SURE Research · Volume 2 26,544 teachers ›

Ten years in the classroom. Under two points of improvement.

Across 26,544 teachers in 69 countries, first-year teachers scored 49.5% on classroom competency and teachers past their tenth year scored 51.4% — a correlation of r = 0.03. Experience alone does not repair teaching quality. Coaching does, and coaching has never scaled.

Read the research See how SURE works

Full reports free to read — no email Methodology & limitations published Figures reusable with attribution

Figure 1 · SURE Research Volume 2

A decade of classroom experience is worth 1.9 percentage points.

80% mastery

First-year teachers

49.5%

10+ years' experience

51.4%

Mean classroom competency score by years of teaching experience
Group	Mean score
First-year teachers	49.5%
Teachers with 10+ years	51.4%
Mastery threshold	80%

Base: 20,429 scored attempts with experience data, pooled 2022–2023. Correlation between years taught and competency: r = 0.03. Participation was voluntary, so the population skews motivated.

51%

Mean competency across the whole assessed population.

Pooled 2022–2023 · 46.8% (2022), 53.4% (2023)

0.49σ

Individualised coaching improves teaching practice by 0.49 standard deviations — among the largest effects education research has produced.

60 causal studies · cited in Vol 2

0–5% → 95%

Training alone moves 0–5% of teachers to use a new skill in class. With coaching on their own lessons, about 95%.

Vol 4 · Joyce & Showers

35 pts

The median teacher carries a 35-point spread between their strongest and weakest domain. No single course fits that profile.

Vol 2 · our data

2.4

Formal observations a tenured teacher receives per evaluation cycle — against roughly a thousand lessons taught a year.

Vol 5 · US district average

The gap is specific, not general. Experience does not repair it. Coaching does — and coaching has never scaled.

Read the four findings together and they write a design brief. Development has to begin from an individual diagnosis, because the median teacher's profile is jagged in its own particular way. It has to name one change at a time, because that is what a good coach does. It has to be continuous rather than episodic, because unexamined classroom time does not compound into skill. It has to observe practice rather than self-perception, because the weakest skill cluster in the data is the feedback loop itself.

And it has to break the expert-hours constraint — because coaching's economics collapse when every observation consumes an expert's hour. That last requirement is the one that has been impossible until now, and it is the reason this product exists.

The product, in one move

One recording. Two views. Nothing else asked of anyone.

A teacher records an ordinary lesson on their phone. That recording is the only input the system ever needs. From it, the teacher gets coaching on their own practice, and the school gets a pattern across the staff.

Recording

41:32

Grade 8 · Algebra · ordinary lesson

Class Insight ReportThe teacher's view · private

Strength · clear questioning Signal · wait-time after questions

08:12 Question posed — 1.1s pause before a student is called on

27:48 Worked example paced well; checks for understanding land

Your one thing

Pause 5 seconds after each question to widen student responses.

The school's view · aggregated

Behaviour mgmt

Assessment use

Communication

Aggregate drift across terms — sample

Patterns across staff. Never a ranking, never an individual.

Sample report shown. Scored on the CPAT framework — six domains, 25 sub-domains, the same architecture the research ran on. If something wasn't in the recording, nobody is asked to supply it.

Aggregate patterns for leaders. Individual detail for the teacher only.

How that is enforced →

How it works

From one lesson to a compounding loop.

Step 01

A teacher records

An ordinary lesson, on their own phone. No camera crew, no announced performance — a phone on the desk.

Step 02

SURE listens

The lesson is scored against CPAT's 25 sub-domains — the same instrument the research runs on. Student data is never captured and is always masked.

Step 03

One thing comes back

The teacher's private report names strengths, signals, and a single specific change — the way a good coach ends a debrief.

Step 04

The pattern emerges

Leaders see domain-level patterns across the staff and over time. Never a ranking, never an individual.

Then it repeats — the next recording measures whether the one thing landed. That is the loop observation never closed.

The framework

Scored on CPAT, not on vibes.

Six domains and 25 sub-domains of classroom practice — the same instrument that scored 1.2M+ responses in the research. Every claim in a report names the sub-domain it rests on, so a signal is checkable against the recording, not an opinion to take on trust.

Foundational Explorer Visionary Leader

Every teacher moves along a named progression — one thing at a time.

25 sub-domains · a sample

Learner behaviour management Use of assessment data Effective classroom communication Curriculum planning & progression Differentiated lesson plans Learner needs identification Teaching strategies & methods Reflective thinking skills Purpose of assessments Use of technology to support learning Effective lesson planning + 14 more

Highlighted: the profession's weakest sub-domain in the research — 39.9% correct across 754,971 responses.

The status quo

The observation cycle, side by side.

Annual observation is the instrument schools already pay for. Here is what it delivers, against what one recording habit delivers.

