---
name: supervision-feedback-and-development
description: >
  Produces developmental feedback on a supervised student's research work that is
  prioritised, specific, and teaches the underlying principle rather than
  correcting the instance. Use for "give feedback on my student's chapter",
  "how do I comment on this draft", "my student isn't progressing and I need to
  say so", "I have 200 comments and that can't be right", "what do I write
  instead of needs more critical analysis", "supervision feedback record",
  "feedback for a student whose English is a second language".
category: 15 Academic University Research
ref: "15.20"
tier: 3
inherits: [K2, K3, K4, K5]
---

# Supervision Feedback and Development

## 1. One-line description

A method for producing supervisory feedback that develops the researcher rather than only repairing the draft: prioritised to two or three things that matter, diagnosed as either a defect in the work or a gap in understanding, written as a transferable principle with the instance as its illustration, and recorded so that development and progression can both be evidenced.

## 2. What this skill is used for

**Read this before you start.** Using an AI system on a student's work may be restricted or prohibited by your institution's policy on AI in supervision and assessment, and it involves processing a named individual's personal data and their unpublished intellectual work. Both questions have to be settled before the draft is opened, not after the feedback is written. Section 5 makes the policy check a required input and Section 12 sets the operating rules. Where your institution has no position, the conservative default is to work on the argument and the structure without uploading identifying material, and to ask the institution before going further.

**The research problem it solves.** Supervisory feedback fails in two ways, and both are ways of being conscientious. The first is volume: a draft returned with two hundred marginal comments, which teaches nothing, because the student cannot see a hierarchy in it and so fixes the surface. The second is abstraction: a comment saying the work "needs more critical analysis", which is true, unhelpful and unactionable, because it names a deficiency without naming what would remove it. Underneath both sits the real tension in supervision: the feedback that most improves this draft is not the feedback that most improves this researcher, and every supervisor navigates that trade-off, mostly without articulating it. This skill makes it explicit, forces a diagnosis before a comment, and converts corrections into transferable principles.

**Where it sits in the research lifecycle.** Throughout a supervision relationship, at every point where a student submits work: a proposal, a literature chapter, a methodology, an analysis, a full draft. It also carries the record that a progression or continuation decision will later have to rest on.

**Typical use cases.**
- Returning a chapter draft with feedback the student can act on in the time available.
- Deciding which two of eleven visible problems to lead with.
- Converting a repeated correction into a principle the student can apply unsupervised.
- Writing the early, clear message to a student who is not progressing at the required rate.
- Separating feedback on writing from feedback on thinking where a student writes in an additional language.
- Building the supervision record that documents what was advised and when.

**Who uses it.** Supervisors and co-supervisors at any level, from a first honours student to a doctoral candidate; postdoctoral and early-career staff supervising for the first time, for whom the prioritisation judgement is hardest; supervisory panels agreeing a common line before a milestone review.

## 3. When to use it

- A student has submitted a draft and you have to return something useful within a bounded amount of your own time.
- You have read a draft, you can see many problems, and you need to decide which ones are this round's feedback.
- The same error keeps recurring across drafts, which means the previous correction did not transfer.
- The student is fixing what you flag and nothing else, which is a signal that your feedback is being read as a checklist.
- You need to tell a student that progress is not adequate, and you want the message to be clear, early, and defensible.
- A student writing in an additional language has produced work whose surface makes its quality hard to judge.
- A progression, upgrade or continuation review is approaching and the feedback record has to show what was advised.
- Two supervisors disagree about what the student should be told, and the disagreement needs resolving before the student hears both versions.

