HSC 2026 final syllabus year published Aug 2026

English Extension 1 question styles to look out for

The question styles to have ready — each linked to its evidence and to a question in the practice paper.

Built by a six-model AI panel and backtested against the hidden 2025 papers — how we did it.

About the 2026 exam & how this page was built

Built from a six-model AI analysis of every HSC English Extension 1 paper, marking guideline and marking-centre feedback report since 2019 — backtested against the 2025 papers before we published it. One thing to hold onto from the start: 2026 is the FINAL year of this syllabus and its prescriptions. Final-year papers tend to be conservative in structure and comprehensive in coverage — the committee showcases the module, it doesn't reinvent it. These are styles to prepare for, not guarantees: our backtest showed the examiners keep the skill and twist the format, so practise the skill chain, not a memorised question.*

The paper itself is settled: 50 marks, two hours plus reading time. Section I — Common Module: Literary Worlds, 25 marks. Section II — your elective essay, 25 marks, close reference to TWO prescribed texts and ONE related text. Every model on the panel puts that architecture at 0.93 or higher.

Section II · 25 marks 6 of 6 models expect this

Section II: one common question for all five electives, TWO prescribed texts + ONE related

A single Question 2 stem applied across all five electives (as in 2020, 2022, 2024, 2025) rather than differentiated questions 2(a)–2(e), with the text requirement fixed at close...

Why we expect this

Section II · 25 marks 6 of 6 models expect this

Literary Mindscapes focus: how FORM constructs interiority, not what a character feels

The Mindscapes question lands on the technical construction of consciousness — focalisation, stream of consciousness, multiple narrators, lyric compression, fragmented chronology,...

Try it → Q2(d) Why we expect this

Section II · 25 marks 6 of 6 models expect this

Intersecting Worlds focus: competing valuations of nature and the human/non-human divide

The Intersecting Worlds question turns on competing conceptualisations of the natural world — beauty, spiritual inspiration, resource, belonging — and whose voices representations of...

Try it → Q2(e) Why we expect this

On this page

1. A writer writing about writing — with a phrase you must thread through
2. Imaginative composing returns — and the stem will name the form
3. A named conceptual constraint inside the creative task
4. The 15 + 10 split: compose, then analyse your own composition
5. The hedge: Section I stays discursive/critical
Where to spend study time
Practice paper
Check our working

Section I — the stimulus and form demands to expect

1. A writer writing about writing — with a phrase you must thread through

What it looks like: an unseen nonfiction extract (roughly 150–350 words) by a novelist or essayist theorising about fiction — storytelling, truth, translation, memory — with an aphoristic phrase quoted from it embedded in the task. Griffiths 2019, Yanagihara 2021, Tan and Atwood 2022, Baldwin 2024, Liu and Solnit 2025: this is the exam's signature stimulus, and five of six models expect it back (panel consensus 0.75).
The traps markers flag (2023, 2024, 2025): engaging with the stimulus superficially — a pre-prepared piece with the extract bolted on scores in the middle bands at best. The quoted phrase must be a conceptual spine, cited and re-earned across the whole response, not name-checked once in the opening.
Practise: take any craft essay or writers-festival address, pull one aphoristic sentence from it, and compose for 40 minutes with that phrase doing structural work. Then check: could a marker reconstruct the phrase from your piece alone?

5 of 6 models expect this — see the evidence

2. Imaginative composing returns — and the stem will name the form

What it looks like: after 2025's purely discursive Section I, all six models have creative composition back in some configuration (61% chance). Since 2022 the stem has always named the required form — imaginative, narrative, discursive, or a stated choice — and the guidelines band control of that form as its own criterion.
The trap (2022, 2023, 2024): treating the named form loosely. A discursive response that drifts into essay, or a narrative that is really description, loses the form criterion outright.
Practise: the same stimulus three ways — narrative, discursive, hybrid — so the form demand on the day is a selection, not a scramble.

