In February 2026, Education Minister Erica Stanford announced a nationally consistent reporting approach for Years 0–10 students, mandating five elements in every report: one of five progress descriptors — Emerging, Developing, Consolidating, Proficient, or Exceeding — for reading, writing, and mathematics; a narrative explanation of why that descriptor was chosen and how parents can support next steps; assessment results from standardised tools; a visual representation of progress over time; and an attendance record.

The framework was developed with principals’ associations and teachers, and trialled in 85 schools involving approximately 12,000 student assessment engagements.[20] It is well designed. The question this document addresses is not whether the reform is sound — it is whether the implementation environment can absorb it.

Alongside it, teachers are learning SMART, a new standardised assessment tool under a five-year, approximately AUD 21 million contract,[1] replacing e-asTTle for twice-yearly assessment of Years 3–10.

The reform does not land in isolation: teachers are simultaneously absorbing SMART, the new reporting framework, legacy data migration, a compulsory technology curriculum from 2027, and Government Target 72 (80% of Year 8 students at expected curriculum levels by December 2030). No single demand is unmanageable — the compound effect is the problem, and TALIS 2024 confirms New Zealand teachers already report change fatigue well above the OECD average.

The OECD’s Teaching and Learning International Survey (TALIS) 2024 provides the most comprehensive international dataset on teacher working conditions. The findings are striking, though they carry an important caveat: New Zealand did not meet TALIS 2024 participation rate standards, and all NZ-specific estimates should be interpreted with caution due to higher risk of non-response bias.[2]

With that caveat noted: New Zealand full-time teachers report working 47.5 hours a week — second-longest in the OECD behind Japan’s 55 — against an OECD average of 41.[3] The gap is not teaching time (21.5 hours against 22.7) but non-teaching work, especially administration (4.1 hours a week against an OECD average of three). More than 30% of teachers experience stress “a lot”, among the highest rates internationally and up five-plus points since 2018;[4] sixty-five per cent cite “changing requirements from authorities” as their top stressor — precisely what the 2026 reform represents.[3]

One finding deserves particular attention: peer-reviewed research by Jerrim and Sims (2021), analysing TALIS 2018 data across five English-speaking systems including New Zealand, found marking and assessment writing is the single task most strongly associated with poor teacher wellbeing[6] — not total hours, specifically the act of writing assessments and marking. Report writing is the concentrated form of that task, repeated 25 times a cycle.

International evidence reinforces this: a commissioned Scottish workload study found reporting processes extend working hours, and that reducing extensive after-hours marking may improve, not compromise, outcomes.[10]

One counterpoint: the NZCER 2024 National Survey found 90% of teachers enjoy their work and 75% report good morale — not every indicator is worsening; the profession cares deeply about the work but is structurally overloaded in specific, identifiable areas.

AI has already arrived in schools. The NZCER 2024 National Survey — a nationally representative sample of 639 teachers from 148 schools — found 46% reported using AI tools in their teaching.[11] A separate, self-selecting NZCER sample of 266 primary teachers found 69% using generative AI weekly.[12] Both point the same way: AI use is widespread, growing, and largely ungoverned.

Most rely on free chatbots: 73% report no school-funded access to premium tools, and only 8% of schools have a policy governing teachers’ AI use.[12] The Ministry has published principles-based guidance on AI in assessment, but no framework exists for AI-assisted report narrative generation.

Current market offerings for AI-assisted report writing operate as prompt-wrapper comment generators: a teacher selects a student, enters a few descriptors, the tool drafts a comment. That addresses the speed of a first draft while leaving unaddressed what makes a report meaningful — the teacher’s voice, the child’s context, the family’s reception needs, institutional review standards, and the longitudinal continuity of the child’s story.

The gap is architectural, not incremental. No New Zealand product, at the time of this analysis, consolidates multiple evidence sources through multi-axis calibration under governance into individually differentiated narrative reporting aligned to the five-element framework.

