Marking the OCR A Level CS NEA: A Teacher's Guide

How to mark the OCR H446 NEA: marks per section, best-fit banding, red flags, moderation rules, and a JCQ/OCR-based checklist for checking AI authenticity.

Marking the OCR A Level CS NEA: A Teacher's Guide

By Yuness, qualified Computer Science teacher and former Head of KS5 Computer Science · Last reviewed 21 November 2026

The OCR A Level Computer Science NEA (the H446 Programming Project, components 03/04) is marked by you, the teacher, out of 70 across six criteria, and then moderated by OCR. You mark each criterion by best fit: read the whole section, choose the band whose descriptor fits the work overall, then decide where in that band it sits. You can only credit work that is the student's own, so marking now goes hand in hand with checking authenticity. That means following JCQ's AI guidance (revised 20 November 2026) and OCR's subject-specific advice on AI in the A Level CS NEA.

The OCR H446 NEA mark scheme at a glance

The Programming Project is worth 20% of the A Level. The marks below come from the OCR H446 specification and the OCR NEA cover sheet. A score of 0 in any criterion means there was no response, or nothing worthy of credit.

CriterionMaxBandsWhat it rewards
Analysis101–2, 3–5, 6–8, 9–10Computational framing, stakeholders, research into existing solutions, requirements, measurable success criteria
Design151–4, 5–8, 9–12, 13–15Decomposition, algorithms, usability, data structures and validation, test data
Iterative development of a coded solution151–4, 5–8, 9–12, 13–15Prototypes for each stage, modular structure, annotation, validation, review
Testing to inform development101–2, 3–5, 6–8, 9–10Testing at each stage, failed tests, justified remedial action
Testing to inform evaluation51, 2, 3–4, 5Post-development testing for function, robustness and usability
Evaluation of solution151–4, 5–8, 9–12, 13–15Success criteria cross-referenced to tests, usability, maintenance, limitations, line of reasoning
Total70

On the cover sheet the two development criteria are grouped as "Developing the coded solution" (25 marks), and the two testing-and-evaluation criteria as "Evaluation" (20 marks). Use the full band descriptors from the specification alongside this guide. The descriptors define the standard. This guide only helps you apply them.

Best-fit marking: what it means in practice

The specification says work does not have to meet every requirement of a band before it can be placed there, and that "marking should be positive, rewarding achievement rather than penalising failure or omissions." In practice:

  • One weak bullet does not pull a section down a band if the rest of the work is consistently in that band.
  • One strong bullet does not lift a section up a band on its own.
  • Once you have chosen a band, use its whole mark range. The specification asks assessors to use the full range of marks, so don't automatically give the middle mark.

Five golden rules before you start

These are my own rules of thumb for applying the specification. They are not OCR rules.

  1. Mark by best fit, not by counting bullets. Read the whole section, decide the band, then place the work within it.
  2. Use the descriptor's own language in your comments. If you are about to write "good design", stop and use the band wording instead, for example that the student has broken the problem down systematically and explained and justified the process. A moderator can then see exactly why you chose that band.
  3. Cross-referencing is the highest-value evidence. Look for explicit links from the Analysis success criteria to the Evaluation, and from the Design test plan to the testing done during development.
  4. Honestly recorded failures are a strength. The top band of Testing to inform development asks for evidence of failed tests and of remedial action with full justification. A student who shows a test failing and then being fixed is doing exactly what the criterion asks.
  5. Past-tense narration where the design should come first is a yellow flag. If the design section says "I made the menu" rather than "the menu will display", the design may have been written after the code.

Section-by-section marking checklists

The "top band" lists below paraphrase the specification. The red flags are patterns from my own marking practice. They tell you where to look harder. They are not OCR rules.

Analysis (10 marks)

To reach 9–10, look for: the problem's features described and justified as solvable by computational methods; stakeholders identified and described, with an explanation of why the solution suits their needs; research into the problem, including existing solutions, with suitable approaches justified; essential features and limitations explained; hardware and software requirements specified and justified; and measurable success criteria identified and justified.