Observation as it exists	With SURE
2.4 formal observations per evaluation cycle	Every lesson a teacher chooses to record
An announced performance, prepared for an audience	Ordinary lessons, as they actually run
One observer's judgement — 0.14–0.37 reliability	Patterns across many lessons, on one published framework
A verdict at the end of a cycle	One specific next step after each recording
An expert hour consumed per observation	No expert hours — the constraint that kept coaching from scaling

Observation figures from SURE Research Volume 5: US large-district averages; single-observation reliability range from published studies.

Which door

Four kinds of buyer. Four different arguments.

Schools

"You have three observations a year to know what happens in a thousand lessons."

Volumes 2 & 3Read the case →

Groups & networks

A group can verify its finances, enrolment and results campus by campus — and cannot verify its teaching anywhere.

Volume 6Read the case →

Systems & standards

Every body that owns a teaching standard sees inside a classroom for hours, once every three to five years.

Volume 5Read the case →

Teacher education & PD

Training moves 0–5% of participants to change practice — and the proof has stopped at the reaction form.

Volume 4Read the case →

Whichever door you came in through — book a walkthrough or start with the research.

SURE Research

Most products cite research.This one is downstream of it.

A standing research programme, not a marketing campaign. We assess classroom competency at scale, publish what we find with the method and the limitations attached, and let anyone check the arithmetic.

26,544

teachers assessed

69

countries

10,562

schools

1.2M+

scored responses

25

sub-domains of practice

Volume 6 · Aug 2026 · Networks One Brand, a Thousand Classrooms 38% of international schools now sit in a group — and the teacher matters 2–3× more than the school. Volume 5 · Aug 2026 · Systems The Standard Nobody Sees One-off observation reliability: 0.14–0.37. An estimated 63% of single-lesson judgements are wrong. Volume 4 · Jul 2026 · Teacher education Why Training Doesn't Stick Transfer after training alone: 0–5%. With coaching on a teacher's own lessons: ~95%. Volume 3 · Jul 2026 · Schools Why Teachers Stop Showing Up About 70% of teachers in their first five years have left or seriously considered it. Replacing one: $11.9k–24.9k. Volume 2 · Jul 2026 · Schools The Gap Experience Doesn't Close 26,544 teachers, 69 countries. A decade moves competency under two points — r = 0.03. Open data · CC BY 4.0 The dataset The aggregate tables behind every volume — for analysts, academics, and anyone who wants to check our arithmetic.

How we research

Stated plainly: participation was voluntary, so this population skews motivated and graduate-qualified — which makes a near-50% average the conservative reading, not the alarming one. The method and the limitations ship inside every volume. Figures are free to reuse with attribution. The research mailing list receives research and nothing else.

Full method & what we don't claim →

Browse the research hub → Full reports free to read · PDFs cost an email

Trust

Rules that make recording feel safe.

SURE only works if teachers record willingly. Every rule here exists to protect that.

Teachers own their reports

The individual Class Insight Report reaches the teacher and no one else. Leadership sees aggregates only.

Never a ranking

Leadership views are patterns across staff and over time. No leaderboards, no per-teacher scores, no league tables.

Students are not the subject

Student data is never captured and is always masked. The recording exists to observe teaching, not children.

Nothing beyond the recording

The teacher starts every recording. If something wasn't in it, nobody is asked to supply it — no forms, no self-reports, no uploads.

Privacy law came first

Recording and retention were designed around India's DPDPA, FERPA and GDPR — researched before the feature was built, not retrofitted.

Procurement answers, ready

A procurement pack — DPA, security overview, data-flow map — walked through with your team before any commitment.

Early-access voices

Coaching that finally feels like trust.

My teachers stopped seeing observation as a threat. They see the patterns, I see the patterns, and the detail stays with them. That changed everything.

Priya NairHead of School, K–8

One tap after a lesson and I get my One Thing. No forms, no waiting for an annual review. It's the first coaching that actually fits my week.

Marcus BellSecondary Teacher

Knowing students are never part of the data, and the audio is only ever a recording teachers choose to make, made this an easy yes for our families.

Hana SatoGrade 5 Teacher

Illustrative quotes shown for this early-access preview.

Mapped to the frameworks you already answer to

Ofsted KHDA CIS IB CAEP World Bank Teach

Questions

Asked in every first conversation.

Who sees a teacher's report?

The teacher, and no one else. The architecture says so, not a policy document: the teacher starts every recording, the individual Class Insight Report reaches them alone, and leadership sees domain-level patterns across the whole staff. Never a ranking, never an individual. What teachers own →

What happens to student voices?

Student data is never captured and is always masked. The system observes teaching — questioning, wait-time, feedback, behaviour management — not children.

How much work does this add for a teacher?

One action: record an ordinary lesson on a phone. No forms, no uploads, no self-assessment. The report and the one thing come back without anything else being asked.

Can I trust an AI's read of a lesson?

