A whole year of teaching, in teachers' own words

Across a single academic year, one secondary academy recorded 1,152 whole lessons through Starlight, turning ordinary classroom practice into private coaching for teachers and an anonymised evidence base for leaders. The data shows a school where high-impact teaching was already well embedded, and where several practices measurably strengthened over the year. Not from lessons prepared for an observer, but from more than a thousand everyday ones.

The challenge

Instructional coaching is one of the most reliable levers for improving teaching, and one of the hardest to scale. Formal observation is episodic, resource-heavy and, however well intentioned, easily experienced as judgement rather than support. Leaders are left with occasional snapshots rather than a continuous, credible picture of what teaching actually looks like week to week.

This academy wanted to grow professional development across the whole staff without adding observation workload, and to build a defensible evidence base for its teaching, particularly the practices that matter most for disadvantaged pupils. It needed to do this in a way teachers would trust: developmental, private, and clearly not a performance-management exercise.

The approach

Teachers recorded whole lessons on their own terms and uploaded the audio. Starlight produced a transcript and a structured, private coaching report within minutes: strengths evidenced in the teacher's own words, a small number of high-leverage next steps, and reflection prompts. Leaders saw none of this at an individual level. Instead they received anonymised, aggregated trends across departments and year groups to inform CPD priorities.

Over the year, 23 teachers recorded 1,152 lessons across 18 subjects and every main year group from 7 to 11, with lessons averaging around 60 minutes. The school then built its own analytical templates inside the platform: a five-part lesson-structure framework aligned to its pedagogical model, and an Ofsted-aligned checking-for-understanding template, with adoption roughly doubling over the year. This is ownership, not passive use: the school shaped the questions Starlight asked of its own teaching.

The results

Two findings stand out. First, the fundamentals of effective teaching were already strong and remarkably consistent, present across 18 subjects and all year groups rather than concentrated in a few classrooms or examination cohorts.

Second, and more unusually, the data documents a year of measurable improvement. Comparing a matched group of teachers who recorded throughout the year, several high-leverage practices strengthened between the first and second halves.

Matched cohort of 15 teachers who recorded throughout the year. Baseline: January–February 2026. Later: March–June 2026.

The formative-assessment story is the clearest. Whole-class checking through mini-whiteboards grew from around a quarter of lessons to close to two-thirds by spring, a genuine shift in individual practice rather than a change in who was recording. It was not confined to STEM: whiteboard-based checking in English Literature grew by 45 percentage points across the year, alongside strong gains in Mathematics and Physics. The explicit naming of misconceptions, a narrow and hard-to-fake marker, tripled in frequency the longer teachers used the platform.

Behind the numbers was something a dashboard cannot capture: teachers asking to take part. When Starlight was offered to staff as a professional development resource, the response was immediate.

“It's not often that you offer something as a CPD resource and get inundated with emails saying ‘me please, me please, me please’.”

What made the difference

Whole-lesson audio gave the AI far richer context than short clips could, so the feedback was specific rather than generic. Because uploading was quick and the report private, coaching became habitual rather than episodic, and that frequency is what turns feedback into growth. And because the architecture put teachers first, with voluntary recording, private-by-default reports and anonymised aggregation for leaders, participation was sustained and even broadened across a full year, which is rare for a reflective-practice platform. Trust was not a compliance overhead here. It was the reason the evidence base exists at all.

Why it matters for disadvantaged pupils

The practices strongest in this dataset, namely explicit vocabulary instruction, structured retrieval, frequent whole-class checking and specific, timely feedback, are precisely those research most robustly associates with narrowing attainment gaps. Whole-class checking removes the option for a quieter or less confident pupil to remain invisible: every pupil's understanding becomes visible at once. Live, in-lesson feedback closes the gap for pupils without academic support at home. Documented at scale from everyday teaching, this is exactly the evidence a school can stand behind for Pupil Premium accountability.

Based on anonymised, aggregated Starlight data. Findings describe classroom practice, not pupil attainment.