Guide to topic selection

CSTA school standards

CSTA is the Computer Science Teachers Association. PK–12 covers the years from pre-kindergarten to grade 12. PK stands for pre-kindergarten and K for kindergarten. That division should not be treated as the same as the grades of the Azerbaijani school system. [1]

CSTA’s 2026 standards organise school computing around five strands:

• Algorithms and design;
• Programming;
• Data and analysis;
• Systems and security;
• Computing and society.

Artificial intelligence is integrated into these strands at a basic level. The earlier page also stressed that the same standards treat AI as a priority. The two statements do not conflict: integration describes the basic level, and priority describes the emphasis. [1]

For deeper study in the upper years the standards name artificial intelligence, cybersecurity, data science, game development, physical computing and software development. These are school-level specialisation tracks, not university degrees.

Physical computing is a program working with the real environment through sensors and electronic devices. A device that measures soil moisture and then starts watering is a project in this direction.

CS2023 university guidance

CS2023 is Computer Science Curricula 2023, guidance for computer-science programmes. The document is joint work by three professional organisations.

CS2023 is aimed mainly at undergraduate study. Although the name says 2023, the final document was approved in 2024. It covers 17 knowledge areas, including algorithms, artificial intelligence, data management, networks, operating systems, security and software engineering. [2], [23]

This approach also takes in applying knowledge, collaboration, responsibility and professional conduct. When you plan a piece of material it is therefore useful to ask not only “What information am I presenting?” but also “What will the learner be able to do after this?” [24]

AP Computer Science Principles

AP is Advanced Placement, a course and examination programme that gives school students work at the start of university level. CSP is Computer Science Principles. The College Board offers the programme. [3]

AP CSP brings together five Big Ideas:

• Creative development;
• Data;
• Algorithms and programming;
• Computer systems and networks;
• The impact of computing.

Besides programming, the course builds a first picture of how the internet works, how to draw conclusions from data, and how technology affects society. CSTA guides school computing, AP CSP guides preparation at the start of university, and CS2023 guides undergraduate study. That difference matters when you choose how deep a topic should go. [1]–[3]

Stack Overflow and GitHub

Stack Overflow is a platform where people discuss programming questions. Its Developer Survey is an annual survey of how programmers use tools, how they work and what they think. The 2025 results used 49,009 responses from 177 countries. Because people were reached mainly through the platform’s own channels, the results should not be treated as a complete picture of all programmers. [4], [25]

Of those who answered the question on using AI, 84 percent said they were using these tools or planned to use them. On a separate question about trust in accuracy, 46 percent expressed distrust and 33 percent expressed trust. The 84 percent figure does not count only people who already use the tools. [26]

From these results one can draw teaching topics such as “How can you check code written by AI?” and “Why can a convincing answer still be wrong?” Those are not recommendations from the survey itself. They are a teaching reading of the results.

GitHub is a platform for storing program code and working on it together. Octoverse is the name of its yearly activity report, not an abbreviation. The 2025 report highlights the growth of AI projects and says that in August of that year TypeScript overtook Python and JavaScript on GitHub’s usage measure. It also reports that Python is used in about half of new AI repositories. A repository is the place that holds a project’s code and the history of changes. [5]

Difference. The earlier page said TypeScript was the most-used language in 2025 and did not name the platform. This page uses the GitHub Octoverse 2025 wording. Both sources point to the same report. These figures describe activity on a platform. They do not directly measure labour-market demand, educational value or how hard a topic is to learn.

Learning research

An empirical study is research based on real observation and collected data. In teaching this may mean interviews with learners, analysis of task answers, and measurements before and after a class.

A concept inventory is a diagnostic test that asks how well core ideas are understood and what misconceptions are present. Wrong answer choices are written to reveal those thought-errors. A tool of this kind for dynamic programming used misconceptions found in earlier studies. [27]

As an example, look at these commands in order:

a = 5
b = a
a = 8

At the end the value of b is 5. b was given the value that a had at that moment; the later command changes only a. The answer “b becomes 8 as well” can show that assignment was understood as a link that stays updated.

A test must be checked for fit with age, language and the topics that are taught. SCS1 — Second CS1 — is an assessment of ideas in introductory programming; CS1 is a name for a first computer-science or programming course. Research on SCS1 stresses the limits of the tool and the value of measuring before and after teaching if you want to judge learning gains. [28]

Thinking like a computer

1. What an algorithm is — and how to turn an everyday problem into steps

An algorithm is a finite list of clear steps, carried out in a definite order, to reach a goal. The hard part is not the word itself, but turning an everyday task into steps that a machine can follow without guessing.