Red flags: a generic stakeholder list with no needs described; no existing solutions discussed; unmeasurable criteria such as "the game will be fun"; a hardware list with no justification for each item. Note that research into existing solutions to similar problems already appears in the 3–5 band, so work with none at all will find it hard to move beyond the lower bands.

For background on the computational-methods side of Analysis, see our notes on computational methods (2.2.2).

Design (15 marks)

To reach 13–15, look for: a systematic decomposition with the process explained and justified; the structure of the solution defined in detail; accurate algorithms, justified as forming a complete solution; usability features justified; key variables, data structures or classes justified, including validation; and test data identified and justified for both iterative and post-development testing.

Red flags: finished implementation code where pseudocode should be; no algorithms or diagrams at all; wireframes with no annotation; test data described only as "I will test the program"; a list of variables with no reasoning.

Iterative development (15 marks)

To reach 13–15, look for: evidence of each stage, linked back to how the problem was broken down in the analysis; prototype versions for each stage; a well-structured, modular solution; code annotated to aid future maintenance; sensible names throughout; validation for all key elements; and review at all key stages.

Red flags: one giant code dump in a single "iteration"; linear development with no prototypes; names such as x, thing or stuff; iterations that are just screenshots with no commentary. Strong projects usually number their versions (v0.1, v0.2 or Prototype 1.1, 1.2) and tie each one to a sub-problem from the analysis.

Testing to inform development (10 marks)

To reach 9–10, look for: testing at each stage of development, evidence of failed tests, and remedial action with full justification.

Red flags: testing described only in prose; every test passing; bugs mentioned with no account of the fix; "I tested the program and it worked".

Testing to inform evaluation (5 marks)

For 5, look for: annotated post-development testing for both function and robustness, plus annotated usability testing. The 3–4 band asks for function and usability, so robustness is what separates the top mark.

Red flags: function tested but no boundary or erroneous cases; usability claimed but no user feedback shown; the student as the only tester.

Evaluation of solution (15 marks)

To reach 13–15, look for: test evidence cross-referenced to each success criterion, each judged met, partially met or not met; how unmet criteria could be addressed; usability features evaluated; maintenance, limitations and future development discussed. The top band also needs a well-developed line of reasoning that is clear and logically structured, because this is where the quality of extended writing is assessed.

Red flags: "the project was successful"; every criterion marked "Yes" with no reasoning; no maintenance discussion; "I would add more features" as the only future work.

Patterns to watch for

  • The all-pass test table. Every final test passes. A student who really debugged their code during development should have at least one boundary or robustness test that showed something. My rule of thumb is that an all-pass table rarely contains real robustness testing. Robustness testing is what separates the 5-mark band from 3–4, so if that evidence isn't there it is hard to justify the top mark for Testing to inform evaluation.
  • Terminology drift. The design pseudocode uses one set of names and the code uses another. This suggests the design was written after the code. OCR's AI guidance makes a related point: "Examiners will not reward reverse engineered designs."
  • The retrofitted analysis. The analysis is in the past tense throughout. Treat this as a reason to look closely at how the project developed and to talk to the student, not as an automatic deduction.
  • The single-stakeholder project. One stakeholder interviewed twice, or one stakeholder plus a generic survey. In my experience this rarely supports the full top band of Analysis.
  • Features deferred in advance. A student who writes "I will not implement X because of time" in the analysis is showing mature scoping, and Analysis can credit that. It doesn't remove the effect of features that were never built, so mark Development honestly.

Calibrating your marks and internal standardisation

If more than one teacher marks NEA, or you teach more than one group, JCQ's Instructions for conducting non-examination assessments (2026–27) require you to standardise marks internally and keep evidence that you did. Their suggested approach is to start with exemplar material, run trial marking with cross-marking, and recheck once marking is under way.

While you are finding your feet, these are my rough benchmarks for the overall total. They are personal anchors, not OCR guidance, so always confirm against the descriptors:

  • A project abandoned mid-development, with no working evaluation, tends to sit in the high 20s to low 30s. Strong analysis can't rescue a missing solution.
  • A modest project that delivers what it scoped, with a brief but honest evaluation, also tends to sit in the low 30s.
  • A delivered solution with one stakeholder, little technical ambition and a sound evaluation against the success criteria tends to sit in the high 50s.
  • A project with several stakeholders interviewed, genuine technical depth, full delivery and a thorough evaluation can sit in the mid-60s.