## 4. When NOT to use it

- **The work is being marked, not developed.** Feedback that carries a grade, or that will be read by an examiner or a board, is assessment and follows a different discipline: criteria first, evidence for every judgement, calibration across candidates. Go to **15.21 Thesis Examination and Marking**. Developmental feedback and summative assessment can both be honest and still say different things; conflating them produces feedback that is neither developmental nor defensible.
- **The academic has not read the work.** Feedback generated from a draft the supervisor has not themselves read is not supervision, it is a text about a text. The named academic is accountable for what the student is told, and cannot be accountable for advice they have not checked against the submission. See §12.1.
- **The issue is conduct rather than capability.** Suspected plagiarism, undeclared AI use, fabricated data or contract cheating are institutional processes with their own thresholds and rights of response. This skill does not assess whether work is a student's own, does not produce an allegation, and must not be used to build one.
- **The student's difficulty is not academic.** Illness, bereavement, caring responsibilities, financial crisis, immigration or visa pressure, and mental health difficulty all present as missed deadlines and thin drafts. Academic feedback delivered onto a non-academic problem makes things worse and delays the referral that would help. `RESEARCHER DECISION REQUIRED` (K5 §2.4): the supervisor decides whether this is a feedback situation or a support-referral situation, and no draft-level analysis substitutes for that judgement.
- **The work is at a stage where feedback is premature.** A student who has produced an exploratory first attempt at a chapter to think with does not need it appraised against final standards. Ask what the draft is for. Detailed critique of a thinking draft teaches the student to stop showing you thinking drafts, which is the worst outcome available.
- **What is actually needed is a plan, not a critique.** Where the problem is that the project is behind, over-scoped or stalled, comments on a chapter are displacement activity. Go to **15.23 Student Progress and Milestone Planning**.
- **The question is whether the thesis will pass.** That is an examination judgement made against the standard of the degree, and giving it as feedback commits you to a prediction you do not control. **15.21** sets out the standard; **15.19 Examiner-Proofing and Defence Strategy** is what the student uses to prepare.
- **No criteria and no stage information are available.** Feedback without knowing what the student was asked to do, and where they are in the programme, is a critique of the text against the reader's private standard. Ask, or stop.

## 5. Required inputs

**Required.**
- **Your institution's position on AI use in supervision and on student data.** Whether an AI system may be used on student work at all, whether the student must be told, and what may be uploaded. If unknown, ask the institution before proceeding. This is not an optional preliminary; it is the condition on everything else.
- **The draft, and what the student was asked to produce.** A chapter judged against the wrong task generates confident, irrelevant feedback.
- **The stage of the programme and the degree being worked toward.** Standards are stage-relative. A literature chapter that is adequate in month four of a doctorate is a problem in month thirty.
- **The supervisor's own reading of the draft.** See §4 and §12.1.

**Optional, and what each one adds.**
- **The marking criteria or the examination criteria the work will eventually meet.** Turns "this is weak" into "this will be judged against a criterion it does not currently meet, here is the gap". It also stops feedback drifting toward the supervisor's personal preferences.
- **Previous feedback given, and the student's response to it.** This is the single most valuable optional input. A problem that has been raised before and persists is a different problem from one appearing for the first time: the first is a gap in understanding, the second may be an oversight. Without the history you will re-issue advice that has already failed.
- **The time the student has before the next milestone.** Determines how much of the feedback is actionable now and what has to be deferred. Feedback that cannot be acted on in the time remaining is a source of paralysis, not improvement.
- **Whether the student writes in an additional language, and whether language support is available.** Prevents the most common misdiagnosis in supervision, which is reading a language difficulty as a thinking difficulty, or the reverse.
- **The supervisory arrangement.** Sole supervisor, joint, or a panel. Where there are several, agree the line before the student is told two things.
- **The student's own account of where they are stuck.** Reframes the feedback around the block rather than around the text, and is frequently more accurate than the text's own signals.

## 6. Questions to ask before starting

1. **Does institutional policy permit an AI system to be used on this student's work, and does the student need to be told?** Determines whether this skill runs at all, and in what form. Default if unanswered: do not upload identifying or unpublished material; work on structure and argument in de-identified form, and check with the institution before the next round.
2. **What is this draft for?** A thinking draft, a formal milestone submission, a near-final chapter and a full thesis draft all require different feedback densities. Default: ask. If unavailable, assume a formal draft and say which assumption you made.
3. **What discipline, degree and stage, and against what criteria?** Never assume. Conventions on voice, citation density, the acceptability of a first-person account, what counts as a contribution, whether a literature review is a chapter or a section, and whether a methodology chapter is expected at all, vary by discipline, by country and by degree structure. Default: ask, and if you must proceed, state the convention you are applying and mark it as an assumption the supervisor should correct.
4. **What was said last time, and what happened?** Determines whether this is a first-time correction or a failed transfer. Default: assume first-time, and flag anything that looks like a recurring pattern for the supervisor to confirm.
5. **How much time does the student have before this matters?** Sets the number of priorities that can honestly be issued. Default: assume one working cycle, and issue no more than three priorities.
6. **Is progress adequate?** A distinct question from whether this draft is good, and one supervisors defer. Default: ask it explicitly on every third submission, and always at a milestone.
7. **Is there anything going on for this student that a supervisor would need to know?** Default: assume nothing, but do not diagnose the absence of it from the draft, and route any concern to the institution's support processes rather than into the feedback.