Example. To make tea: boil water, put a bag in a cup, pour the water, wait, remove the bag. If a step is missing — “wait how long?” — the algorithm is incomplete.

2. Decomposition, patterns and abstraction

Decomposition means breaking a large problem into smaller ones. A pattern is what repeats. Abstraction means keeping the features that matter for the goal and setting the secondary details aside.

Example. When you choose the shortest route to school, the links between streets and the distances matter; the colour of the buildings usually does not.

Another example. Planning a school trip: tickets, food and a timetable (decomposition); every bus ride needs a start time (pattern); ignore seat colours and keep “how many seats” (abstraction).

First programs

3. Variables as changing stored values

A variable is a named place that holds a value the program can change later. It is not a mystery symbol from algebra; it is storage with a name.

Example. score = 0, then later score = score + 10. The place named score now holds 10.

Another example. a = 5, b = a, a = 8. At the end b is still 5, because b was given the value that a had at that moment.

4. The difference between assignment and mathematical equality

Assignment writes a value into a variable. In many languages the equals sign does not claim that both sides are already the same, as in mathematics.

Example. x = x + 1 is nonsense as an equation, but as assignment it means “take the current x, add one, store the result back in x.”

5. Conditions and Boolean expressions

A condition asks a yes-or-no question. A Boolean value is only true or false. The program chooses the next step from that answer.

Example. If temperature > 30, write “hot”; otherwise write “normal”.

6. Nested conditions and compound logic

Conditions can sit inside other conditions, and several tests can be joined with “and” or “or”. The difficulty is keeping the order of decisions clear.

Example. If it is a weekday and it is raining, take the bus; if it is a weekday and it is dry, walk; otherwise stay home.

7. How loops work and why an infinite loop starts

A loop repeats the same steps. It must have a way to stop. An infinite loop starts when the stopping condition never becomes true.

Example. “While the tank is not full, pour water” works only if each pour actually raises the level. If the tap is closed, the loop never ends.

8. A function versus a command that only prints a result

A function is a named piece of work that can take inputs (parameters) and give back a result (a return value). Writing a number on the screen is not the same as returning it for later use.

Example. A function add(2, 3) should return 5 so another part of the program can store it, not only display “5”.

9. Systematic debugging instead of changing code at random

Debugging is finding errors by testing a hypothesis, not by changing lines at random. First you fix the expected result, then you compare it with the actual result. This is hard to learn and hard to teach, because experienced programmers often rely on methods they do not spell out. [10]

Example. If a total is wrong, check the first input, then the addition, then the display — one step at a time — instead of rewriting the whole program.

How computers store information

10. Binary numbers, bits, bytes and hexadecimal

Computers store information as bits — 0 or 1. Eight bits make a byte. Hexadecimal (base 16) is a shorter way to write the same binary values.

Example. In a common text encoding the letter A is the byte 01000001, often written as 41 in hexadecimal.

11. How text, images, sound and video are represented digitally

A computer does not “see” a picture or “hear” a song. It stores numbers that stand for samples of colour, brightness or air pressure, then rebuilds the media from those numbers. A picture can be explained as pixels and their colour values.

Example. A black-and-white image can be a grid of 0s and 1s: 0 for white, 1 for black.

12. Lossy and lossless compression

Compression makes a file smaller. In lossless compression the original data can be restored completely. In lossy compression some of the data is thrown away so the file becomes smaller.

Example. A ZIP archive of text is usually lossless. A strongly compressed photo may look slightly softer because some colour detail was discarded.

Networks and safety

13. The difference between the Internet, the Web and a browser

The Internet is a system of networks linked to one another. The Web is one service that runs on that infrastructure: pages tied to addresses. A browser is the program that lets you look at that web content.

Example. The Internet is the road system; the Web is the shops along the road; the browser is the car you use to visit a shop.

14. IP addresses, DNS, routers, packets and HTTP

An IP address is a machine’s network number. DNS (Domain Name System) maps a domain name to a network address. A router forwards packets — small slices of data. HTTP (Hypertext Transfer Protocol) sets the request-and-response rules for web resources.