The governance gap is acute given student report data involves minors. A hallucinated claim about a child’s progress is not an inconvenience — it is a professional liability event for the teacher and potentially harmful to the child; that risk is structural to any AI-assisted reporting approach, including this one. The trust architecture must be at least as rigorous as anything applied to adult-facing AI tools, because the human in the loop is time-poor and reviewing output in batches of 25.

To understand the scale of the reporting task, the author developed a per-student time model based on stated assumptions — not measured data, but a structure for practitioner validation.[13] The model estimates the current reporting burden for a 25-student class at approximately 42 hours per cycle, including an IEP adjustment for the roughly 16% of students with additional learning needs. As a cross-check, two practising school leaders independently estimated approximately 50 hours per cycle — the same order of magnitude.[14]

The 42 hours break down across four phases: data collection and collation (16.8 hrs), report narrative writing (17.9 hrs), the peer review chain (5.1 hrs), and administrative completion (2.2 hrs) — 84 hours annually across two reporting rounds.

The model projects three scenarios for the engine’s impact once calibrated and operational: conservative (48.4 annual hours, 35.6 hours saved), moderate (33.0 hours, 51.0 saved), and optimistic by the third cycle (24.8 hours, 59.2 saved). Even the conservative scenario returns approximately one full working week a year — time for professional learning, differentiated planning, informal parent communication, or simply going home before dark during reporting season.

The institutional picture is larger still. For a 12-classroom primary school, the estimated total reporting burden is approximately 565–590 hours per cycle — over 1,100 hours a year across teacher direct time, syndicate leader review, deputy principal oversight, principal review, and coordination overhead.[15] These are illustrative estimates requiring practitioner validation, but the order of magnitude — roughly 140 eight-hour working days, more than half a full-time teaching position — is telling.

The review chain is not merely an overhead cost. It serves a dual function that most workload analyses overlook — a quality-assurance track (syndicate leader, deputy principal, principal review) and a professional-mentoring track through which institutional voice is shaped and individual teachers develop their reporting craft.

The full analysis explores this in depth — request the complete paper.

The value proposition of the School Report Narrative Engine is relationship transformation, not productivity.

Under the current model, the formal reporting cycle is the primary channel through which schools communicate a child’s progress to their whānau — two touchpoints a year; a parent finds out in July what happened in February. The engine enables a dual-output model: every real-time classroom observation captured through its mobile input channel can feed both the formal reporting log and an optional informal note sent home the same week, reviewed and sent by the teacher. (The mobile capture pipeline carries its own open privacy question, recorded in Volume II, Section 10.) This changes the whānau relationship from transactional to relational — a parent who has received three or four informal notes across a term arrives at the formal report already trusting the teacher’s understanding of their child.

The Equity Dimension

The transformation potential is not distributed equally, and that is precisely the point. Teachers in lower-decile schools — working with Pasifika families, recent migrants, or families facing poverty and housing instability — have the hardest time building relationships and the most to gain from a tool that makes warm, culturally attuned parent communication achievable within the working day. The greatest transformation potential sits where the relationship gap is widest: this is an equity argument, not a productivity one.

Making the New Way Achievable

The 2026 reform asks teachers to do more with new tools in a context where they are already under pressure. Every hour returned from reporting is an hour available for learning SMART, understanding progress descriptors, or professional learning. The engine makes the new way achievable, not the old way faster.

A Note on Scope

This document analyses one component of teacher workload: the reporting burden. It does not claim that reporting efficiency addresses the structural conditions of teaching — class sizes, staffing ratios, pay, or the demands collective agreement negotiations rightly address; tool-based efficiency and structural workload reduction are complementary, not substitutive. Nor does it resolve who should fund such tooling — a model where teachers pay personally for professional infrastructure is itself a workload-equity question collective processes rightly examine.

The engine is built on five structural elements, each requiring distinct calibration, addressing dimensions current market offerings leave untouched.