Treat it as a signal, not a verdict. Scoring runs on CPAT — six domains, 25 sub-domains, published method — and every claim in a report names the sub-domain it rests on, so it can be checked against the recording itself. The research behind the instrument, limitations included, is public.

What does it cost?

Pricing is banded per teacher, per year — no per-recording metering, so nobody is taxed for practising. Pilots are fixed-price and fixed-duration. The full band sheet is shared in the first walkthrough, before any commitment.

What do leaders actually see?

Domain-level patterns across the staff and their movement over time — where the school is strong, where next term's PD should aim. Aggregates only, on the same 25 sub-domains the research uses.

Does it work?

The mechanism is the best-evidenced intervention in the field: individualised coaching moves practice by 0.49σ, and coaching on a teacher's own lessons moves transfer from 0–5% to roughly 95%. Our own product efficacy is a younger claim — we publish numbers with denominators when we have them, and nothing before.

Which frameworks does it align to?

Reports map to the frameworks schools already answer to — Ofsted, KHDA, CIS, IB, CAEP and World Bank Teach — so the evidence lands in language an inspection already speaks.

Correct-response rate by sub-domain, ranked weakest to strongest
Sub-domain	Correct
Learner behaviour management	39.9%
Professional ethics & values	41.7%
Curriculum planning & progression	42.5%
Learner needs identification	44.8%
Use of assessment data	45.1%
Integrating tech, pedagogy & content	45.8%
Purpose of assessments	46.5%
Inter-disciplinary lesson planning	47.0%
Types of assessment	48.4%
Learning theories in practice	49.3%
Differentiated lesson plans	49.3%
Theories of child development	51.4%
Activities for PD plan	53.4%
Responsible use of technology	56.5%
Accountability towards learners	57.1%
Classroom resources & safety	57.3%
Effective lesson planning	59.4%
Reflective thinking skills	59.7%
Types of curriculum	60.4%
Delivery of a discipline-specific lesson	62.1%
Teaching strategies & methods	62.7%
Purpose of curriculum design	63.1%
Use of technology to support learning	63.9%
Approaches to discipline-specific teaching	66.2%
Effective classroom communication	71.9%

What	Access	Why
Full report, HTML, all chapters	Free	Citation, links, and a place on the shortlist
Key findings pages	Free	Built to be quoted and lifted
Methodology & limitations	Free	This is the validation surface. Hiding it defeats the point.
Figures — PNG, SVG, CSV, embed	Free	Attribution travels with the chart
Research series subscription	1 field	A standing channel across a long decision cycle
Designed PDF	3 fields	The artefact forwarded to the rest of the buying group
The dataset, CC BY 4.0	Email + org	Analyst and academic reuse under an attribution licence
"Where does your school stand" benchmark	Progressive	An interaction, not information — and a real intent signal

Across 26,544 teachers in 69 countries, mean classroom competency was 51%.

A decade of experience moves it under two points. Five volumes work through what sits inside that number, and what it asks of the people who run schools.

Every volume is free to read in full. The method and the limitations ship with each one, and the figures are free to reuse with attribution. Participation in the assessment was voluntary, so the population skews motivated and graduate-qualified — which makes a near-50% mean the conservative reading, not the alarming one.

Showing all 5 volumes

Groups & networks August 2026 ≈18 min read One Brand, a Thousand Classrooms 38% of international schools now belong to a group — and the teacher a student is assigned explains 2–3× more of the variation in achievement gains than the school they attend. Inside this paper The network era has arrived — 52% of schools in development are group-affiliated The spread that breaks the promise lives inside each campus, not between them Three million lessons hosted a year; roughly 0.2% ever seen Read the paper 4 figures · method & limitations included
Systems & standards August 2026 ≈22 min read The Standard Nobody Sees A single lesson observation scores between 0.14 and 0.37 on reliability. An estimated 63% of one-off lesson judgements are wrong. Inside this paper What the reliability of one observation actually is, and what it can carry How often a one-off judgement lands on the wrong teacher 2.4 formal observations per cycle against ~1,000 lessons taught a year Read the paper 4 figures · method & limitations included
Teacher education July 2026 ≈20 min read Why Training Doesn’t Stick Training alone produces classroom implementation in roughly 0–5% of participants. With coaching on teachers’ own lessons: about 95%. Inside this paper What survives the walk from a training room back to a classroom Coaching on a teacher’s own lessons: 0.49 SD on instruction across 60 causal studies Where the proof stopped — why the reaction form became the evidence Read the paper 4 figures · method & limitations included
Schools July 2026 ≈22 min read Why Teachers Stop Showing Up About 70% of teachers in their first five years have left the classroom or seriously considered it. Replacing one costs $11,900–$24,900. Inside this paper What the first five years do to a teacher, and how early the decision forms The full cost of one replacement, hiring and lost practice included How that cost compounds across a staff, term after term Read the paper 4 figures · method & limitations included
Schools July 2026 ≈24 min read The Gap Experience Doesn’t Close First-year teachers scored 49.5% on classroom competency; teachers with ten or more years scored 51.4% — a correlation of r = 0.03. Inside this paper A decade of experience moves competency under two points The gap is jagged, not general — which sub-domains move and which never do What a pooled mean of 51% conceals across 26,544 teachers Read the paper 4 figures · method & limitations included

Process How we research Method, independence, the dataset, and what this series does not claim. Open data The data set All 25 sub-domains with their correct-response rate and base. CC BY 4.0.

