parts and the same access time wherever the data is, so defragmentation brings no speed benefit and the additional writes shorten its life. The operating system uses TRIM on a solid-state drive instead, to tell the drive which blocks are no longer needed.

The answer that scoresWith a dynamic link library the routine's code stays in a separate file. The program records that it needs the DLL, and when the program runs the operating system loads the DLL into memory and the program calls into it. Because the code is not copied into the executable, the executable is smaller and takes less disk space; one copy of the DLL can be shared in memory by several programs running at once; and the DLL can be replaced with a corrected version without the programs being recompiled. A static library works the other way: the code is copied in at link time, so the executable is larger but does not depend on a separate file being present.

The answer that scoresEncryption converts plaintext into ciphertext so that the data cannot be read without the key, which protects confidentiality in transit and at rest. It does not guarantee integrity: an authorised user with the key can still change the data, and corruption on the disk or in transit still happens. Integrity is protected by validation at entry, verification during entry and transfer, access rights that limit who may write, and an audit trail that records who changed what.

The answer that scoresA range check confirms only that the value lies between a lower and an upper limit, for example that an age is between 11 and 18. An age of 15 for a customer who is really 16 passes the check, so validation cannot establish that the data is true. Confirming accuracy needs verification against the source, such as a visual check against the passport or a second entry of the same data.

The answer that scoresWith byte parity an extra bit is added to each byte so that the number of 1 bits is always odd, or always even. The receiver counts again and detects that an error has occurred, but not where it is, so the byte is re-sent and nothing is corrected. With block parity there is a parity bit for every row and every column, so a single flipped bit fails both its row check and its column check and the intersection gives its position, which allows the receiver to flip that bit back.

The answer that scoresPhishing is a fraudulent email or message that appears to come from a trusted organisation and persuades the recipient to click a link to a fake login page or open an attachment, so credentials are captured or malware is installed. Pharming needs no message and no click: the attacker poisons a DNS server's cache or alters the local hosts file so that a correctly typed domain name resolves to the attacker's server, and the browser shows the genuine URL while the user is on a copy of the site.

The answer that scoresA firewall controls which connections may pass between networks, using rules based on source and destination IP address, port number and protocol, so it stops an attacker reaching a service that should not be exposed and can block outbound traffic to a command server. It cannot inspect the content of traffic it permits, so a virus inside an allowed email or download is not stopped by it. Anti-virus software scans those files and attachments, matching known signatures and watching for suspicious behaviour, and quarantines anything it finds.

The answer that scoresFree software, as defined by the Free Software Foundation, means software that gives the user four freedoms: to run it for any purpose, to study and change it (which requires the source code), to redistribute copies, and to distribute modified versions. 'Free' refers to freedom, not price - free software can be sold, and much of it is. Shareware is different: it is a proprietary product distributed free of charge for a trial period or with some features disabled, and the user is expected to pay to continue using it. The source code is not supplied and copies must normally be passed on unaltered, so shareware is not open source and not free software.

The answer that scoresOpen source software remains the copyright of its authors; what changes is that the copyright is licensed on generous terms. Under a permissive licence such as MIT or Apache 2.0 a company may use the code even inside a closed-source product, provided it keeps the copyright notice. Under a copyleft licence such as the GPL, a company that distributes a modified version must release that version under the GPL with its source available, so it cannot take the code, improve it and lock the improvement away. Using the code outside the terms of the licence is copyright infringement, exactly as it would be with commercial software.

The answer that scoresCopyright protects the fixed expression of a work: the actual source code, the text, the image, the recording. It does not protect the idea, the algorithm or the function, which is why two programmers may each write a program that sorts a list, provided they do not copy each other's code. Buying software transfers a licence to use it under stated conditions, not the copyright: the buyer may not copy it for others, publish it, or adapt it beyond what the licence allows, and doing so is infringement. The copyright, and the right to grant further licences, stays with the owner - which is why the licence terms, not the price, decide what the buyer may do.

The answer that scoresA professional body such as the BCS or the IEEE represents a profession rather than negotiating pay. It publishes a code of conduct or code of ethics that members must follow, provides training, publications and continuing professional development, awards recognised qualifications such as Chartered IT Professional status, and investigates complaints against members. A member who breaches the code can be suspended or expelled, which can cost them the title their career depends on. Membership is voluntary, and that is the point: joining is a public commitment to a published standard, it gives the public and employers someone to hold the professional to, and it gives the member guidance and support when an employer asks for something unethical.

The answer that scoresAI is software that performs tasks normally needing human intelligence by learning patterns from data rather than following rules written for every case; a large but rule-based program is not AI. Its impact has to be separated into three categories. Socially it improves access to services and diagnosis but can inherit bias from its training data, enable surveillance and produce convincing fake content. Economically it raises productivity and creates new roles, while displacing routine cognitive work and concentrating market power in the firms that own the models and the computing power. Environmentally it supports climate modelling, energy optimisation and precision farming, but training and running large models consumes large amounts of electricity and water for cooling, and the hardware creates electronic waste and demand for mined minerals.