OCR's support materials include exemplar programming projects with moderator commentary at high, mid and low levels. Read them before your first marking session, and compare candidates' work against them directly.

Checking the work is the student's own in the AI era

What JCQ requires of teachers

JCQ's AI Use in Assessments: Your role in protecting the integrity of qualifications was last revised on 20 November 2026. Its key requirements are:

  • Students cannot be credited for work that isn't their own. Using AI without acknowledging it is misuse, which is malpractice.
  • Staff should not accept work they suspect was produced with AI tools without investigating further, and should only authenticate work they are confident is the student's own.
  • Where a student uses AI, the acknowledgement must name the tool and the date. The student must keep their prompts and the AI output in a non-editable form, such as screenshots, with a brief explanation of how they used it. "ChatGPT" on its own is not an acceptable reference.
  • Even when AI use is acknowledged, you must not reward parts of the work where AI did what the criteria ask the student to do. Keep a clear record wherever AI use has affected a mark.
  • The November 2026 revision adds a requirement to give NEA students JCQ's Information for Candidates: AI document before they begin the work.
  • If you use AI to help you mark, it cannot be the sole marker. A human must review all the work and remains responsible for the mark.

The JCQ NEA instructions add the general rules. You must know a student's usual standard well enough to judge whether the work is within their capability. If you can't confirm the work is the student's own, don't accept it for assessment.

What OCR says about AI in the A Level CS NEA

OCR's blog Guidance and support for the use of AI in A Level Computer Science NEA (updated 4 February 2026) is the subject-specific guidance. The H446 specification itself (version 3.0, April 2026) also now tells teachers delivering the NEA to follow JCQ's AI guidance, and says students can only be awarded marks for what they can do themselves. In summary, the blog says:

  • Analysis: AI can help generate ideas and speed up research, but final ideas should not come from it. Analysis needs real interaction with stakeholders.
  • Design: AI use should generally be discouraged. Students should attempt each part independently first. Warning signs include suspiciously perfect designs, pseudocode that reads like code, and sudden redesigns from one week to the next.
  • Implementation: AI may suggest methods or help with troubleshooting, but the student must show how they adapted what it suggested. Asking AI to write a whole method or algorithm is unacceptable. Warning signs include erratic work rates, shifts in coding style, and code growing much faster than in lessons.
  • Testing and evaluation: AI use in testing should be very limited, and using AI to write the evaluation is unacceptable. A student who can't connect their earlier test data to their own code may have used AI without acknowledging it earlier in the project.
  • Your department: OCR recommends a department AI policy that matches your school's policy, teaching students to reference AI before the NEA starts, and keeping evidence of AI use in an appendix.

Talking to students about AI

Ofqual's AI in coursework resources for schools (updated 18 November 2026) include a ready-to-deliver lesson plan, a classroom poster, an assembly presentation and newsletter text for parents. An Ofqual blog from March 2026, when the first resources were published, argues that beyond the rules there is a bigger reason: "when AI does the work, they miss out on the learning." Have that conversation early, well before the deadline.

Practical, defensible checks

No single check proves anything. JCQ warns that AI detectors can give both false positives and false negatives, so treat any detector result as one part of an overall judgement. These checks are supported by the JCQ and OCR guidance above, and they fit naturally into a CS project:

CheckWhat to do
Version historyAsk for Git commits or dated copies of the code and write-up for each iteration. JCQ lists a history that shows "a single large insertion of finished text rather than progressive drafting" as an indicator of possible misuse.
Short, regular conversationsAsk the student to talk you through a function, a bug fix or a design decision. JCQ suggests short verbal discussions at various stages, and cloze-style questions where you read part of a sentence from their work and ask them to finish it.
Supervised work and observing codingJCQ recommends doing enough work in class under direct supervision to support authentication. Watching students code in lessons tells you how they actually work. A Computing at School session on OCR NEA marking (March 2026) also recommended observing students coding and asking them to explain their work.
Compare with classworkCompare the NEA with supervised work you already have: coding style, naming, the techniques they normally use, how fluently they write. Work far above the student's usual standard, with no evidence of how they got there, is a JCQ indicator.
Ask to recreate a fixAsk the student to explain or reproduce a remedial action from their testing tables. OCR suggests talking to students about how they reached their solutions and how well they understand them.