6 of 6 models expect this — see the evidence

3. A named conceptual constraint inside the creative task

What it looks like: every imaginative task since 2022 has carried one concrete compositional constraint (divergent voices; a world within a world; a crossing that changes a character). The panel's favourite 2026 variant, called independently by three models: two characters who attach incompatible meanings to the same word, object or remembered event — with the divergence structuring the piece.
The trap (2022, 2023, 2024): bolting the constraint on as a plot event or closing twist. Markers use the constraint as the primary line between conceptual and superficial engagement — it has to organise the piece from the first paragraph.
Practise: write openings (just openings) that establish a constraint structurally in 150 words: two voices, a frame, an object that means two things. Speed at conceptual set-up is the scarcest skill in the hall.

5 of 6 models expect this — see the evidence

4. The 15 + 10 split: compose, then analyse your own composition

What it looks like: part (a), a 15-mark narrative from the stimulus; part (b), a 10-mark analysis of how YOUR OWN part (a) constructed a literary world. Used in 2019 and 2024; five models rate its return (consensus 0.44, the strongest single-format call) because it is the only exam vehicle for outcome EE12-5 and a final-year committee wants comprehensive coverage.
The traps (2024): retelling your story instead of analysing it; failing to quote your own writing as evidence; describing what happens rather than naming the choices — voice, structure, motif, perspective — and their intended effect.
Practise: after every practice piece, write a 10-minute reflection quoting yourself twice and using metalanguage. It feels strange precisely because it is undertrained.

Try it → Q1 5 of 6 models expect this — see the evidence

5. The hedge: Section I stays discursive/critical

What it looks like: a single 25-mark response extending an extract's ideas (2021, 2025), or a comparison of two short extracts. All models keep it live at lower probability (33% chance).
The trap (2021, 2025): summarising the extract instead of extending it, and thin literary range — 2025 feedback explicitly asked for discussion supported by a range of literary texts.
Practise: keep one discursive piece warm; don't bet the section on it, but don't walk in without one.

5 of 6 models expect this — see the evidence

Section II — the elective essay

The panel expects one common question across all five electives (70% chance) with a high-order directive — evaluate, justify, to what extent — back after two years of personal-voice stems (64% chance). The traps here are the same four the marking centre has flagged for seven straight years: technique-listing instead of an evaluative thesis (2022–2025), three mini-essays instead of one synthesised argument (2019–2022), evidence drawn only from openings and iconic moments (2019–2023), and a related text that doesn't sit squarely in the elective (flagged every year 2019–2024 — the single strongest separator markers report).

Question focus by elective — what the verbs will point at:

Literary Homelands (63% chance): whose perspectives of home are privileged, marginalised or silenced — and whether form (stage adaptation, poetic sequence, film narration) enacts or resists that silencing. Evaluate perspective and power, don't recount a migrant journey.
Worlds of Upheaval (64% chance): formal rupture as the argument — how broken or innovative form constructs competing social values. Watch the elective's least-tested clause: texts' power to activate change. Name context precisely; "industrialisation" as a gesture scores nothing.
Reimagined Worlds (64% chance): how the imagined world makes the familiar strange to reappraise the composer's own society. Range beyond the famous set-pieces — 'The king was pregnant', the Pale Man, the opening of Kubla Khan are exactly the evidence markers are tired of.
Literary Mindscapes (65% chance): how form constructs interiority — focalisation, stream of consciousness, lyric compression — not what a character feels. Separate the composer from the constructed voice.
Intersecting Worlds (65% chance): competing valuations of nature — beauty, spirit, resource — and whose voices speak for the natural world. Handle the fragmented Wordsworth and Winton selections as whole works, and refuse the conservationist-versus-exploiter binary.

Practise: one essay per fortnight against a rotating verb (evaluate / justify / to what extent), always three texts in one argument, always one piece of evidence per text from its second half.