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Research volume · August 2026

One Brand, a Thousand Classrooms

38% of the world’s 15,075 international schools now belong to a group — and the variation that decides whether a group’s promise is kept sits inside each campus, not between them: the teacher a student is assigned explains 2–3× more of the variation in achievement gains than the school they attend (Nye, Konstantopoulos & Hedges, 2004).

Volume 6 · Written for groups and networks · ~18 min read · Read free · PDF needs email

The four findings

What this volume found

52%

of international schools currently in development are group-affiliated

Among international schools in development at the time of ISC Research’s 2026 snapshot, 52% are group-affiliated — a higher share than the 38% already open. Source: ISC Research (2026), Global International Schools Snapshot 2026.

2–3×

more achievement-gain variation is explained by the teacher a student gets than by the school they attend

In a study of randomised class assignments, the teacher a student is assigned explained 2–3× more of the variation in achievement gains than the school the student attends; in reading, the between-teacher variance component was over twice the between-school component at grade 2 and over three times at grade 3. Source: Nye, Konstantopoulos & Hedges (2004). Population and base are as published; we did not re-analyse the study.

17–28%

annual teacher turnover in international schools

Annual teacher turnover in international schools is reported as a range of 17–28%. We have not been able to attach a single base, period or source to that range, so we publish it as a range and mark the gap rather than imply a precision we do not have.

Executive summary

The promise a group makes is made at network level, experienced at classroom level, and managed at the one level in between where the signal cancels out #

A family enrolling at any campus, in any city, is choosing the same education. Every network brochure makes a version of this promise, and every network leader knows the uncomfortable truth underneath it: the group can verify its finances, its enrolment, its fee income and its exam results campus by campus, and it cannot verify its teaching anywhere. This report examines that gap with the evidence available, reading the research on teacher effectiveness and school networks against ISC Research’s market data and our own competency data on 26,544 teachers. Four findings follow.

The network era has arrived, and it concentrates the risk. 38% of the world’s 15,075 international schools now belong to a group, and 52% of the schools still in development are group-affiliated: the sector’s future is majority networked. A group inherits the classroom-visibility problem every school has, multiplied by every campus it operates, and its brand concentrates the consequences: one weak campus is no longer one school’s problem, it is every sister campus’s problem.

The spread that breaks the promise lives inside each campus. The research on teacher effectiveness is unambiguous: which teacher a student gets matters 2–3× more than which school they attend, and most of the variation in teaching quality sits within schools rather than between them. Our own data puts a 35-percentage-point spread inside the median teacher’s own profile. The campus averages a head office tracks are calm surfaces over exactly the variation parents actually experience, classroom by classroom.

Oversight thins exactly as the group grows. A 60-campus group with 3,000 teachers hosts roughly three million lessons a year; a five-person group academic team observing lessons every school day would see about 0.2% of them. Meanwhile annual teacher turnover of 17–28% means each campus’s staffroom substantially rebuilds itself between head-office visits, and the newest campuses, where the brand is least embedded, hire fastest and churn most.

The group’s instruments report too late and train too broadly. Exam results arrive annually and describe teaching the students concerned have already lived. Group professional development calendars deliver the same content to jagged, campus-specific profiles, and the transfer evidence shows training without classroom-level follow-up moves roughly 0–5% of teachers to change practice. Neither instrument can locate a weakening campus while it is weakening.

The closing sections describe what the evidence specifies instead: a network operating layer for teaching quality, built on ordinary recorded lessons, that gives every teacher a private coach, every principal a campus pattern, and the group a continuous implementation picture of its own promise, campus by campus, practice by practice, term by term. The group that runs its brand on evidence will be the one whose flagship is not a place but a standard.

The PDF edition

Get Volume 6 as a PDF#

The whole volume is free to read on this page. The PDF is the same research in a form you can circulate to a board or a principals’ group — and it carries the extended chapters for findings three and four, which are summarised but not yet expanded in this HTML edition.

PDF edition

Three fields, and we disclose all three#

Work email, name, role. That is the entire form, and nothing you have just read sits behind it.

A work address reaches you more reliably than a personal one — but either will do, and we will not chase you for the other.

Three fields, disclosed in full: work email, name, role. Front-end demo — this form is not wired to anything yet, so nothing is sent and nothing is stored.