The answer that scoresA foreign key is a field in one table that holds the value of the primary key of another table, so it is how the relationship is implemented. It may repeat and it may be null, so it is not unique. In a one-to-many relationship such as one customer placing many orders, the foreign key CustomerID sits in the Orders table, the many side, and repeats once for each order that customer places. In a many-to-many relationship there is no single place for it: a junction table is created holding both foreign keys, and those two fields together form its composite primary key.

The answer that scoresThe table is in 2NF but not in 3NF because it contains a transitive dependency. CustomerName and CustomerTown both depend on CustomerID, and CustomerID is not the primary key, so they depend on the key only indirectly. To reach 3NF, the customer details move into their own table, Customer(CustomerID, CustomerName, CustomerTown) with CustomerID as the primary key, and Order becomes Order(OrderID, OrderDate, Total, CustomerID) with CustomerID as a foreign key referencing Customer. Each fact is then stored once, so changing a customer's town means changing one field.

The answer that scoresThe statement must name the records to be changed: UPDATE Customer SET Town = 'Bradford' WHERE CustomerID = 51; The WHERE clause identifies the single record by its primary key, so only that customer moves. Exactly the same rule applies to DELETE FROM Customer WHERE CustomerID = 51;, and to a query, where omitting WHERE returns every record instead of the ones asked for.

The answer that scoresDDL, the Data Definition Language, is used for all creation and modification of the structure: CREATE DATABASE, CREATE TABLE, ALTER TABLE, and the declaration of data types, primary keys and foreign keys. The DBMS records these definitions in its data dictionary. DML, the Data Manipulation Language, is used for all queries and maintenance of the data: SELECT, INSERT INTO, UPDATE and DELETE FROM. Both are written in SQL, which is the industry standard for each.

The answer that scoresEvery field in the SELECT list that is not inside an aggregate must appear in the GROUP BY clause: SELECT CustomerID, COUNT(OrderID) FROM Orders GROUP BY CustomerID; If the surname is genuinely needed, it has to be brought in from the Customer table with an INNER JOIN on CustomerID, and it still cannot be shown alongside a count of orders unless the grouping is by customer and the query returns one row per customer. The safe check is to compare the SELECT list with the GROUP BY list before writing anything else.

The answer that scoresIn a file-based approach the same data is duplicated across separate files owned by separate applications, so an update made to one copy leaves the others holding a different version and the data becomes inconsistent. Each program is written round the exact layout of its file, so widening a field means rewriting every program that reads it. There is no way to ask a new question without writing a new program, access is granted to a whole file rather than to individual fields, and nothing centrally prevents two users overwriting each other's changes or a program bypassing the validation rules.

The answer that scoresAn algorithm is a solution to a problem expressed as a sequence of defined steps - defined meaning that each step has only one meaning and the order is fixed, so that anyone following it gets the same result. It is language-independent: the same algorithm can be implemented in several programming languages, and the program is that implementation, written with the syntax of one language so that a computer can execute it. Abstraction is the process of removing or hiding the detail that is not needed for a particular purpose so that only the essential features of a problem or system are represented. Its purpose is to reduce complexity so the problem is small enough to understand, solve or model, and it produces an abstract model containing only essential details - such as a tube map that keeps stations and interchanges but discards true distances because the map exists for route planning.

The answer that scoresDecomposition is breaking a large problem down into smaller sub-problems, each of which can be solved separately, until each is small enough to be written and tested on its own. The sub-problems are chosen by function - what each part has to do - so a school results system splits into reading the marks in, validating them, calculating grades, producing the report and saving the results. Each sub-problem becomes a program module, either a procedure or a function. Because the split happens before coding, each module can be given to a different programmer, can be tested in isolation with its own test data, can be reused elsewhere in the program, and can be changed later without affecting the other modules, which is what makes a decomposed program easier to maintain.

The answer that scoresInitialise the counter and the total before the loop, change the counter inside it, and close the loop properly: Count <- 0, Total <- 0, then WHILE Count < 10 DO, INPUT Mark, Total <- Total + Mark, Count <- Count + 1, ENDWHILE. Because the number of repetitions is known in advance, a count-controlled loop is the better choice and removes the risk entirely: FOR Count <- 1 TO 10, INPUT Mark, Total <- Total + Mark, NEXT. The average is then calculated after the loop, Average <- Total / 10, and output once, not inside the loop.

The answer that scoresAbstraction removes the detail that is not needed for the purpose, so that only the essential features are represented, and what is essential is decided by the purpose of the model. A model of a car for a driving simulator must keep mass, engine power and grip because they change how it drives, and can discard paintwork and seat material because they do not. Remove something that does matter - wind resistance from the flight model, the overdraft limit from the bank record - and the model gives wrong answers, which is worse than no model. So the skill is not removing as much as possible; it is deciding correctly what is essential, and the benefit that follows is a model that is cheaper to build, easier to understand and test, easier to change, and safe to experiment with.