Also on the radar

A visual stimulus — one model's strong dissent: a minimalist image forcing on-the-spot conceptual translation. Low consensus, high disruption if it lands; do one image-stimulus practice piece.
Five separate elective questions 2(a)–2(e) sharing one quoted stem — three models flag it as the final-year full-coverage gesture (p ≈ 0.35).
A triadic directive — texts "echo, unsettle or oppose" a stimulus claim (2022, 2023 pattern): calibrate a different verdict per text; a blanket agree/disagree caps you.
The module descriptor's untouched clause — 'private, public and imaginary worlds' has never been quoted in a question. Committees like to close a syllabus by returning to unexamined descriptor language.
A short shared stimulus in Section II (as in 2025) rather than a bare question — treat any Section II extract as a phrase-threading exercise, the same skill as Section I.

Where to spend less time

There is no restable content in this course — five electives, one module, everything sits in every paper. What the panel does deprioritise: memorising a polished creative piece to transplant (three consecutive years of constraints exist precisely to break that strategy), and preparing more than one related text to a high polish. One conceptually substantial related text, known whole-text deep, beats two known thin.

The honest fine print

We backtested this method on the 2025 papers (Chemistry and Maths Extension 1) before publishing anything. The results were good but humbling: our high-confidence topic calls went 37/37, and roughly half of all question-level predictions recognisably appeared — but the examiners inverted a flagship format all six models had called one way, migrated predicted questions into different sections and sizes, and broke two five-of-six-model streaks outright. In Extension 1 the whole panel over-predicted breadth. English Extension 1 adds its own caveat: with no fixed content taxonomy, our calls are about stimulus type, form demands and question focus — the most invertible kind of prediction there is. Prepare the skill chains above — threading a phrase, controlling a named form, analysing your own writing, arguing one evaluative thesis across three texts — and whichever way the committee turns the final paper of this syllabus, you're covered.

Want to check our working? every call above, with each model's own prediction

Section I stimulus: an unseen writer-on-craft nonfiction extract, with a phrase quoted into the task 5 of 6 models expect this

ee1-s1-writer-on-craft-extract Section I 15–25 marks gatekeeper consensus 0.75

The Section I stimulus is again a short prose extract (roughly 150–350 words) by a writer theorising about fiction — storytelling, truth, translation, memory, the reader's part in making a world — rather than a literary extract for analysis. The Griffiths 2019 / Yanagihara 2021 / Tan–Atwood 2022 / Baldwin 2024 / Liu–Solnit 2025 lineage. Four models further expect an aphoristic phrase quoted from the extract embedded in the stem, which the response must thread throughout (2021, 2024, 2025 pattern) — the stimulus cannot be read once and discarded.

Each model's own prediction

Claude Fable 5: nonfiction writer-on-storytelling extract with quoted phrase, p 0.85
Grok 4.6: writer on fiction, truth or language, p 0.86
Claude Opus 5: metafictional prose extract by novelist/essayist, quoted phrase embedded, p ~0.80
DeepSeek V4: prose extract stimulus assumed in both Section I formats, p 0.70
GPT-5.6 Sol: short nonfiction extract about storytelling anchors the narrative, p 0.56
dissent: gemini-3.1-pro predicts a minimalist VISUAL or bare aphoristic stimulus instead (p 0.75) — see watch_list

Marker-feedback lineage: Marking feedback 2021, 2023, 2024, 2025 (stimulus engagement and threading the quoted idea)

Section II: one common question for all five electives, TWO prescribed texts + ONE related 6 of 6 models expect this

ee1-s2-common-question-two-plus-one Section II 25 marks structural consensus 0.70

A single Question 2 stem applied across all five electives (as in 2020, 2022, 2024, 2025) rather than differentiated questions 2(a)–2(e), with the text requirement fixed at close reference to TWO prescribed texts and ONE other text of the student's own choosing (the 2023–2025 'exactly one' wording, not 'at least one'). The panel is near-unanimous on the text requirement (0.85–0.99 across all six models); the common-vs-differentiated call is softer (0.55–0.85), with three models flagging a five-question reversion as their bold call.