1 · The promise a network makes #

The value of a school group is a claim about sameness. A family relocating from Singapore to Dubai keeps the same curriculum, the same standards, the same idea of good teaching; a family choosing the newest campus is promised the quality of the oldest one. The brand is the promise that geography does not matter. This is what justifies the premium, drives the cross-border enrolment pipeline, and makes an acquisition worth more inside the group than outside it.

Every function of a modern group can verify its part of that promise except one. Finance consolidates by campus and reports monthly. Admissions and enrolment are live dashboards. Facilities, safeguarding and compliance run on audits and checklists. Examination results arrive annually, benchmarked across the network. And teaching, the activity all of it exists to support, and the thing the family is actually buying, reports through none of these systems. What a head office knows about classrooms is what campuses tell it, plus what a small academic team sees on visits: the same two instruments, self-report and the occasional observed lesson, whose limits earlier volumes of this research series documented for schools and for the bodies that set their standards.

A network is, in this sense, a standard-setter in miniature. It writes a teaching model; every staffroom in the network enacts it; the centre cannot see the enactment. This volume examines what that blindness costs a group specifically, because the group case is sharper than the single-school case in every direction: the variation is larger, the turnover is faster, the sampling is thinner, and the brand ties every campus’s reputation to the classroom the centre has never seen.

2 · Finding one: the network era has arrived, and it concentrates the risk #

The international schools sector is consolidating into groups, quickly. ISC Research’s 2026 snapshot counts 15,075 international schools serving 7.7 million students with 730,000 staff and USD 69.3 billion in annual fee income. Of those schools, 38% are now affiliated to a group. The forward pipeline is more striking still: 52% of schools currently in development are group-affiliated. The sector’s growth since 2020, 8% more schools, 15% more students, 27% more fee income, is increasingly group-shaped growth. What was once a market of standalone schools with a few large operators is becoming a market of networks.

Consolidation professionalises almost everything it touches: procurement, governance, safeguarding, facilities, marketing, admissions. But it also concentrates a risk that standalone schools carry only singly. A group’s campuses share a name, and the name is doing commercial work at every campus simultaneously. A quality failure at one campus is priced into all of them; a single inspection downgrade, a single cohort of poor results, a single exodus of teachers travels through the brand faster than any operational report travels up it. The larger the network, the larger the surface area of the promise, and the smaller the share of it the centre has ever seen.

The concentration cuts the other way too, and this is the opportunity the rest of this report builds toward: a group is also the one structure in education that can fix teaching quality at scale, because it owns the standard, employs the teachers, and answers to itself. What it has lacked is not authority but sight.

3 · Finding two: the spread that breaks the promise lives inside each campus #

Where should a group expect its quality problem to live? The instinctive answer is “at the weak campus”, and it is the wrong answer. The research on teacher effectiveness has converged on a finding with awkward implications for any organisation that manages quality at building level: the variation that matters is inside the buildings.

Nye, Konstantopoulos and Hedges, working with randomised class assignments, found that the teacher a student is assigned explains 2–3× more of the variation in achievement gains than the school they attend; in reading, the between-teacher variance component was over twice the between-school component at grade 2 and over three times at grade 3. Hanushek and Rivkin’s review of the value-added literature reaches the same structural conclusion: the variation in teacher effectiveness is large, and much of it sits within schools rather than between them.

Figure 1

In reading at grade 3, the teacher a student is assigned accounts for over three times the achievement-gain variation that the school does

The teacher a student is assigned (between-teacher component)

over 3×

The school a student attends (between-school component)

1× (reference)

Relative size of variance components in student achievement gains, reading, grade 3, with the between-school component scaled to 1. Source: Nye, Konstantopoulos & Hedges (2004), How Large Are Teacher Effects?, Educational Evaluation and Policy Analysis, 26(3), 237–257. The published finding is “over three times” at grade 3 and “over twice” at grade 2; the upper bar is drawn at exactly three times, the stated lower bound, so the chart cannot overstate the result. Study population and base are as published; we did not re-analyse the data. Every value is printed beside its bar, so no meaning here depends on colour.

Relative size of variance components in achievement gains (reading, grade 3), between-school component scaled to 1. Source: Nye, Konstantopoulos & Hedges (2004).
Variance component	Relative size (between-school = 1)
Between-teacher (the teacher a student is assigned), grade 3	Over 3
Between-school (the school a student attends), grade 3	1 (reference)
Between-teacher, grade 2, for comparison	Over 2

Our own data shows what this looks like from inside the teacher. The median teacher carries a 35-percentage-point spread between their strongest and weakest domain, and 24% of teachers are simultaneously below 40% in one domain and above 70% in another. And the profession’s weakest skills cluster in the fundamentals that decide an ordinary lesson: in the 2023 edition, learner behaviour management drew correct responses on 39.9% of 15,047 attempted responses, curriculum planning and progression on 42.5% of 30,165, and use of assessment data in achieving learner outcomes on 45.1% of 30,205.