## 7. Step-by-step methodology

**Step 1. Settle the policy and data position, and record it.**
Before the draft is opened: confirm what your institution permits, decide what will and will not be shared with an AI system, and note the decision. Where the policy is silent, the defensible default is that unpublished student work and identifying details are not uploaded, and that the AI-assisted part of the work is confined to structuring and testing the supervisor's own reading. Record one line in the supervision file: what was used, on what, and what the student was told. *Correct result: a written position, taken before reading, that you would be content to show the student and the institution.*

**Step 2. Establish the frame: task, stage, criteria, history.**
Write down, in four lines: what the student was asked to produce, where they are in the programme, the criteria the work will eventually be judged against, and what was said last time. This takes two minutes and prevents the most common failure in supervision feedback, which is judging a draft against the reader's imagined ideal rather than against the task and the stage. Where a discipline convention is doing work in your judgement, name it now, because the student may be operating in a different one legitimately. *Correct result: four lines that a colleague could read and then predict roughly what standard you are about to apply.*

**Step 3. Read the whole draft once, straight through, without annotating.**
Resist the pen. The purpose of the first pass is to answer one question: what is this trying to argue, and does it get there? Annotation on a first pass produces a list of local defects and destroys the ability to see structure, because attention spent on sentence three is not available for the relationship between sections two and four. At the end of the pass, write the argument of the draft in two sentences, in your own words. If you cannot, that inability is the finding, and it is almost always the most important thing you will tell the student. *Correct result: a two-sentence statement of the draft's argument, or an explicit note that no argument could be reconstructed and at what point the thread was lost.*

**Step 4. Inventory the problems without ranking them.**
Second pass, now with a pen. List everything, briefly, with a location. Do not filter yet, and do not write comments to the student. This list is for you. Sort each entry into one of four categories, because the category determines the response, and mixing them is what produces two hundred undifferentiated comments:

| Category | Definition | Response |
|---|---|---|
| **Blocking** | The work cannot proceed until this is resolved. The research question does not match the design, the analysis does not answer the question, the theoretical frame is not being used. | Lead. Everything else waits. |
| **Structural** | The content exists but the architecture defeats it. Order, signposting, the missing section, the argument that is present in the material and absent in the writing. | Address explicitly and in full. |
| **Developmental** | A recurring pattern that reveals a gap in the student's understanding of how research is done or written. | Teach the principle, once, with two instances. |
| **Surface** | Expression, referencing format, typography, individual sentence problems. | One general note, one marked example, and no more. |

*Correct result: a categorised inventory, held privately, that you have not yet sent anyone.*

**Step 5. Diagnose each significant entry: defect in the work, or gap in understanding?**
This is the step that distinguishes supervision from proofreading. For each blocking, structural and developmental entry, ask what would have had to be true for the student to have avoided it. If they know the principle and did not apply it here, it is a defect in the work and the response is to point at it. If they do not know the principle, or hold a wrong version of it, it is a gap in understanding, and pointing at the instance produces a corrected instance and an uncorrected student. The diagnostic evidence is pattern and history: one occurrence with the principle applied correctly elsewhere is a defect; three occurrences, or one after the same correction last time, is a gap. Where you cannot tell, ask a question rather than issuing a correction. "You have cited these six studies in a row; what is the claim they are collectively supporting?" tells you more than any amount of marking up. *Correct result: every significant entry carries a diagnosis citing its evidence (occurrence count, prior feedback) rather than an impression of the student.*