Each model's own prediction

Gemini 3.1 Pro: single shared stem, p 0.85; TWO+ONE mandate p 0.99
DeepSeek V4: common evaluative essay across electives, p 0.80
GPT-5.6 Sol: one 25-mark frame for all electives, p 0.76
Grok 4.6: unified stem, p 0.64; TWO+ONE p 0.93
Claude Opus 5: single common Question 2, p 0.60
Claude Fable 5: common design, p 0.55

Marker-feedback lineage: Marking feedback 2019–2024 (related-text misalignment, mini-essay structure); 2025 general commentary

Literary Mindscapes focus: how FORM constructs interiority, not what a character feels 6 of 6 models expect this

ee1-s2-mindscapes-form-constructs-interiority Section II 25 marks discriminator consensus 0.65

The Mindscapes question lands on the technical construction of consciousness — focalisation, stream of consciousness, multiple narrators, lyric compression, fragmented chronology, cinematic interiority — as the mechanism by which readers experience another mind. Every model separates describing a character's psychology (mid-band) from analysing the composer's construction of interiority and what it does to the responder's own sense of identity (top band). Secondary territory: memory reshaping rather than preserving the self; the gap between private consciousness and public identity.

Each model's own prediction

Gemini 3.1 Pro: disruption of linear narrative and voice immerses reader in fractured psychological states, p 0.90
DeepSeek V4: language's inadequacy for interior experience, p 0.80
GPT-5.6 Sol: voice/focalisation/chronology expose private-vs-public tension, p 0.64
Grok 4.6: constructed voices give access to interior worlds, composer vs persona, p 0.62
Claude Opus 5: form-mandating stem on constructed interiority, p 0.50
Claude Fable 5: form constructs interiority; describing vs analysing discriminator, p 0.45

Marker-feedback lineage: Marking feedback 2022–2025 (descriptive rather than conceptual analysis; device-listing)

In the practice paper: Q2(d)

Intersecting Worlds focus: competing valuations of nature and the human/non-human divide 6 of 6 models expect this

ee1-s2-intersecting-competing-valuations-of-nature Section II 25 marks discriminator consensus 0.65

The Intersecting Worlds question turns on competing conceptualisations of the natural world — beauty, spiritual inspiration, resource, belonging — and whose voices representations of nature privilege, unsettling the human-centred view. Three models specifically warn the discriminator is holistic handling of the 'fragmented' prescriptions (the Wordsworth selection and Winton's Island Home as whole works, not one coastline image or a single poem), and refusing the conservationist-versus-exploiter binary.

Each model's own prediction

Gemini 3.1 Pro: collision of human and non-human forces challenges cultural paradigms, p 0.90
DeepSeek V4: challenge the human-centred view of nature, p 0.80
GPT-5.6 Sol: unsettle the human–nature divide via competing cultural/economic/spiritual/ecological perspectives, p 0.65
Grok 4.6: reappraise assumptions about use, beauty and belonging, p 0.62
Claude Opus 5: which valuation of nature a text privileges and at what cost, p 0.45
Claude Fable 5: competing conceptualisations; whose voices privileged, p 0.45

Marker-feedback lineage: Marking feedback 2019–2023 (openings-and-iconic-moments evidence; related-text alignment)

In the practice paper: Q2(e)

Reimagined Worlds focus: speculative form makes the familiar strange to reappraise the real 6 of 6 models expect this

ee1-s2-reimagined-familiar-made-strange Section II 25 marks discriminator consensus 0.64

The Reimagined Worlds question asks how imagined settings, altered realities and disrupted narrative conventions re-evaluate perceptions of the composer's (and responder's) actual world — speculation, satire and fantasy as instruments of insight rather than escape. Three models name the same evidence trap: candidates leaning on celebrated set-pieces ('The king was pregnant', the Pale Man scene, the opening of Kubla Khan) instead of ranging across whole texts; the top band treats the invented world as an argument about the composer's society.