Figure 2

Every one of the ten weakest sub-domains sits below 50% correct — and they are the skills an ordinary lesson runs on

Learner behaviour management

39.9%

Professional ethics and values

41.7%

Curriculum planning and progression

42.5%

Learner needs identification

44.8%

Use of assessment data in achieving learner outcomes

45.1%

Integrating technology, pedagogy and content

45.8%

Purpose of assessments before, during and after a lesson

46.5%

Inter-disciplinary lesson planning

47.0%

Types of assessment

48.4%

Learning theories in teaching practice

49.3%

The ten weakest of the 25 sub-domains measured in the 2023 edition of our competency research, ranked by correct-response rate. Each estimate is a simple proportion of correct answers among attempted responses; bars are drawn on a 0–100% axis and every value is printed beside its bar. Base for the sub-domain analysis: 754,971 attempted responses from 15,548 teachers, 2023 edition. Per-sub-domain bases range from 15,047 to 45,295 attempted responses across these ten rows and are listed row by row in the table behind this chart and in the annexure. Tenth place is shared: learning theories in teaching practice and differentiated lesson plans both score 49.3%, and the annexure carries both.

The ten weakest of 25 sub-domains by correct-response rate, 2023 edition. Base overall: 754,971 attempted responses from 15,548 teachers.
Rank	Sub-domain	Core domain	Correct (%)	n (attempted responses)
1	Learner behaviour management	Learning Environment	39.9	15,047
2	Professional ethics and values	Ethics and Values	41.7	15,127
3	Curriculum planning and progression	Curriculum, Pedagogy and Technology	42.5	30,165
4	Learner needs identification	Curriculum, Pedagogy and Technology	44.8	30,185
5	Use of assessment data in achieving learner outcomes	Learning Assessment & Data	45.1	30,205
6	Integrating technology, pedagogy and content	Curriculum, Pedagogy and Technology	45.8	30,242
7	Purpose of assessments before, during and after a lesson	Learning Assessment & Data	46.5	30,187
8	Inter-disciplinary lesson planning	Student Engagement	47.0	30,204
9	Types of assessment	Learning Assessment & Data	48.4	45,295
10	Learning theories in teaching practice	Curriculum, Pedagogy and Technology	49.3	30,210

For a network, the consequence is precise: the campus average, the unit at which head office perceives quality, is the exact resolution at which the problem is invisible. Two campuses can post identical results while one holds a cluster of classrooms in quiet difficulty. The brand promise is made at network level, experienced at classroom level, and managed at the one level in between where the signal cancels out.

4 · Finding three: oversight thins exactly as the group grows #

A 60-campus group with 3,000 teachers hosts roughly three million lessons a year; a five-person group academic team observing lessons every school day would see about 0.2% of them. That is arithmetic on a stated group shape, not a measurement of any particular group.

Meanwhile annual teacher turnover of 17–28% means each campus’s staffroom substantially rebuilds itself between head-office visits, and the newest campuses, where the brand is least embedded, hire fastest and churn most.

5 · Finding four: the group’s instruments report too late and train too broadly #

Exam results arrive annually and describe teaching the students concerned have already lived.

Group professional development calendars deliver the same content to jagged, campus-specific profiles, and the transfer evidence shows training without classroom-level follow-up moves roughly 0–5% of teachers to change practice (Joyce & Showers, 2002). Neither instrument can locate a weakening campus while it is weakening.

Annexure: all 25 sub-domains in our research data #

Correct-response rate for every sub-domain measured, ordered from weakest to strongest. Each estimate is a simple proportion of correct answers among attempted responses; n is the number of attempted responses in our research data.

Annexure

Eleven of the 25 sub-domains sit below 50% correct, and the weakest of them are the fundamentals of an ordinary lesson

Learner behaviour management

39.9%

Professional ethics and values

41.7%

Curriculum planning and progression

42.5%

Learner needs identification

44.8%

Use of assessment data in achieving learner outcomes

45.1%

Integrating technology, pedagogy and content

45.8%

Purpose of assessments before, during and after a lesson

46.5%

Inter-disciplinary lesson planning

47.0%

Types of assessment

48.4%

Learning theories in teaching practice

49.3%

Differentiated lesson plans

49.3%

Theories of child development

51.4%

Activities for professional development plan

53.4%

Responsible use of technology

56.5%

Accountability towards learners

57.1%

Classroom resources and safety

57.3%

Effective lesson planning

59.4%

Use of reflective thinking skills for effective teaching and learning

59.7%

Types of curriculum

60.4%

Delivery of a discipline-specific lesson

62.1%

Teaching strategies and methods

62.7%

Purpose of curriculum design and development

63.1%

Use of technology to support teaching and learning

63.9%

Approaches to discipline-specific teaching

66.2%

Effective classroom communication

71.9%

All 25 sub-domains measured in the 2023 edition, ordered weakest to strongest. Base overall: 754,971 attempted responses from 15,548 teachers. Per-sub-domain bases range from 15,047 to 60,385 attempted responses and are listed row by row, with each row’s core domain, in the table behind this chart. Bars are drawn on a 0–100% axis and every value is printed beside its bar, so no meaning depends on colour.