**Step 6. Select the priorities. Two, or three at the outside.**
Take the blocking items first; if there are more than two, the rest are next round. If there are none, take the one or two structural items with the largest effect on the work's ability to make its case, and the one developmental item most likely to recur. The selection rule is leverage, not severity: pick the problems whose resolution fixes or reveals others. A misaligned research question is high leverage, because half the local problems in the chapter are consequences of it. Inconsistent tense is low leverage, however irritating. Everything not selected is either held for the next round or issued as a single collected note with no expectation of action this cycle. Say explicitly which is which, or the student will treat all of it as equally required and spend the cycle on the wrong half. *Correct result: a numbered list of two or three priorities, plus an explicitly labelled "not this round" set.*

**Step 7. Write each priority as principle, instance, consequence, action.**
Four parts, in this order, and the order matters because the principle is what transfers.

- **Principle.** The general rule, stated so it applies beyond this draft. "A literature review earns its place by organising the field into positions that disagree, not by listing what each author found."
- **Instance.** One located example, with a page or section reference, so the abstraction is anchored. Two if the pattern needs demonstrating; never seven.
- **Consequence.** What this costs the work, in the terms of the criteria or the examination. "As written, section 2.3 gives an examiner no way to see where your study sits, so the contribution claim in chapter 5 has nothing to stand on."
- **Action.** A specific, bounded thing to do next, that fits the time available. "Rewrite 2.3 around the three positions on this question, and put your own position last."

Then apply the specificity test to every sentence: could a competent student act on it without asking what you meant? "Needs more critical analysis" fails. "You describe each study's findings but never say whether you find its method convincing; for the four studies you rely on most, add one sentence each on what its design can and cannot establish" passes. The rewrite of a vague comment is almost always longer, and that is the correct trade. *Correct result: each priority occupies a short paragraph containing all four parts and passes the specificity test.*

**Step 8. Separate feedback on writing from feedback on thinking, and say which is which.**
These are different faults with different remedies, and students routinely misread one as the other. Poor expression can conceal sound reasoning; fluent expression can conceal none. The test is to restate the passage's claim in plain language. If you can, and the claim is sound, the problem is expression, and it goes to the surface tier or to a language service. If you cannot, or the restated claim collapses, the problem is thinking, and editing will not fix it. State the diagnosis in as many words, because "this section is unclear" leaves the student unable to tell whether to rewrite the sentences or rethink the argument. This matters most with students writing in an additional language: a supervisor who marks up grammar for six months and then finds the argument was never there has failed the student twice, and one who assumes weak English means weak reasoning has failed them differently. Assess the reasoning first, on its own terms, and handle the language separately. *Correct result: each priority labelled as a thinking issue or a writing issue, and the student told which.*

**Step 9. Set the tone deliberately, and check it against the rigour test.**
Rigorous and discouraging are not the same thing, and the difference is not softening. It is three things: criticism attaches to the work and not the person; every criticism names what would fix it; and what is working is stated plainly, because a student who does not know what to keep will change it. Avoid both failure registers. The unmarked pile of criticism, with no hierarchy and no praise, reads as a verdict on the student. The compliment sandwich buries the substance between two pleasantries, and the student takes away the pleasantries. Say the difficult thing directly and early, then explain it fully; where the feedback is severe, say that it is severe and why, rather than letting the student discover the severity by counting comments. *Correct result: read your feedback as if you were the student. If the dominant effect is confusion about what to do next, or a sense of global inadequacy, it is not finished.*

**Step 10. Where progress is not adequate, say so now, in writing, with the specifics.**
The kindest version of this message is the early one. A student told in month nine that the work is behind can recover; the same student told in month twenty-six cannot, and will reasonably ask why nobody said so. The message has four parts: the standard expected at this stage, what has been produced against it, what specifically must change, and by when it must be visible. Avoid both evasions: delivered as a comment on the draft it reads as one more item on a list, and delivered only verbally it leaves no record and both parties later remember it differently. This is a dated written supervision message, separate from the draft feedback. `RESEARCHER SIGN-OFF REQUIRED` (K5 §2.5, §2.8): the wording, the timing and the judgement that progress is inadequate belong to the named academic. *Correct result: a dated written message in the supervision record stating the gap and the required change, without stating a consequence the academic is not authorised to impose.*