Element 1: Teacher Voice Calibration

Each teacher writes differently, and reports carry professional identity. The engine ingests a teacher’s previous reports as calibration data and learns their voice — sentence structure, vocabulary range, how they balance specificity with warmth — so output reads like the teacher who knows the child.

Element 2: Student Context Differentiation

Every child requires structurally different narrative, not just different data. A child on an IEP requires a fundamentally different narrative structure than a child who is exceeding expectations; the Ministry explicitly states that children with additional learning needs should not be automatically assessed as Emerging.[21]

Element 3: Whānau Reception Awareness

How you communicate progress to a Pasifika family differs from how you communicate to a recent migrant family. Cultural context is central to Te Tiriti o Waitangi obligations and the New Zealand Curriculum, and the teacher holds this knowledge through the whānau relationship — no consumer AI tool even knows to ask about it.

Element 4: Longitudinal Child Development Register

At every school transition, a teacher’s accumulated understanding of a child either does not transfer or arrives as a thin data extract. The engine maintains a child-owned longitudinal register, keyed to the National Student Number (NSN), that persists across transitions and bridges assessment-framework discontinuities (Te Whāriki Learning Stories in ECE through to NZC progress descriptors). This element activates only once a dedicated trust architecture is resolved.

Element 5: Peer Review Chain Calibration

The engine profiles each reviewer’s correction patterns and pre-applies institutional standards at generation time (reviewer awareness and consent for this profiling is an open design question; Volume II, Section 10), shifting the reviewer’s time from catching predictable compliance gaps to genuinely developmental feedback. The core performance metric is the first-pass approval rate through the review chain.

The contrast is stark: today’s market tools offer generic tone selection only, leaving whānau reception, longitudinal continuity, and review-chain calibration unaddressed — the five gaps the concept architecture is built to close.

Privacy by Non-Ingestion

The child’s name never enters the engine. The National Student Number (NSN) is the primary key for all student context; the teacher knows which NSN belongs to which child, the engine does not need to. The reporting log lives on the teacher’s own school drive — no hosted platform, no data sovereignty question. In a context where SMART was, at the time of this analysis, under active OIA scrutiny on data sovereignty questions,[19] this architecture is designed to minimise re-identification risk rather than to manage it after the fact. That comparison is scoped to offshore-transmission and hosted-platform risk; whether NSN-keyed context itself constitutes personal information under the Privacy Act remains an open question (Volume II, Section 10).

The buyer is the individual teacher, not the school or the Ministry. Schools will not move without MoE approval, and MoE moves at its own pace — but a teacher facing 30 reports, a two-week deadline, and the review chain’s iteration cost will pay for something that gives them their evenings back.

Teachers who use AI already pay for their own subscriptions; the product is the calibration workspace and profiling tools that make an existing AI subscription produce professional-grade reports instead of generic comments, priced so a teacher does not need institutional approval to buy it.

The operating cycle: the teacher maintains a structured reporting log; each round, they share it with the engine, which reads it cold, drafts reports grounded in the accumulated context, and returns an updated log for the teacher to review, edit, and submit. The engine is stateless between sessions — the log is the continuity mechanism.

Institutional adoption follows from the ground up: when enough teachers in a staffroom are using it and the principal notices reports arriving consistently cleaner through the review chain, adoption spreads organically — no procurement tender, no Ministry sign-off at market entry.

This ground-up path sits in deliberate tension with the governance case made earlier, and the tension deserves naming rather than glossing: a teacher adopting any AI reporting tool individually — including this one — should disclose that use to school leadership, and the concept treats that disclosure as a condition of professional use.

This is not a hosted platform: the product is methodology and tooling — calibration workspace, voice profiling, context framework, whānau guidance, and review chain profiling tools, built on the school’s own drive and management system.

This document is built on assumptions — some grounded in published research, others professional estimates — offered for testing against the experience of people who actually write school reports, review them, and live with the consequences. This document is designed to start that conversation.