All 25 sub-domains, 2023 edition: core domain, correct-response rate and base n, ordered from weakest to strongest. Base overall: 754,971 attempted responses from 15,548 teachers.
Sub-domain	Core domain	Correct (%)	n (attempted responses)
Learner behaviour management	Learning Environment	39.9	15,047
Professional ethics and values	Ethics and Values	41.7	15,127
Curriculum planning and progression	Curriculum, Pedagogy and Technology	42.5	30,165
Learner needs identification	Curriculum, Pedagogy and Technology	44.8	30,185
Use of assessment data in achieving learner outcomes	Learning Assessment & Data	45.1	30,205
Integrating technology, pedagogy and content	Curriculum, Pedagogy and Technology	45.8	30,242
Purpose of assessments before, during and after a lesson	Learning Assessment & Data	46.5	30,187
Inter-disciplinary lesson planning	Student Engagement	47.0	30,204
Types of assessment	Learning Assessment & Data	48.4	45,295
Learning theories in teaching practice	Curriculum, Pedagogy and Technology	49.3	30,210
Differentiated lesson plans	Student Engagement	49.3	45,297
Theories of child development	Curriculum, Pedagogy and Technology	51.4	30,025
Activities for professional development plan	Teacher Growth & Professional Development	53.4	30,189
Responsible use of technology	Ethics and Values	56.5	15,095
Accountability towards learners	Ethics and Values	57.1	15,109
Classroom resources and safety	Learning Environment	57.3	30,211
Effective lesson planning	Student Engagement	59.4	30,209
Use of reflective thinking skills for effective teaching and learning	Teacher Growth & Professional Development	59.7	30,213
Types of curriculum	Curriculum, Pedagogy and Technology	60.4	30,248
Delivery of a discipline-specific lesson	Student Engagement	62.1	30,215
Teaching strategies and methods	Curriculum, Pedagogy and Technology	62.7	30,263
Purpose of curriculum design and development	Curriculum, Pedagogy and Technology	63.1	30,212
Use of technology to support teaching and learning	Curriculum, Pedagogy and Technology	63.9	30,231
Approaches to discipline-specific teaching	Curriculum, Pedagogy and Technology	66.2	30,202
Effective classroom communication	Learning Environment	71.9	60,385

Methodology and limitations #

This volume is built on three sources, and it is worth being exact about which claim comes from which. Our own competency data: 26,544 teachers assessed across 69 countries and 10,562 schools, producing more than 1.2 million scored responses, in the 2022 and 2023 editions. The programme measures six domains; the 25 sub-domains used in Figure 2 and the annexure were measured in the 2023 edition, on a base of 754,971 attempted responses from 15,548 teachers. Pooled mean competency is 51% — 46.8% in the 2022 edition and 53.4% in the 2023 edition. Sector data: the counts of schools, students, staff and fee income are ISC Research’s, from the Global International Schools Snapshot 2026, not ours. External literature: the teacher-effect findings are cited as published from Nye, Konstantopoulos and Hedges (2004) and Hanushek and Rivkin (2012), and the transfer evidence from Joyce and Showers (2002).

What this evidence cannot tell you #

Participation in our assessment was voluntary, so the population skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national workforce, and no figure on this page should be read as a national average.
The assessment measures knowledge-for-practice, not enacted classroom behaviour. A sub-domain score describes what a teacher knows to do, not what an observer would see them do.
The 2022 and 2023 editions share no common items, so no growth claim can be made between them. The pooled 51% sits between 46.8% and 53.4% because the editions differ, not because teachers improved.
The 17–28% turnover range carries no single base, period or source we can stand behind, so we publish it as a range and mark the gap rather than dress it up.
The teacher-effect and transfer findings are external literature, cited as published. We did not re-analyse those datasets, and their populations are not ours.
The oversight arithmetic in finding three — 60 campuses, 3,000 teachers, roughly three million lessons, about 0.2% seen — is an illustrative calculation on a stated group shape, not a measurement of any particular group.
We have no named school case studies, no product efficacy results and no adoption numbers, and we will not publish any until they exist.
Findings three and four appear here in summary only; the extended chapters are in the PDF edition and their HTML editions are in preparation.

The full method sits on How we research, the dataset is described on The dataset, and the standing methodology note is at Methodology.

References #

Hanushek, E. A., & Rivkin, S. G. (2012). The Distribution of Teacher Quality and Implications for Policy. Annual Review of Economics, 4, 131–157.
ISC Research (2026). Global International Schools Snapshot 2026.
Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD.
Nye, B., Konstantopoulos, S., & Hedges, L. V. (2004). How Large Are Teacher Effects? Educational Evaluation and Policy Analysis, 26(3), 237–257.
Our own competency data, 2022 and 2023 editions — described on How we research and The dataset.