**Step 11. Record what was given, and close the loop next time.**
Log the date, the priorities issued, the diagnosis behind each, and what the student was asked to do. At the next submission, check each prior priority: acted on, partly acted on, not acted on, or acted on in a way that shows the principle did not transfer. That fourth outcome matters most developmentally and is invisible without the record. The same record is what a progression panel or an appeal will require, and reconstructing it from memory afterwards is unconvincing. Per K2 §5, each piece of advice is tied to the location in the draft that prompted it. *Correct result: a supervision log where each entry names the advice, its location, and the outcome observed at the next submission.*

## 8. Analytical framework

Feedback is built on one chain, applied per issue:

    Observation → Diagnosis → Principle → Instance → Consequence → Action → Check

The chain has two load-bearing joints. The first is *Observation → Diagnosis*, where a defect in the work is separated from a gap in the student's understanding. Skipping it produces feedback that repairs drafts and does not develop researchers, which is the characteristic failure of conscientious supervision. The second is *Principle → Instance*, and its direction is deliberate. Feedback that runs instance-first ("this sentence does not follow from the previous one") teaches the student about that sentence. Feedback that runs principle-first, with the instance as illustration, teaches them something they can carry to the next chapter you will not read.

Sitting across the chain is the priority tier, which determines how much of the chain each issue gets:

    Blocking → full chain, leads the document
    Structural → full chain
    Developmental → principle and two instances, no exhaustive list
    Surface → one note, one example, no chain

The governing discipline is subtractive. Every issue you find has to earn its place in the feedback by displacing something else. If nothing is displaced, nothing was prioritised.

## 9. Output format

Feedback is issued as a short document, not as marginal comments alone. Marginalia may accompany it; they never replace it, because marginalia have no hierarchy.

**1. Header.** Student, draft, date, what the draft was asked to be, the stage, and one line on the AI and data position taken under Step 1.

**2. What this draft does well.** Two or three sentences, specific, naming what to keep. Not a preamble; a functional instruction.

**3. The argument as I read it.** Your two-sentence reconstruction from Step 3, offered as a check. Where you could not reconstruct one, say so here and say where the thread was lost. This section does more work than any other, because a student seeing their argument described back to them can see instantly whether it survived the writing.

**4. Priorities for this round.** Numbered, two or three, each with principle, instance and location, consequence, action. Each labelled thinking or writing.

**5. Noted but not for this round.** A short collected list, explicitly not required work this cycle. Includes the surface tier, in one paragraph, with one marked example.

**6. Questions I want you to answer.** Where the diagnosis was uncertain (Step 5), the question that would resolve it. This converts a guess about the student into a conversation with them.

**7. Next step and date.** What is submitted next, and when.

**When the draft is thin, the format must not manufacture substance.** A draft that does not yet contain an argument gets a short feedback document saying so, one priority, and a meeting. Do not populate section 4 with three priorities because it has three slots (K4 §1), and do not compensate for a thin draft with dense surface annotation, which is the most common way a supervisor appears to have engaged without engaging.

## 10. Quality checks

Run before the feedback is sent.

1. Did you read the whole draft yourself before any of this was written?
2. Is the number of priorities three or fewer, and is everything else explicitly marked as not-this-round?
3. Does every priority state a principle that applies beyond this draft?
4. Does every criticism carry a location the student can find?
5. Could a competent student act on every sentence without asking what you meant?
6. Is each priority labelled as a thinking issue or a writing issue?
7. Where a problem has appeared before, does the feedback say so and change its approach rather than repeat the correction?
8. Is any judgement about the student, as opposed to the work, present in the text? Remove it.
9. Is the feedback proportionate to the stage, or is it applying final standards to an early draft?
10. Does the document say what is working, specifically enough to be actionable?
11. If progress is inadequate, is that said plainly, in its own message, dated, rather than implied through the volume of comments?
12. Is anything in the feedback a matter of your preference rather than of the criteria or the discipline's conventions, and if so, is it labelled as preference?
13. Is the record updated, with each priority tied to its location, so the next round can check transfer?
14. Would you be comfortable if this document were read by the student, a co-supervisor and a progression panel? It may be.