The answer that scoresA complete identifier table gives every identifier, a description and a data type, and every quantity has its own name: StudentMark (INTEGER - the mark for the current student, out of 100), MarkTotal (INTEGER - the running sum of all marks entered), StudentCount (INTEGER - how many students have been processed), AverageMark (REAL - MarkTotal divided by StudentCount), IsPass (BOOLEAN - TRUE when StudentMark is 40 or more). The names are meaningful and consistently styled, they are unique, and the data types are chosen to fit the values each one can take, with AverageMark as REAL because a mean need not be a whole number.

The answer that scoresThe telephone number should be a STRING, because it is not used in arithmetic, it may begin with 0, and it may contain spaces, brackets or a leading + for an international number; storing it as an INTEGER would drop the leading zero and could not hold the other characters. The price should be a REAL, because it can have a fractional part and is used in arithmetic such as adding up a total or working out change. If the program must avoid the rounding errors of binary floating point, the price can instead be stored as an INTEGER number of pence and divided by 100 only when it is displayed.

The answer that scoresA swap needs three assignments through a temporary variable: Temp <- Names[Index], then Names[Index] <- Names[Index + 1], then Names[Index + 1] <- Temp. The first statement saves the value that is about to be overwritten, the second moves the smaller value down, and the third puts the saved value into the upper position. The whole swap sits inside IF Names[Index] > Names[Index + 1] THEN ... ENDIF, and a Boolean flag is set to TRUE inside the same IF so that the outer REPEAT ... UNTIL Swap = FALSE loop knows another pass is needed.

The answer that scoresThe search must initialise both an index and a flag, and its loop condition must contain both conditions: Found <- FALSE, Index <- 1, then WHILE NOT Found AND Index <= 500 DO, comparing ProductCode[Index] with SearchCode, setting Found <- TRUE when they match and otherwise incrementing Index. After the loop, test the flag: if Found is TRUE then output Index, otherwise output a message such as "Code not found". Because the index is incremented while the value is not found, it ends at 501 for an unsuccessful search, which is exactly why the flag rather than the index decides the output.

The answer that scoresTo add a pupil without losing the existing records the file is opened FOR APPEND, which positions the write at the end and leaves the existing lines untouched; FOR WRITE would be used only when the whole file is being created again. To read it, the program opens FOR READ, then loops WHILE NOT EOF("pupils.txt") DO, reading one line with READFILE inside the loop and processing it, and closes the file with CLOSEFILE when the loop ends. Testing EOF before the read means the loop stops cleanly after the last line instead of reading past the end.

The answer that scoresA stack is last in, first out: each page visited is pushed onto the top of the stack, and the back button pops the most recent page off the top, which is exactly the page the user expects to see. A queue is first in, first out: items are enqueued at the rear and dequeued from the front, so the oldest item is removed first, which suits a print spooler where jobs must print in the order they were sent. The back button therefore needs a stack, because the most recent page must be the first one removed.

The answer that scoresREPEAT, then INPUT Answer, then UNTIL Answer = "Y". The condition after UNTIL is the exit condition: the body repeats while that condition is FALSE and stops as soon as it is TRUE, so the condition must describe the state that ends the loop. The pre-condition equivalent is WHILE Answer <> "Y" ... ENDWHILE, where the condition is the continuation condition - carry on while the answer is not Y. Same job, opposite-looking conditions, and the test sits at opposite ends of the body.

The answer that scoresFUNCTION Average(Total : INTEGER, Count : INTEGER) RETURNS REAL, with RETURN Total / Count as the last statement, and ENDFUNCTION. The call belongs in an expression, for example OUTPUT "Mean mark: ", Average(Sum, N), where the returned REAL replaces the call and is then printed by the OUTPUT that surrounds it. The header promises a REAL, the RETURN supplies exactly that type, and no CALL is used because a function is not a procedure.

The answer that scoresPROCEDURE AddMark(Mark : INTEGER, BYREF Total : INTEGER), with Total <- Total + Mark in the body and ENDPROCEDURE, called as CALL AddMark(56, Sum). BYREF before the parameter means the parameter refers to the caller's own variable, so the assignment changes Sum itself. The rule is to ask what the module must give back: if nothing, pass by value, which protects the caller's data; if it must change a value in the caller, that parameter needs BYREF, and a procedure that must return two changed values needs BYREF on both.

The answer that scoresFOR Index <- 1 TO 30 ... NEXT Index for an array declared ARRAY[1:30], because both bounds are included and the count is upper bound minus lower bound plus one. For five repetitions, FOR Index <- 1 TO 5 ... NEXT Index with nothing inside the body that touches Index, because NEXT is what advances the counter. The bounds should come from the declaration, and the counter should be left to the loop: a FOR loop whose counter is assigned inside the body no longer has a known number of repetitions, which was the only reason for choosing it.