Each model's own prediction

Gemini 3.1 Pro: speculative settings interrogate the values of the composer's reality, p 0.90
DeepSeek V4: reconsider the boundaries between reality and possibility, p 0.80
GPT-5.6 Sol: make the familiar strange; reconsider assumptions about humanity, p 0.62
Grok 4.6: speculative/marvellous form unsettles certainty; evidence beyond set-pieces, p 0.62
Claude Opus 5: the return journey — what the imagined world sends the responder back to see, p 0.45
Claude Fable 5: re-evaluating perceptions of the real via speculation/satire/fantasy, p 0.45

Marker-feedback lineage: Marking feedback 2019–2023 (iconic-moment evidence); 2022–2025 (theme-vs-values confusion)

In the practice paper: Q2(c)

Worlds of Upheaval focus: formal rupture mirrors social rupture — and the untested 'activate change' clause 6 of 6 models expect this

ee1-s2-upheaval-form-breaks-with-convention Section II 25 marks discriminator consensus 0.64

The Upheaval question centres on how disrupted or innovative form (industrial montage, Gothic frame, absurdist stasis, lyric elegy) constructs competing social values in periods of rupture — analysis of literary form over historical recount. Two models converge on the elective descriptor's least-examined clause: the potential of texts to ACTIVATE change in attitudes and social circumstances, asking whether literature merely witnesses upheaval or works on readers. Precision of context (not gestures at 'industrialisation' or 'revolution') is the named discriminator.

Each model's own prediction

Gemini 3.1 Pro: structural and linguistic disruption reflects paradigm shifts, p 0.90
DeepSeek V4: upheaval worlds expose limits of social and political structures, p 0.80
GPT-5.6 Sol: disrupted forms represent individuals seeking justice/certainty/restoration; witness vs activate, p 0.61
Grok 4.6: form constructs competing social values, p 0.62
Claude Fable 5: 'activate change' clause plus challenging literary conventions, p 0.45
Claude Opus 5: activate-change clause; formal rupture as the argument, p 0.45

Marker-feedback lineage: Marking feedback 2019–2023 (context gestured at, not argued); elective descriptor clause untested 2019–2025

In the practice paper: Q2(b)

Literary Homelands focus: whose perspectives of home are privileged, marginalised or silenced 6 of 6 models expect this

ee1-s2-homelands-privileged-and-silenced-voices Section II 25 marks discriminator consensus 0.63

The Homelands question turns on constructed voices and points of view: how texts privilege, marginalise or recover cultural perspectives of home, place, belonging and otherness — migrant and colonial voices weighed against the perspectives they displace — and whether form (stage adaptation, poetic sequence, film narration) enacts or resists that silencing. Secondary territory: home as refuge and cage; belonging as inherited versus made; representations of place that simultaneously connect and divide.

Each model's own prediction

Gemini 3.1 Pro: narrative conventions construct place to unsettle reader expectations, p 0.90
DeepSeek V4: challenge dominant assumptions about belonging and identity, p 0.80
Grok 4.6: reappraise assumptions about belonging and otherness; voice/structure/staging as constructing homeland, p 0.62
GPT-5.6 Sol: voices and perspectives unsettle inherited assumptions about homeland/belonging/otherness, p 0.58
Claude Fable 5: privilege, marginalise or silence cultural perspectives of home, p 0.45
Claude Opus 5: whose perspectives a homeland text privileges, marginalises or silences, p 0.45

Marker-feedback lineage: Marking feedback 2019–2024 (related-text misalignment; recount over evaluation)

In the practice paper: Q2(a)

Section II directive verb: evaluate / justify / to what extent returns after the personal-voice stems 6 of 6 models expect this

ee1-s2-evaluative-directive-returns Section II 25 marks discriminator consensus 0.64

After 2024's 'justify' and 2025's 'how does X shape your understanding', the panel expects a high-order evaluative directive — 'evaluate', 'justify', 'to what extent' — demanding a calibrated judgement sustained across all three texts, likely still framed around the candidate's own argument ('your texts', 'your understanding'). Marking feedback has flagged listed techniques instead of an evaluative thesis every year 2022–2025; several models predict the stem will chain multiple components (quoted idea + concept + text requirement) so the top band hinges on keeping every element active.