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The two-hundred-comment return** | Every page is dense with marginalia and no priority is visible | Categorise privately, then select two or three (Steps 4 and 6) |
| **Surface capture** | The student's revision fixes every flagged sentence and none of the structure | Lead with structure, and mark surface items explicitly as not-this-round |
| **"Needs more critical analysis"** | A comment that names a deficiency without naming its remedy | Apply the specificity test in Step 7 to every sentence |
| **Correcting the instance** | The same error returns, corrected in the flagged place only | Diagnose gap versus defect, and lead with the principle (Steps 5, 7) |
| **The private standard** | Feedback that reflects how the supervisor would have written it | Frame against task, stage and criteria first (Step 2) |
| **Language read as thinking** | Grammar marked heavily; the argument never assessed | Restate the claim in plain language and diagnose separately (Step 8) |
| **Thinking read as language** | Fluent prose accepted; the hollow argument passes | Reconstruct the argument before annotating (Step 3) |
| **The deferred hard message** | Progress concerns first stated near submission | Ask the progress question at every milestone (Step 10) |
| **The compliment sandwich** | The substantive criticism sits in the middle and is the shortest part | State the difficult thing first, then explain it |
| **The unrecorded conversation** | A verbal warning both parties remember differently | Written, dated, in the supervision record (Steps 10, 11) |
| **AI-fluent feedback the supervisor has not verified** | Plausible, well-structured comments referring to things not in the draft | The supervisor reads the draft; every located reference is checked (K4 §6.4) |
| **Invented specifics** | Feedback citing a page, a section or a claim that does not exist in the submission | Every location reference verified against the actual document before sending |
| **Feedback as a checklist** | The student asks "is that everything?" | Say what the principle is, and that applying it is the work |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated.

1. **These skills assist an academic's judgement, they do not replace it.** Assessment, examination and progression decisions are the responsibility of the named academic and cannot be delegated to an AI system. The skill never assigns a mark, never produces feedback to be passed to a student unread, and never makes a progression or examination recommendation the academic has not independently reached. Institutional regulations and marking criteria take precedence over anything in this skill, and where a student's work may have integrity concerns, that is an institutional process and not a judgement this skill makes.

2. **Settle the policy and data question before the work, not after.** Institutional rules on AI in supervision, and on processing a student's personal data and unpublished work, govern whether this skill runs at all. Where the position is unknown, do not upload identifying or unpublished material, and say so in the output. Do not infer permission from the absence of a prohibition.

3. **Never generate feedback on a draft the supervisor has not read.** Where only an extract, a summary or a description of the work is available, the output is confined to what that material supports, and says explicitly what it is based on. Never write feedback in the register of having read the whole document.

4. **Never cite a location, a page, a section, a source or a quotation from the student's work that has not been verified against the document.** A feedback comment referring to something not in the draft costs the supervisor their credibility with that student permanently, and it is the single most likely AI failure in this task.

5. **Never characterise the student.** Feedback describes work, not people. Do not infer effort, ability, motivation, engagement or trajectory from a text. A thin chapter is evidence about a chapter; it is not evidence about a person, and the causes that produce it are frequently invisible in the document.

6. **Never diagnose an integrity concern, and never hint at one.** Do not comment on whether work appears AI-generated, plagiarised or not the student's own. Detection claims of this kind are unreliable, they carry serious consequences for the student, and they belong to an institutional process with rights of response. Where something concerns the supervisor, the route is the institution, not the feedback document.

7. **Never state or imply an outcome the academic has not reached.** No prediction that a thesis will pass or fail, no implied mark, no statement about whether a student should continue. Progression language is reserved to the academic and the institution (K5 §2.8).

8. **Where a judgement rests on discipline convention, name the convention and mark it as assumed.** Expectations about voice, structure, citation density, theory use and what constitutes a contribution differ by field, country and degree. Applying one field's conventions to another's student invisibly is a common and damaging failure.

9. **Do not resolve an uncertain diagnosis by choosing the more coherent story** (K5 §2.6). Where the evidence does not distinguish a gap in understanding from a lapse in execution, output the question to put to the student rather than the diagnosis.