If you have concerns

Timing matters. Before the student signs the declaration of authentication, the centre can deal with concerns under its own policy and does not have to report them to OCR. After the declaration is signed, the head of centre must report the case to OCR on form JCQ/M1. OCR's blog gives malpractice@ocr.org.uk as the address. If you suspect malpractice or can't authenticate a student's work, JCQ's NEA instructions say you must inform a member of your senior leadership team.

Feedback, marks and moderation: the rules that catch people out

  • General advice vs. specific help. General advice does not need to be recorded. Help beyond that, such as detailed feedback on drafts or correcting code, must be recorded and taken into account when you mark. Don't give marks for anything you effectively did for the student.
  • No provisional marking followed by redrafting. JCQ says teachers must not provisionally assess work and then allow the student to revise it, and failing to follow its rules on assistance constitutes malpractice.
  • Tell students their marks. Students must be told their centre-assessed marks in time to ask for a review before marks are submitted to OCR. Leave room for this when you plan backwards from the deadline for centre-assessed marks, which is 15 May 2027 for the June 2027 series according to JCQ's key dates.
  • Paperwork. Complete an NEA cover sheet for each candidate. The specification says candidates and centres declare the work is the candidate's own on the centre authentication form (CCS160), which you keep until after the review-of-results deadline.
  • Moderation. After internal standardisation you submit marks, and OCR then requests a moderation sample that includes work from across the range of attainment (centres with fewer than 10 candidates submit all their work). The sample is uploaded through OCR's Submit for Assessment service or posted to the moderator, depending on the entry option you chose. Keep marked work secure until the review-of-moderation deadline has passed.

Final checklist before you record a mark

  • I have read the whole section before placing it in a band.
  • I have used best fit, not bullet-counting.
  • My comment uses phrases from the band descriptor and gives page numbers.
  • The section marks add up to a total that makes sense as an overall judgement.
  • Any help beyond general advice, and any AI use, is recorded and reflected in the mark.
  • If I changed a mark, I have written down why.
  • I have noted any patterns, such as all-pass tables or terminology drift.
  • I would be comfortable defending this mark to a moderator.

FAQ

How many marks is the OCR A Level Computer Science NEA worth?

70 marks, which is 20% of the H446 A Level: Analysis 10, Design 15, Iterative development 15, Testing to inform development 10, Testing to inform evaluation 5 and Evaluation of solution 15.

Can students use AI in the OCR A Level CS NEA?

Only in limited ways, and every use must be acknowledged with the tool, the date, the prompts and the output. OCR's guidance allows AI for generating ideas, research and troubleshooting, provided students show how they adapted what it gave them. It discourages AI in design and in testing and evaluation. Unacknowledged AI use is malpractice, and acknowledged AI work still can't be credited as the student's own.

Can I use an AI detector to check NEA work?

You can, but JCQ says detectors produce false positives and false negatives. Use any result as one part of an overall judgement, alongside version history, conversations with the student and comparison with their classwork.

What happens if I suspect AI misuse after the student has signed the declaration?

It must be reported to OCR on form JCQ/M1. If the student hasn't yet signed, your centre can deal with it internally under its own policy.

Can I give students feedback on their NEA drafts?

You can give general advice. Anything more specific must be recorded and taken into account in the mark. You must not mark work provisionally and then let the student improve it.

Is there a student version of this guide?

Yes. Our OCR A Level Computer Science NEA guide for students covers project ideas, each section and the AI rules from the student's side. It is useful to share at the start of Year 12.

Authenticity checks are much easier when you already know how each student codes. If you'd like a regular picture of that from ordinary lessons, our free Python platform for schools marks every answer as it is submitted, and its Summarise button shows you the most common mistakes and where each student got stuck. It covers GCSE, A Level and IB. For more A Level material, see the OCR A Level Computer Science hub.