Each model's own prediction

Gemini 3.1 Pro: 'evaluate' coupled with manipulation-of-form phrasing, p 0.90
DeepSeek V4: evaluative-verb common essay, p 0.80
Grok 4.6: Evaluate / To what extent verb set is due back, p 0.62
GPT-5.6 Sol: evaluate or justify with qualified judgement, p 0.61
Claude Fable 5: evaluate/justify framed around a personal argument, p 0.45
Claude Opus 5: judgement-demanding stems throughout; triadic gradation the variant, p 0.45

Marker-feedback lineage: Marking feedback 2022, 2023, 2024, 2025 (technique-listing without evaluation; components dropped)

In the practice paper: Q2

Section I form demand: imaginative composition returns after the all-discursive 2025 6 of 6 models expect this

ee1-s1-imaginative-composing-returns Section I 15–25 marks gatekeeper consensus 0.61

Every model has creative composition back in Section I in some form — whole-section 25-mark imaginative task, a 15-mark part (a), or an imaginative/discursive choice — after 2025 mandated a purely discursive response. The stem is expected to NAME the required form (named in 2022, 2023, 2024, 2025), with control of that form banded as its own criterion; the discriminator is control of imaginative form under a conceptual constraint, not plot ingenuity. Grok's cycling note: discursive was fully tested in 2025 and imaginative composing is the rested skill; the final-year committee showcases the module's signature creative demand.

Each model's own prediction

Gemini 3.1 Pro: single 25-mark imaginative response, p 0.85
Claude Fable 5: imaginative component returns (whole task or 15-mark part a), p 0.75
DeepSeek V4: 15-mark stimulus narrative, p 0.70
GPT-5.6 Sol: 15-mark narrative component, p 0.52
Claude Opus 5: form named in the stem, p 0.88; narrative task keyed to extract, p 0.48
Grok 4.6: imaginative literary world task, p 0.36

Marker-feedback lineage: Marking feedback 2022, 2023, 2024 (form control banded separately; constraint as discriminator)

The imaginative task carries a named compositional constraint — favourite: incompatible meanings 5 of 6 models expect this

ee1-s1-named-conceptual-constraint Section I 15–25 marks discriminator consensus 0.50

Whatever the format, the creative task names one concrete conceptual constraint the piece must be built around (the 2022–2024 pattern: divergent voices, a world within a world, a crossing that changes a character). The panel's favourite specific variant, called independently by three models: two characters or figures who attach INCOMPATIBLE MEANINGS to the same word, object or remembered event, with the divergence structuring the piece rather than arriving as a closing twist. Markers use the constraint as the primary line between conceptual and superficial engagement.

Each model's own prediction

DeepSeek V4: narrative exploring a stated conceptual tension, p 0.70
GPT-5.6 Sol: two characters, incompatible meanings of one word/object/event, p 0.56
Claude Opus 5: mandated constraint of the 2022–2024 type; irreconcilable ways of seeing, p 0.48
Claude Fable 5: named constraint (told story diverging from lived truth; world within a world), p 0.40
Grok 4.6: divergent-meanings task; divergence must structure the piece, p 0.36

Marker-feedback lineage: Marking feedback 2022, 2023, 2024 (constraint handling as the discriminator; one-dimensional voices)

Section I splits 15 + 10: compose, then analyse your OWN composition (2024 format restored) 5 of 6 models expect this

ee1-s1-split-15-10-self-reflection Section I 10–15 marks discriminator consensus 0.44

The 2019/2024 two-part structure returns: (a) a 15-mark narrative from the stimulus, (b) a 10-mark reflective analysis of how the candidate's own part (a) constructed a literary world — quoting one's own writing as evidence and justifying narrative choices in metalanguage. Four models note the same institutional motive: the reflection is the only exam vehicle for outcome EE12-5, and a final-year committee seeking comprehensive outcome coverage has every incentive to reinstate it. Opus's caution: the single 25-mark task still carries his base rate (0.60), keeping the split as the strong standing alternative rather than the favourite.