The answer that scoresIf the branches depend on two variables or on compound conditions, use nested IF statements, one condition per IF, each closed with its own ENDIF. If the choice really is one identifier against a list, use CASE OF Identifier with labels that are single values or ranges such as 80 TO 100, and finish with OTHERWISE : followed by the default action, so an unexpected value is handled deliberately rather than falling through. Ranges in a CASE must not overlap either, because the clauses are tested in sequence and the first match wins.

The answer that scoresA syntax error breaks the rules of the language, so the translator cannot produce code and reports the fault at compile time - a missing ENDIF, an unbalanced bracket, a mistyped keyword. A logic error is valid code that does the wrong thing: it uses < where it needed <=, so a loop runs one iteration too few, or it adds where it should multiply. The program runs to completion and gives an incorrect result, so the fault is found by comparing output with the expected result, by a dry run with a trace table, or by a walkthrough - never by the translator. A run-time error is the third type: the code is valid and the logic may be right, but when a particular instruction executes with particular data, such as a division by zero or an array index out of range, the program halts.

The answer that scoresA dry run is a person working through the pseudocode or code by hand, keeping a trace table with a column for each variable and a row for each step, writing down the value of every variable each time it changes, and comparing the final output with the expected output. No computer is used, so it can be done on the design before the program exists and it exposes logic errors early. Running the program slowly, or stepping through it in a debugger, is a different technique: the computer is executing the code and you are watching the values it produces.

The answer that scoresA structure chart shows the solution decomposed into a hierarchy of modules: the main program at the top, the sub-tasks it calls below it, and their sub-tasks below them, with each box labelled with what that module does. The lines show calls, and the labelled arrows show the parameters passed - data passed down to a module and values returned up to the module that called it. It does not show the order in which instructions execute, and it contains no selection or iteration. The order of execution and the decisions belong in pseudocode or a flowchart for each individual module, which is why the two diagrams are used together, not instead of one another.

The answer that scoresAlpha testing is carried out in-house by the developers and their own testers before release, in controlled conditions where the people who wrote the code are available to investigate anything that goes wrong. Beta testing releases the near-final version to real users outside the organisation, who use it on their own hardware with their own data and report the faults they find; it is the only way to discover problems caused by the variety of machines, configurations and patterns of use in the real world. Acceptance testing comes after both: the client tests the system against the requirements specification to decide whether it can be accepted.

The answer that scoresCorrective maintenance fixes a fault that has been found in use: a calculation that gives the wrong total, a report that crashes on one customer's record, a validation rule that lets through an impossible date. Perfective maintenance improves a system that is working as specified: making a search faster, using less memory, redesigning a screen that users find confusing, adding a report they have asked for, or restructuring the code so that future changes are easier. Adaptive maintenance is the third type and is caused from outside the system - a new operating system, replacement hardware, a new file format, or a change in the law such as a new tax rate. The test is to ask why the change is being made: something was wrong, something outside changed, or something working is being made better.

The answer that scoresA set is a composite data type holding an unordered collection of values of the same base type, in which each value can appear at most once. Because order carries no meaning, two sets with the same members are equal however they were built, and adding a value that is already present changes nothing. The operations are adding and removing a value, testing membership with IN, and combining sets by union, intersection and difference. A set is the right type when the question is membership, such as which optional subjects a pupil takes, and an array is the right type when order and repetition matter.

The answer that scoresA random file, also called a direct file, stores each fixed-length record at an address calculated from its record key by a hashing algorithm. To find a record the program hashes the key and goes straight to that address, so one disk read is enough however large the file is; there is no searching. The order of the records is meaningless, which is why a sequential report from such a file needs a separate sorted index, and the cost of the method is wasted space and the need to handle collisions, where two keys hash to the same address.

The answer that scoresThe mantissa holds the significant digits, so m mantissa bits give m - 1 significant binary digits and set the precision: with an 8-bit mantissa the gap near 1 is 2^-7, and with a 12-bit mantissa it is 2^-11. The exponent sets the range: e exponent bits give a two's complement exponent from -2^(e-1) to 2^(e-1) - 1, so 4 bits give -8 to +7 and 8 bits give -128 to +127. Moving a bit from the mantissa to the exponent therefore widens the range of values that can be held but makes the values coarser, so the two cannot be increased together in a fixed word length.

The answer that scoresThe mantissa 00010110 is 0.0010110, or 0.171875, and the exponent 0100 is 4, so the value is 0.171875 x 16 = 2.75. The mantissa begins 00, so shift it left twice to 01011000 and decrease the exponent by two, from 4 to 2. The normalised pattern is 01011000 0010, and the check is 0.1011 x 2^2 = 0.6875 x 4 = 2.75, which is the original value. The rule is that a normalised positive mantissa begins 01 and a normalised negative mantissa begins 10.