10. **Cap confidence on any inference about transfer or pattern at the evidence available** (K3 §3.1). One occurrence is not a pattern. Where prior feedback is not supplied, say that recurrence could not be assessed rather than assuming a first occurrence.

## 13. Best-practice principles

- **Feedback is a teaching act with a scarce budget.** The constraint is not your time, it is the student's capacity to change several things at once. Two changes made are worth twenty listed.
- **The first pass has no pen.** Structure is only visible before annotation starts, and it is where nearly all the leverage lives.
- **Say the argument back to them.** A two-sentence reconstruction of what a chapter argues is the highest-value pair of sentences in most feedback documents, and it costs nothing once the first pass is done.
- **Lead with the hardest thing.** Difficult feedback placed late is read as an afterthought; placed first and explained fully, it is read as the point.
- **Distinguish the criteria from your taste, out loud.** Students cannot tell the difference and will spend equal effort on both. Saying "this is a requirement" or "this is my preference" buys a great deal of trust.
- **A recurring error is a curriculum, not an irritation.** The third appearance means the earlier correction did not carry a principle, and the response is to teach it, not to correct it again more firmly.
- **Ask before you diagnose.** One question resolves more supervision uncertainty than any amount of close reading, and it models the behaviour you want.
- **Praise is functional.** A student who does not know what worked will remove it in revision. Specific praise is instruction, not encouragement.
- **The early hard conversation is the kind one.** Supervisors defer it to protect the student from discouragement, and thereby remove the time in which recovery was possible.
- **Never let feedback carry a decision it has no authority to make.** Feedback that implies an outcome creates an expectation someone else will have to break.
- **Write for the second reader.** Assume a co-supervisor, a panel or an appeal will read it. Feedback written to that standard is also clearer to the student.
- **Record it as you go.** The supervision log is the only thing that survives a change of supervisor.

## 14. Worked example

Generic fictional scenario, academic, social sciences.

**INPUT**

A supervisor receives the second draft of a literature chapter from a taught masters student, eight weeks before the dissertation deadline. The student writes in an additional language. The first draft, returned three months ago, carried nineteen marginal comments, fourteen on expression and referencing and five on content. This draft is longer, the referencing is nearly clean, the sentences are noticeably better, and the supervisor's instinctive reaction on skimming is that it has improved.

**PROCESS**

*Step 1.* The institution permits AI-assisted preparation of feedback but requires that no unpublished student work be uploaded externally and that the supervisor's own reading is the basis of anything sent. The supervisor works from their own notes and records that in the file.

*Step 2.* Frame: a critical literature review chapter of roughly 4,000 words; eight weeks from submission; criteria requiring critical engagement rather than description; previous feedback dominated by surface items.

*Step 3.* Whole read, no pen. The supervisor then tries to write the chapter's argument in two sentences and cannot. Three sections each summarise a group of studies competently, with no statement anywhere of what the field disagrees about or where this study will sit. That inability is the finding.

*Step 4.* Inventory: eleven entries. One blocking (no organising argument, therefore no position for the study to occupy). Two structural (the theoretical frame appears in section 3 and is never used; the chapter ends without leading anywhere). Two developmental (studies reported by findings only, never by method quality; every paragraph opens with an author name rather than a claim). Six surface.

*The judgement call.* The supervisor's first instinct is that the student has improved, because the visible change is large. Steps 3 and 4 contradict it: the surface improved because the last feedback was three-quarters surface comments, and the student did exactly what was asked. What did not improve is what was barely mentioned. The diagnosis is therefore not that the student is not progressing; it is that the previous feedback taught the surface, and the supervisor is the proximate cause of this draft's shape. That changes both the content of this round's feedback and its tone.

*Step 5.* Diagnosis on the blocking item: gap in understanding, not defect. The student has produced a competent summary of a body of work, which is what a literature review is in some contexts and disciplines, and nothing suggests they know a masters-level review is expected to organise a field into contested positions. Evidence: the pattern is complete and consistent across all three sections, done carefully, which is not what carelessness looks like.