Example. When you type a page address, DNS is asked for the number, then HTTP asks that machine for the page, which arrives in packets.

15. Passwords, phishing, malware, privacy and two-factor authentication

Phishing is an attempt to obtain a password or another secret through a deceptive message or site. Malware is software written to harm or spy. Two-factor authentication (also called two-factor confirmation) uses two checks from different categories.

Example. On a school account the password is the first check; a confirmation through a separate device or app can be the second factor.

Another example. A message that looks like a school portal but asks you to “confirm your password” on an unknown site is a common phishing pattern.

16. Encryption, hashing and digital signatures

Encryption protects the confidentiality of data: only someone with the right key can read it. Hashing makes a short digital fingerprint of the data; you cannot usefully turn that fingerprint back into the original. A digital signature helps check integrity and origin.

Example. A messaging app may encrypt the text on the way. A site may store a hash of your password rather than the password itself.

AI and data

17. How AI differs from ordinary software

Ordinary software follows rules a person wrote. Many AI systems, especially machine learning, build their behaviour from training data — large sets of examples — rather than from a complete handwritten rule book. Not every AI method works in the same way.

Example. A spam filter is not given every possible junk sentence. It is shown many examples of spam and ordinary mail, then it estimates which new message looks more like spam.

18. Training data and the role of examples in machine learning

The examples used in training shape what the system can do and where it will go wrong. If the examples are narrow or biased, the answers will be too.

Example. A model trained only on summer photos may mislabel snow as a defect in the picture.

Another example. A model trained only on pictures taken in one setting may perform poorly in another setting.

19. Why AI can hallucinate or repeat bias

A hallucination is when an AI system produces information that sounds convincing but is false or unsupported. The system is not looking up a checked fact; it is predicting a likely sequence of words. It can also repeat bias present in the training data. An explanation should show the difference between an answer that sounds right and an answer that has been checked.

Example. A chatbot may invent a paper title that sounds real, or repeat a stereotype that appeared often in the data.

20. How to check information and code produced by AI

Treat AI output as a draft. Check claims against a reliable source, and run code before you trust it. This is listed as a school-level skill, not only a specialist one. The 2025 Stack Overflow results also support this check as a practical topic. [4], [26]

Example. If a model writes a date or a formula, open the cited page or test the formula with a known example.

21. Collecting data, graphs, misleading pictures and bias

Data can mislead even without AI. A graph can hide the scale, a sample can miss a group, and a colourful chart can make a weak pattern look strong.

Example. A bar chart that starts at 90 instead of 0 can make two close results look far apart.

Making things

22. The basics of websites, games, mobile apps and robotics

These are the first practical settings in which the ideas above meet a real audience: a page that loads, a game that keeps score, a phone app that stores a setting, or a robot that senses and moves. They are listed as fundamentals, not as a demand that every pupil build all four. A simple game project can explain conditions, loops and events together.

Example. Even a simple game needs a variable (the score), a loop (the game continues) and a condition (the player has won).

Computing and society

23. Copyright, digital identity, disinformation and responsible use

Computing is not only technique. Pupils also need to know who owns a work, how an online identity can be copied, how false claims spread, and what responsible use looks like. Showing the sources of images, sound and text in a project ties copyright to the technical work.

Example. Putting an image into a school project without credit can breach copyright even if the image was easy to download.

How programs run

1. Program execution models and tracing state

A program is a sequence of changes to state — the current values of variables and the next instruction. Tracing means writing down that state after each step. Many later misconceptions are linked to the lack of a mental model of this process. [8], [9]

Example. For x = 2; x = x * 3 a trace is: start with no x; after the first line x is 2; after the second line x is 6.

2. Scope, lifetime, parameter passing and aliasing

Scope is where a name is visible. Lifetime is how long a stored value or object exists. Parameter passing is how inputs enter a function. Aliasing means two names refer to the same object, so a change through one name is visible through the other.

Example. If a function receives a list and appends an item, the caller’s list may change too — both names pointed at the same list.

Another example. If two variables point at the same object, a change to the object is visible through both references. Pointing one of the references at a different object is a different operation.

3. References, pointers and memory management

A reference or pointer holds the location of a value, not the value itself; the exact behaviour depends on the language. Memory management is deciding when that storage is created and released. Indirection can remain difficult even after several years of study. [14]

Example. Two variables can point at one customer record. Updating the address through either variable changes the same record.