Each model's own prediction

DeepSeek V4: (a) 15 narrative + (b) 10 self-analysis, p 0.70
GPT-5.6 Sol: 15-mark imaginative + linked 10-mark critical analysis, p 0.52
Claude Fable 5: 2024 reprise as bold call, p 0.35–0.40
Claude Opus 5: split restored, p 0.33 (single task still 0.60)
Grok 4.6: 1(a) 15 + 1(b) 10 with own-quotation requirement, p 0.32
dissent: gemini-3.1-pro holds a single undivided 25-mark task

Marker-feedback lineage: Marking feedback 2019, 2024 (self-analysis quality; EE12-5 coverage)

In the practice paper: Q1

The live alternative: a full 25-mark critical/discursive extension of unseen extract(s) 5 of 6 models expect this

ee1-s1-discursive-critical-alternative Section I 25 marks alternative format consensus 0.33

The hedge every model carries at lower probability: Section I stays analytic/discursive as in 2021 and 2025 — a single 25-mark response extending one extract's ideas, or a critical comparison of two short unseen extracts (one manifesto on writing, one fiction) judging which account of why literary worlds are made is more persuasive. The discriminator named across models is a synthesised personal thesis about composer–text–responder rather than sequential treatment or a device catalogue. Prepare it as insurance, not as the main bet.

Each model's own prediction

Claude Opus 5: discursive/analytical extension of one writer-on-fiction extract, p 0.55; two-extract comparison, p 0.22
GPT-5.6 Sol: two-extract comparative evaluation, p 0.38
DeepSeek V4: single discursive extension alternative, p 0.30
Grok 4.6: two-extract critical comparison, p 0.22
Claude Fable 5: 2021/2025-pattern full critical response, p 0.20

Marker-feedback lineage: Marking feedback 2021, 2025 (extension vs summary; range of literary reference)

Transparency notes on this subject's evidence

deepseek's Section I part (b) prediction quotes 'something real behind the words' — the 2024 Baldwin task phrase — as if predicting it for 2026; clustered here as a FORMAT call (self-analysis part b) with the phrase-level content discounted as corpus echo.
fable and grok both cite real past stimulus phrases ('The king was pregnant', Pale Man, Kubla Khan; 'practise our own humanity') as evidence-trap or lineage references — verified against extract/english_ext_1.md as legitimate historical citation, not prediction leakage.
No model predicted a question tied to a single prescribed text; per the clustering rule none would have been clustered if it had.

Per our corpus-echo rule, agreement on details not derivable from the evidence pack counts as one vote, never as independent confirmation.

Watch list — worth having ready (7)

Gemini's dissent: a minimalist VISUAL stimulus (ambiguous landscape/abstract image) forcing on-the-spot conceptual translation (gemini 0.75 as format, 0.35 as bold call; opposed by the five extract-predicting models)
Five differentiated elective questions 2(a)–2(e) return as a final-year full-coverage gesture, sharing one quoted stem (fable 0.35, grok 0.36, opus 0.32) — breaking the 2024–2025 common-question run
A triadic/gradated directive — texts 'echo, unsettle or oppose' the stimulus claim — inviting a different verdict per text (opus trend call; grok's five stimulus-stance variants 0.28–0.34)
The task built verbatim on the module descriptor's one never-quoted clause, 'private, public and imaginary worlds' (fable 0.30)
A short poem as the Section I stimulus instead of prose (deepseek 0.35)
The stem widened to invite literary texts from the candidate's own wider reading, reversing the 2020–2021 'outside texts unnecessary' instruction (opus 0.22, citing 2025 feedback)
Section II carries a short shared stimulus/quoted statement rather than a bare conceptual question (opus's favoured design; deepseek dissents at 0.80 for a no-stimulus evaluative essay)