The answer that scoresThe denary fraction 0.1 is 0.0001100110011... in binary, which recurs, so it cannot be represented exactly in a finite number of bits. With an 8-bit mantissa it is normalised to 0.1100110 x 2^-3 and stored as 01100110 1101, which is 0.099609375 - an error of about 0.00039. That is a rounding error, and it accumulates when such values are added repeatedly or multiplied, so a program that tests two calculated values for equality may get the wrong answer. The consequences are handled by comparing with a tolerance, by holding money in integer pence, and by allowing enough mantissa bits for the accuracy the problem needs.

The answer that scoresTCP/IP is the protocol suite used by the internet, organised as four layers: Application, Transport, Internet and Link. Email is handled at the application layer by SMTP, which sends the message from the client to its mail server and between mail servers. The transport layer, using TCP, splits the message into numbered packets, adds port numbers and a checksum, acknowledges delivery and retransmits anything missing. The internet layer, using IP, adds the source and destination IP addresses and lets routers forward each packet, and the link layer puts each packet into a frame with MAC addresses for transmission over the physical medium.

The answer that scoresSMTP is the push protocol: it sends email from the client to the outgoing mail server, and then from one mail server to another until the message reaches the server holding the recipient's mailbox. POP3 and IMAP are the pull protocols used by the recipient's client to collect the message. POP3 normally downloads the messages to one device and deletes them from the server, so nothing is synchronised. IMAP leaves the messages on the server and synchronises folders and read/unread state, so the same mailbox can be read on a phone and a laptop.

The answer that scoresCircuit switching reserves a dedicated path for the whole duration of a connection, so bandwidth and delay are guaranteed and data arrives in order, which suits continuous real-time traffic such as a telephone call or live video. It is unsuitable for the internet's bursty data traffic because the reserved capacity is wasted during silences, the connection takes time to set up, the number of simultaneous connections is limited, and the loss of any link on the path ends the connection. Packet switching is used instead: links are shared only while packets are actually being sent, and packets can be rerouted around a failure.

The answer that scoresA router receives a packet, reads the destination IP address from its header, and looks that address up in its routing table to find the best next hop. It then forwards the packet to the next router on that route, and the next router repeats the process, so the packet crosses the internet one hop at a time. The destination IP address identifies the network and host that the packet is ultimately for, while the MAC addresses in the frame identify only the two devices at the ends of the current link, and they are replaced at every hop. If a link is congested or has failed, the router can choose a different next hop.

The answer that scoresIn packet switching each packet is sent independently and routers decide the next hop for each one, so packets from the same message may travel by different routes and arrive out of order, with different delays. Each packet carries a sequence number, and the transport layer at the destination uses it to reassemble the message in the correct order. Packets that fail the error check are discarded, and TCP arranges for missing or corrupt packets to be retransmitted. Delay that varies from packet to packet is called jitter, and it is why real-time audio and video need priority handling.

The answer that scoresA RISC processor uses a small set of simple, fixed-length instructions, so a program needs more of them, which makes the code larger and puts more work on the compiler. Against that, each instruction is short and predictable: most take one clock cycle, all take the same time to decode, and the processor can overlap them in a pipeline so that one instruction completes every cycle. Fixed length and simple addressing also keep the pipeline free of stalls. A CISC processor needs fewer, longer instructions, which keeps the code small, but each may take several cycles and the varying lengths make pipelining harder. Neither is simply better: RISC wins on power and predictable speed, which is why it dominates phones and embedded systems, while CISC wins on code size and compatibility with existing software.

The answer that scoresThe processor checks for interrupts at the end of each fetch-decode-execute cycle. If one is pending it finishes the instruction it is executing, then saves the processor state - the program counter and the contents of the registers - usually by pushing them onto a stack. It loads the address of the interrupt service routine from the interrupt vector table into the program counter and executes that routine to service the device. When the routine ends, the saved state is restored and the original program resumes at exactly the instruction it had reached, so from the program's point of view nothing has changed except the time taken.

The answer that scoresA full adder has inputs A, B and Cin and outputs Sum and Cout, and needs eight rows in binary counting order. Sum = A XOR B XOR Cin, so it is 1 when an odd number of inputs are 1; Cout = A.B + Cin.(A XOR B), so it is 1 when at least two of the three inputs are 1. For example, with A = 1, B = 1, Cin = 1 the outputs are Sum = 1 and Cout = 1, because 1 + 1 + 1 = 11 in binary. The half adder has only A and B and no carry in, which is why it cannot be used beyond the least significant column of a multi-bit addition.

The answer that scores(A.B)' = A' + B', and (A + B)' = A'.B'. The rule is: break the bar, change the sign underneath, and invert each term. Applying it to a longer expression works one bar at a time from the outside in, so (A.B + C)' = (A.B)'.C' = (A' + B').C'. Checking a single row of the truth table, such as A = 1, B = 0, C = 0, confirms it: the left side is (0 + 0)' = 1 and the right side is (0 + 1).1 = 1.

The answer that scoresOnly blocks of 1, 2, 4, 8 or 16 cells in a rectangle may be grouped, and each block must be as large as possible; blocks may overlap and may wrap round the edges of the map. In a three-variable map, a block of two removes one variable, a block of four removes two, and a block of eight removes all three, leaving a constant. A diagonal pair cannot be grouped, so the 1s it contains must be covered by legal blocks - enlarging a group, overlapping an existing group, or wrapping to the other side of the map - before the expression is written.