4. Recursion and the recursive call stack

Recursion is a function calling itself, directly or indirectly, until a base case that needs no further call. Each call waits on the call stack until the inner call finishes. An explanation should show both the new calls being created and the results coming back. [13]

Example. factorial(4) waits for factorial(3), which waits for factorial(2), until factorial(1) returns 1.

5. Object identity, classes, objects, inheritance and polymorphism

A class is a description; an object is one instance. Identity asks which object you have, not only whether two objects hold the same numbers. Inheritance lets a class reuse another’s structure. Polymorphism lets the same interface produce different behaviour in different objects. A constructor takes part in setting an object’s starting state when it is created. [12]

Example. Circle and Rectangle can both answer “area”, but each calculates it differently.

Data structures and algorithms

6. Lists, linked structures, stacks, queues, trees, heaps and graphs

A data structure is a way of organising values so that some operations stay cheap. A list keeps order; a stack is last-in, first-out; a queue is first-in, first-out; trees and heaps organise hierarchy; a graph models connections. Here a heap is a structure; the same word is also used for a region of memory. [11]

Example. A browser’s Back button behaves like a stack: the last page you opened is the first one you return to.

7. Sorting, searching and graph algorithms

These are standard methods for putting items in order, finding an item, or walking along connections. A student needs to see both the method and when it is the right method.

Example. Finding a name in a sorted class list can use binary search: repeatedly cut the remaining list in half.

8. Big-O, worst case and the space–time trade-off

Big-O is used to give an upper bound on how an algorithm’s demand for resources grows as the input grows; it is not a count of exact seconds. Worst-case analysis asks about the most expensive legal input. A space–time trade-off means using more memory to save time, or the reverse.

Example. Looking at every item once is linear time. Looking at every pair of items grows much faster as the list lengthens.

9. Greedy algorithms, divide and conquer, and dynamic programming

A greedy method takes the locally best step. Divide and conquer splits a problem, solves the parts, and combines them. Dynamic programming stores answers to overlapping subproblems so they are not recomputed. Choosing the technique and writing the recurrence are common difficulties. [20], [27]

Example. Making change with the fewest coins can be greedy if the coin system allows it; some systems need a more careful stored-solution method.

Data and databases

10. Relational modelling, keys, joins and normalisation

A relational database stores data in tables. A key uniquely identifies a row. A join combines rows from more than one table. Normalisation organises tables so that repetition and update problems are reduced.

Example. A Students table and a Courses table meet in an Enrolment table. Enrolling the same student on several courses shows why the result can contain several rows.

11. Aggregation, subqueries and window functions in SQL

SQL (Structured Query Language) is the language used to ask questions of a database. Aggregation summarises many values as a sum, a count or an average. A subquery sits inside another query. A window function computes across related rows while still keeping the separate rows in the result. Large-scale analysis found frequent student errors with joins, subqueries and GROUP BY. [16]

Example. University student and enrolment tables can explain a JOIN. First show that the join increases the number of rows, then show how GROUP BY gathers those rows by course or by student.

12. Transactions, isolation, locking and database consistency

A transaction treats related operations as one piece of work: they should finish together or fail together. Isolation decides how far parallel transactions may see one another’s intermediate results. Locking prevents a clash on the same data. Consistency means the stored facts still obey the rules after the change.

Example. A bank transfer should not subtract from one account unless it also adds to the other.

Systems

13. Processes, threads, scheduling and virtual memory

A process is the environment of a running program; it usually has its own memory. A thread is a line of execution inside that environment. Scheduling decides which work the processor runs next. Virtual memory separates the addresses a program sees from physical memory. Students may memorise these mechanisms and still struggle to choose between solutions. [17]

Example. A browser can keep one process per tab so that one crashed page does not take down the others.

14. Race conditions, synchronisation, deadlock and nondeterminism

When several pieces of work overlap, the result can depend on the order of the steps. A race condition is that dependence on order. Synchronisation coordinates threads. Deadlock (also called tıxac in the earlier Azerbaijani page) is a standstill in which each side waits for a resource the other holds. Nondeterminism means an incorrect program may appear to work on some runs. One correct run does not prove that every legal order is correct. [15]

Example. Two threads both read “1 seat left” and both sell the last seat.

Another example. Splitting the update of the same counter by two threads into steps shows a race condition.