*Step 6.* Two priorities, given eight weeks. (1) The organising argument. (2) Appraise method, not only findings, for the four studies the dissertation most relies on. Everything else is held: the theory-frame problem will resolve as a consequence of (1), and the surface items go into one paragraph.

*Step 7.* Priority 1 is written principle first: a critical review earns its place by mapping what the field disagrees about and showing where this study stands, not by reporting what each author found. Instance: sections 2.1 and 2.2, where six studies appear in publication order with no statement of what divides them. Consequence: there is no position for the dissertation's contribution claim to stand on, and the criteria require critical engagement explicitly. Action: identify the two or three points on which this literature disagrees, restructure around those, and put the student's own position last. A one-page restructure is requested before any rewriting, so the shape can be checked before three weeks are spent on it.

*Step 8.* Both priorities are labelled thinking, and the feedback says in as many words that the writing improved substantially, that it is no longer the limiting factor, and that the remaining work is entirely about the argument. This matters for a student who may otherwise read continued criticism as continued evidence that their English is the problem.

*Steps 9 to 11.* The document opens by naming the improvement specifically, then states the blocking issue directly and explains that the previous round's emphasis contributed to it. Progress is judged adequate, with the stated caveat that priority 1 must be visible in the restructure within ten days for that to remain true. The supervision log records both priorities, their locations, and the checkpoint.

**OUTPUT**

A two-page feedback document with two priorities, an explicit "not this round" list, a request for a one-page restructure before rewriting, and a dated checkpoint. Marginal annotation is limited to one marked example per surface category. `RESEARCHER SIGN-OFF REQUIRED` on the progress judgement and the checkpoint condition (K5 §2.5).

## 15. Advanced usage

**Feedback across a supervisory panel.** Where two or three supervisors comment, the failure mode is contradictory advice delivered in parallel, which students resolve by following the more senior or the more recent voice. Run the inventory and selection steps jointly, issue one document, and where the panel genuinely disagrees, say so and say how the student should proceed. A named disagreement is legitimate research training; an invisible one is a trap.

**Designing feedback sequences rather than rounds.** Plan which capabilities are taught when: argument structure early, methodological justification in the middle, contribution framing late. Without a sequence, each round's priorities are set by whatever the current draft happens to expose, and some capabilities never get taught because they never happened to fail visibly.

**The handover file.** Where supervision transfers, the record from Step 11 is the deliverable: priorities issued, transfer observed, current diagnosis. Written as you go it takes minutes; reconstructed at handover it takes a day and is less accurate.

**Where the standard approach does not fit.** For practice-based, creative or professional doctorates, the artefact and the written component are assessed against different criteria and often by different people, so run the frame in Step 2 twice. For students working in a supervisor's second language, or in a discipline the supervisor knows adjacently rather than directly, mark clearly which comments are within your expertise and which are a general reader's response; students otherwise weight both equally. Where a student is being supervised at a distance with infrequent contact, increase the proportion of feedback given as questions, because the conversational repair that normally corrects a misread comment is not available.

## 16. Skill chain

**Recommended previous skills:**
- **15.23 Student Progress and Milestone Planning.** Hands over the stage, the milestone position and the time available, which set how many priorities can honestly be issued.
- **15.08 Research Design and Methodology Chapter** and **15.10 Data Analysis Chapter Development.** Supply the standards a methodology or results chapter is being read against, so feedback is framed against convention rather than preference.

**Recommended next skills:**
- **15.21 Thesis Examination and Marking.** Takes over where feedback becomes assessment.
- **15.23 Student Progress and Milestone Planning.** Takes a feedback round that revealed a scope or pace problem and converts it into a revised plan.
- **15.19 Examiner-Proofing and Defence Strategy.** What the student uses on a full draft once developmental feedback has done its work.

**Runs well alongside:**
- **15.22 Research Rigour Audit**, where a student's study design needs auditing independently of how the chapter reads.
- **15.05 Academic Writing Structure and Argumentation**, which is what a student is pointed to when the diagnosis in Step 8 is a writing issue.
- **13.04 Bias Detection**, for checking that feedback across a cohort is not being applied differentially.

---
A Yazi Supplied Skill and resource.