The answer that scoresA virtual machine is a software implementation of a computer that runs on shared physical hardware under a hypervisor. Its processor time, memory and storage are allocated from the host, so a VM can never exceed the host's real resources and is usually slightly slower than running the same software directly, because some instructions have to be trapped and translated and because it competes with the other guests. What the VM does give is flexibility: several operating systems on one machine, isolation from the host, snapshots and cloning, and machines that can be created, moved or destroyed in minutes - with the trade-off that the host is a single point of failure and that a hypervisor flaw exposes every guest.

The answer that scoresVirtual memory is a memory management technique in which the pages of a process that are not currently needed are held on disk in a page or swap file, while the active pages stay in main memory. A page table maps each virtual page to a frame, and when the processor refers to a page that is not present a page fault occurs: the OS finds a frame, loads the page from disk, updates the page table and restarts the instruction, blocking the process meanwhile. This allows a program larger than physical memory to run and allows more processes to be resident, but it does not add RAM and it slows execution, because a page fault is a disk access. If faults happen constantly the system thrashes.

The answer that scoresPaging divides the address space into fixed-size pages and physical memory into frames of the same size, invisible to the programmer, mapped by a page table from page number to frame number. Because all pages are the same size there is no external fragmentation, though the last page of a program wastes part of its frame. Segmentation divides the program into variable-length segments that match its logical structure, such as code, data and stack, with a segment table holding a base address and a limit for each. Segments are visible to the programmer, can be shared between processes and can be given their own protection, but their variable size leaves gaps in memory, so external fragmentation can occur and compaction may be needed.

The answer that scoresAn interpreter translates and executes the source one statement at a time and produces no translated version, so there is no compile step and the program starts straight away. It also stops at the first error it reaches, pointing at the exact statement, which makes it much easier to debug - the reasons it is used for teaching and development. But translation is repeated on every run, so an interpreted program is slower than a compiled one, and the source code must be present, so it cannot be hidden. A compiler translates the whole program once into object code, producing an executable that runs at machine-code speed without the compiler or source and can be distributed commercially, at the cost of a compile step, a list of errors rather than one, and the need to recompile for a different platform.

The answer that scoresA blocked process is waiting for an event outside the processor, such as an I/O operation completing. When the device finishes it raises an interrupt, the kernel's handler notes that the event has occurred, and the process is moved from blocked to ready - it is now able to run but does not have the processor. It runs only when the scheduler selects it from the ready queue and performs a context switch, restoring its program counter and register values from its process control block. So the sequence is blocked -> ready -> running, and the time spent in ready depends on the scheduling routine and on the priorities of the other processes.

The answer that scoresLexical analysis reads the source as characters and groups them into tokens - keywords, identifiers, constants and operators - discarding whitespace and comments and building the symbol table. Syntax analysis takes that token stream and checks it against the grammar of the language, expressed in BNF or syntax diagrams, building a parse tree and reporting syntax errors such as a missing bracket. Code generation then converts the parse tree into object code, and optimisation improves that code before it is finally assembled and linked into an executable.

The answer that scoresRead the expression left to right, pushing operands and applying operators to the top two stack values. Push 8, push 3; the `-` pops 3 first and 8 second, so the calculation is 8 - 3 = 5, and 5 is pushed. Push 4; the `*` pops 4 and 5 and multiplies them, so the result is 20. Writing the stack after each step makes the order explicit and prevents the error.

The answer that scoresThe sender obtains the company's public key, which is published and may be given to anyone, and encrypts the message with it to produce ciphertext. Only the matching private key, which the company keeps secret, can decrypt that ciphertext, so the message is confidential even though the ciphertext crosses a network the sender does not control. This scales to the public because the public key can be published to the whole world without weakening the encryption: it can lock data but cannot unlock it. The weakness is that the sender has no proof the public key really belongs to the company, which is the problem a digital certificate solves.

The answer that scoresA digital certificate binds a public key to an identity and is signed by a certificate authority that the client already trusts. It contains the subject's name or domain, the subject's public key, the issuer, the validity dates, a serial number and the CA's digital signature. During the TLS handshake the server presents the certificate; the client checks the CA's signature up the chain of trust to a root certificate in its own trust store and checks the domain name matches, so it knows the public key belongs to the real server and not to an impostor. The encryption is a separate step: the client uses that now-trusted public key to send a session key, or agrees one with Diffie-Hellman, and the data is then encrypted symmetrically. The certificate prevents a man-in-the-middle attack; it does not itself encrypt.