15. Network protocols, routing, reliability and congestion

A protocol is an agreed conversation between machines. Routing chooses a path. Reliability asks whether lost data is detected and resent. Congestion is overload on a shared path.

Example. A video call may drop picture quality when the path is congested rather than freeze forever.

16. Authentication, authorisation, encryption and secure software design

Authentication answers “Who are you?” Authorisation answers “What are you allowed to do?” Encryption protects data on the way or in storage. Secure design treats these as part of the system from the start. In cybersecurity, vulnerability, threat, risk and a protective measure are often mixed up. [19]

Example. Logging in proves identity; being a student still does not grant the right to change another student’s marks.

17. Distributed systems, replication, consistency and consensus

In a distributed system the parts run on different computers and communicate over a network. Replication keeps copies of data in more than one place. Those copies can disagree, so the system needs a consistency rule and sometimes a consensus method for agreeing on the next value.

Example. Two copies that are temporarily out of contact are a good starting point for explaining how data is brought back into agreement.

Another example. If two people type at once in a document saved on several servers, the system must decide whose edit remains.

18. Cloud computing, containers, orchestration and deployment

Cloud computing rents servers, storage and networks as a service. A container makes it easier to run an application together with what it depends on. Orchestration coordinates the placing, starting and managing of many containers. Deployment is putting a new version into service.

Example. A course website may run in a container so that the same image works on a laptop and in the cloud.

Machine learning and large models

19. Evaluating machine learning, overfitting and underfitting

Overfitting means the model fitted the peculiarities of the training data too closely and then fails on new examples. Underfitting means the model did not learn the main relationships well enough. Evaluation must use data the model has not already seen.

Example. A model that names every student in last year’s class but cannot place a new student has overfitted.

20. Data leakage, class imbalance, model bias and explainability

Data leakage is accidentally giving the model information it would not have at prediction time. Class imbalance means the example categories are unequal in number. Bias here can be a systematic distortion of results for certain groups. Explainability is being able to say, in a way people can follow, which factors affected the result.

Example. Mixing the test data into training is an example of leakage in evaluation.

Another example. If a medical model sees the hospital name, it may “predict” from that hospital’s typical patients rather than from the symptoms.

21. Tokenisation, vector representations, attention, retrieval and hallucination in large language models

An LLM is a large language model. Tokenisation splits text into the pieces the model processes; a token is not always a whole word. A vector representation (also called an embedding) expresses those pieces as numbers. An attention mechanism weighs relations among pieces in the context. Fetching information from an outside source can give an answer extra grounding, but it does not automatically guarantee correctness.

Example. The same word may become one token or several, which is why some models stumble on unusual spellings or codes.

Software work

22. Testing software, architecture, version control and continuous integration

Testing checks that code does what was intended. Architecture is the large-scale structure. Version control (often Git) records who changed what. Continuous integration (CI) regularly merges code changes and runs automated checks.

Example. A team can see who last changed a login page and whether the tests still pass after that change.

23. Functional, logic and concurrent programming paradigms

A paradigm is a general approach to building a program. Functional style emphasises functions and values; logic style emphasises relations and search; concurrent style emphasises work that overlaps in time. Concurrency is the overlapping progress of several pieces of work; parallelism is work that truly runs at the same time.

Example. The same “shortest path” problem can look like function composition in a functional language, like a constraint in a logic language, and like several searches overlapping in a concurrent model.

24. Compilers, interpreters and programming-language semantics

An interpreter carries out a program directly. A compiler first translates it into another form. Semantics is the meaning of the language’s constructions — what a program is allowed to do, not only how it is written.

Example. First tracing the same short program line by line, then asking “what exactly does this language do with x = x + 1?”, helps separate semantics from the mechanism of compilation or interpretation.

Theory of computation

25. Automata, computability, reductions, P, NP and NP-completeness

An automaton is a formal model described by states and transitions. Computability asks whether a problem can be solved by an algorithm at all. A reduction is a transformation that lets one problem be solved through another. P is the class of decision problems solvable in polynomial time. For NP, a suitable proof of a yes-answer can be checked in polynomial time. NP does not mean “not polynomial”. An NP-complete problem is in NP, and every NP problem can be reduced to it in polynomial time.

Example. If you can turn every instance of problem A into an instance of problem B, then a fast method for B would also give a fast method for A.