The answer that scoresSSL/TLS protects a connection between a client and a server: it gives confidentiality by encrypting the data with a symmetric session key, integrity by authenticating each record so alterations are detected, and authentication by the server presenting a digital certificate that the client verifies up the chain of trust. When the data reaches the server it is decrypted so the application can use it, and from then on it is only as safe as the server's own storage controls - access rights, database encryption and backups. TLS also does not protect against a server that is itself malicious or already compromised, and it does not hide the fact that a connection took place. A complete answer to 'how are customer details protected' names both transit encryption and encryption at rest.

The answer that scoresThe two types are used together, each for what it does best. Asymmetric cryptography solves the key distribution problem: the server publishes a public key inside its certificate, and the client uses it to send a randomly generated session key, or agrees one using Diffie-Hellman, without any shared secret existing beforehand. Symmetric cryptography then encrypts the actual data with that session key, using AES, because it is fast enough for bulk traffic. The asymmetric operation happens once per connection and the symmetric cipher handles everything else, so the connection has the security of the first and the speed of the second. Ephemeral session keys also give forward secrecy, because a new key is generated for each connection and discarded afterwards.

The answer that scoresQuantum key distribution distributes a symmetric key by encoding each bit in the polarisation or phase of a single photon. An eavesdropper cannot copy an unknown quantum state and cannot measure it without disturbing it, so interception introduces errors that the two parties detect when they compare a sample of the key over a public channel; if the error rate is too high the key is discarded and the process restarts. Its benefits are that security rests on physics rather than on computational difficulty, so it is not threatened by a future quantum computer, and that eavesdropping is detected rather than hidden. Its drawbacks are that it needs specialised and expensive single-photon hardware, that range is limited because photons are lost in fibre and cannot be amplified, that key rates are low, that it is point-to-point, and that it still requires an authenticated classical channel - so it does not remove the need for certificates.

The answer that scoresA expands the node with the lowest f = g + h, where g is the cost of the best route found so far and h is the estimated cost from that node to the goal. It returns a shortest path only if the heuristic is admissible, meaning h never overestimates the true remaining cost, because then the estimate can delay a node but can never make a cheap route look more expensive than it is. Straight-line distance is admissible for road travel, since no road is shorter than the straight line between its ends, and with such a heuristic A is both complete and optimal. With an inadmissible heuristic A* may still return a route quickly, but the route may not be the cheapest, so the guarantee is lost.

The answer that scoresDijkstra's algorithm is only valid when every edge weight is zero or positive. The reason is the finalising step: when the node with the smallest tentative distance is selected, the algorithm assumes that every other route to that node must pass through a node that is already further away and then add a weight, and with non-negative weights that can never produce a smaller total. A negative weight breaks that assumption, because a route that looks longer can become shorter when the negative edge is used, so a finalised distance may later turn out to be too large. Algorithms that can handle negative weights are different ones, and Dijkstra's algorithm cannot cope with a negative cycle at all.

The answer that scoresSupervised learning uses labelled training data: each example is supplied with the correct output, the model learns the mapping from input to output, and its performance can be measured by comparing its predictions with the labels. It covers classification, where the output is a category such as spam or not spam, and regression, where the output is a number such as a price. Unsupervised learning uses unlabelled data: the model is given the inputs only and looks for structure in them, typically by clustering similar records together or finding items that occur together. There is no correct answer to compare against, so the result has to be interpreted by a person, but no expensive labelling is needed.

The answer that scoresA trained network produces its output by a forward pass: the data enters at the input layer, each unit multiplies its inputs by the weights on its connections, adds a bias and applies an activation function, and the result passes forward layer by layer until the output layer gives the prediction. Back propagation happens during training. After the forward pass, the output is compared with the known correct output to give an error; that error is then propagated backwards through the layers, using the derivatives of the activation functions, to work out how much each weight contributed to it. Each weight is adjusted slightly in the direction that reduces the error, by an amount set by the learning rate, and the whole process is repeated for many examples over many epochs.

The answer that scoresMachine learning is the broad approach in which a model is learned from data instead of rules being written by hand, and it includes many methods, of which regression, decision trees and neural networks are examples. Deep learning is machine learning carried out with an artificial neural network that has many hidden layers. The depth is what matters: the layers learn a hierarchy of features, from simple edges or sounds in the early layers to whole shapes or words in the later ones, so the network can work from raw data without a person deciding which features to measure. Deep learning usually does need large labelled data sets and substantial computing power, which is why it became practical only once both were widely available, but the defining feature is the many-layered network.

The answer that scoresRegression is a supervised learning method whose output is a number on a continuous scale: a price, a temperature, a demand, a lifespan. Linear regression fits a straight line y = mx + c to the training data, choosing the gradient and intercept by least squares so that the total squared distance between the data points and the line is as small as possible; multiple regression extends this to several inputs, with one coefficient each. Classification is the other supervised task, and its output is a category, such as spam or not spam, so a spam filter is a classifier rather than a regression model. Logistic regression is the exception that explains the confusion: it produces a probability between 0 and 1, which is then turned into a class by a threshold, so despite its name it is used for classification.

The answer that scoresA binary search requires the data to be sorted into the order used by the comparisons, because it discards half of the remaining items at every comparison on the assumption that everything before the middle is smaller and everything after it is larger. It also requires the middle item to be reachable directly, which is true of an array but not of a linked list that must be walked from one end, and the items must be comparable. When those conditions hold, a search of 1 000 000 items takes at most 20 comparisons; when they do not, the only safe option is a linear search, or sorting the data first, which costs far more than the search saves on a single pass.

The answer that scoresA bubble sort without a swap flag always makes n - 1 passes and is O(n^2) whatever the data, which is one reason the flag is added. With the flag the sort ends as soon as a pass makes no swaps, so sorted data costs one pass of n - 1 comparisons, O(n), and data in reverse order still costs about n(n - 1)/2 comparisons and swaps, O(n^2). An insertion sort is also O(n) on sorted data and O(n^2) in reverse order, but it moves each item only as far as it needs to go, so on data that is nearly in order it does much less work than a bubble sort, which moves an item only one place per pass. On small arrays with a small constant cost per comparison, an insertion sort is usually the faster of the two, but the honest answer is that it depends on the initial order and on n.

The answer that scoresA queue is a first in, first out structure: enqueue adds an item at the rear and dequeue removes and returns the item at the front, moving the Front pointer on one place (with MOD in a circular queue). A stack is a last in, first out structure: push adds an item at the top and pop removes and returns the item at the top, decrementing the Top pointer. Both operations begin with a check: dequeue and pop must test for an empty structure and report underflow, and enqueue and push must test for a full one and report overflow.

The answer that scoresEvery recursive routine needs a base case - a condition that returns a result without making another call - and a general case that calls the routine with an argument that has moved closer to that base case. For example, Factorial(N) returns 1 when N = 1, and otherwise returns N * Factorial(N - 1), so the argument falls by one at every call until the base case is reached. Each call is given a stack frame holding its parameter and return address, and the frames are popped as the returns unwind back up to the original call; with no base case the frames are never popped and the stack is exhausted.

The answer that scoresBig O specifies the growth of the time taken or the memory used as the number of data items n increases, with constants and lower-order terms dropped. An O(n) algorithm's work grows in step with n, while an O(n^2) algorithm's work grows with the square of n, so as n becomes large the O(n) algorithm wins clearly - a linear search of 1 000 000 items may need 1 000 000 comparisons where a binary search needs about 20. For a small n, however, the constant cost of each step matters more than the growth: a bubble sort on twenty items can be quicker than a more elaborate algorithm with a large constant per item, which is why general-purpose libraries switch to a simple sort for small sections.

The answer that scoresA programming paradigm is a style or approach to writing programs, defined by the features a language provides and by the way the problem is broken down. Low-level, imperative (procedural), object-oriented and declarative programming are the four in this syllabus. A language is not a paradigm: Python can be written as a procedural program or as an object-oriented one, and Java can be written in a largely procedural style even though it is designed for objects. Low-level, imperative and object-oriented programming are all imperative, because the programmer writes the steps; declarative programming is not, because the programmer writes facts and rules and the system decides the order.

The answer that scoresImmediate addressing puts the value itself in the operand, so LDM #25 places the number 25 in the accumulator and no memory access is needed to fetch it; the value is fixed when the program is written. Direct addressing puts the address of the value in the operand, so LDD 25 would load the contents of memory location 25. Indirect addressing puts the address of an address in the operand, so LDI 25 would look in location 25, find an address stored there and load the contents of that second address. Indexed addressing adds the index register to the operand address, and relative addressing treats the operand as a displacement from the current instruction.

The answer that scoresA serial file holds records in the order they were written and has no ordering by any key, so it is read from the beginning when every record is being processed; a log of transactions or a backup is serial. A sequential file holds records in order of a key field such as an account number, is read in that order, and is updated in a batch by merging the master file with a transaction file to produce a new master file. A random file holds fixed-length records at positions that can be calculated from the key, so SEEK moves the pointer straight to a record and it can be read or replaced without reading the records before it, which is what makes it suitable for looking up and updating one record at a time.

The answer that scoresValidation checks input against the rules of the problem - a mark between 0 and 100, an age that is a positive whole number - and it is written with IF statements as part of the normal logic. Exception handling deals with an error or unexpected event raised while the program runs, such as a file that does not exist, a value that cannot be converted to a number, or a division by zero. It is appropriate where the failure is outside the program's control, and the TRY block should contain only the operations that can fail, with the EXCEPT block doing something useful: reporting the problem, using a default, asking for another file name or closing files cleanly.

The answer that scoresInheritance is an is-a relationship: a SavingsAccount is a BankAccount, so the subclass inherits the attributes and methods of the superclass, adds an interest rate, and overrides CalculateInterest with its own version. Containment or aggregation is a has-a relationship: a Car has an Engine, so the Engine is an attribute of the Car and the Car's methods call the Engine's methods. A Car cannot inherit from an Engine, because a car is not an engine; inheriting there would give the Car the Engine's attributes directly and break as soon as the Engine class changed.