If you run a group and want to see what a continuous, campus-by-campus picture of teaching would actually look like, here is how the platform works.

Read next

Where this evidence goes next

For groups and networks

The same evidence, arranged around a head office’s decisions

What a group can do with a finding that says the variation it needs to manage is inside its campuses, not between them.

Volume 6 · Key findings

The four findings of this volume on one page

38% group-affiliated, 52% of the pipeline, the 2–3× teacher effect, and 17–28% turnover.

Volume 5 · August 2026

The Standard Nobody Sees

A single lesson observation scores between 0.14 and 0.37 on reliability — the instrument a head office leans on hardest.

Volume 4 · July 2026

Why Training Doesn’t Stick

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises it to approximately 95% (Joyce & Showers, 2002).

Volume 2 · July 2026

The Gap Experience Doesn’t Close

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023.

All research volumes

Cite this

Quoting Volume 6

Permalink
/#/research/one-brand-a-thousand-classrooms/
Open this volume at its permanent path. Every chapter heading, figure and headline statistic on this page carries a frozen id, so you can link to a single finding rather than the whole volume.

Plain citation
SURE (2026). One Brand, a Thousand Classrooms. Research volume 6, August 2026.

APA
SURE. (2026, August). One brand, a thousand classrooms (Research volume 6).

No individual author byline is assigned to the volumes in this series, so the corporate author is the correct one to cite. If you are quoting a figure, quote it with its base: the sub-domain figures rest on 754,971 attempted responses from 15,548 teachers in the 2023 edition, and the teacher-effect figure belongs to Nye, Konstantopoulos and Hedges (2004), not to us.

Licence
We have not yet published a formal reuse licence for this series. Until we do: quote from this volume with attribution to SURE and a link back to this page, and ask us first if you need something broader than that.

Subscribe

One email per volume, with the findings and their bases in the email itself, so you can decide from the email whether the volume is worth your time.

Subscribe to the research

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Research volume · August 2026

The Standard Nobody Sees

A single lesson observation scores between 0.14 and 0.37 on reliability, and an estimated 63% of one-off lesson judgements are wrong — and that observation is the instrument through which a body that owns a teaching standard finds out whether the standard is enacted.

Volume 5 · Written for curriculum bodies, inspectorates and systems · ~22 min read when the full manuscript publishes · Read free · PDF needs email

The four figures

What this volume argues from#

Volume 5 publishes three headline findings, and the third of them carries two numbers that only mean anything set against each other — so it is shown here as two cards, marked as the two halves of one published finding rather than as two findings.

0.14–0.37

is the reliability of a single lesson observation as a measure of the teaching it stands for

A single lesson observation scores between 0.14 and 0.37 on reliability. Source: SURE research, Volume 5, August 2026. Published as a range rather than a point; no midpoint is published and none is calculated here. The full source note is not reproduced on this page, so the coefficient type, the base (n) and the period are not shown, and none is supplied in their place.

63%

of one-off lesson judgements are wrong, on the estimate this volume argues from

An estimated 63% of one-off lesson judgements are wrong. Source: SURE research, Volume 5, August 2026. Published as an estimate and quoted here as one. Its complement is not published and has not been calculated here. The full source note is not reproduced on this page, so no base (n) and no period are shown, and none is supplied.

2.4

formal observations a tenured teacher receives per evaluation cycle in the average large US district

A tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. Source: SURE research, Volume 5, August 2026. This card and the next are the two halves of that one published finding. US-derived. The length of an evaluation cycle is not published with it on this page, and none is supplied.

~1,000

lessons that same teacher teaches in a year — the population those 2.4 observations are drawn from

A tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. Source: SURE research, Volume 5, August 2026. The 1,000 is published as a round approximation and is not treated as an exact count. No share of lessons observed is stated here, because the two figures are counted over different periods and no cycle length is published with them.

Executive summary

A standard written for thousands of classrooms is verified through a sample that thin, with an instrument that unstable #

A teaching standard is a description of what good practice looks like inside a room. Every body that writes one — a curriculum authority, an inspectorate, an accreditation scheme, a ministry, the owner of a framework such as Ofsted, KHDA, CIS, IB, CAEP or World Bank Teach — depends in the end on someone going into that room and watching. This volume is about that instrument, and about what the published evidence says the instrument can carry. Three findings follow, and they approach the same instrument from three directions.

One observed lesson is an unstable measure. A single lesson observation scores between 0.14 and 0.37 on reliability. The figure is published as a range rather than a point, and no midpoint is published or calculated here. The full source note for it is not reproduced on this page, so the coefficient type, the base and the period are not shown and are not supplied. What the range describes is one observed lesson used as a measure of the teaching it is meant to represent — not observation as a practice, and not any named framework or protocol.

Most of what a one-off judgement concludes is wrong. An estimated 63% of one-off lesson judgements are wrong. It is published as an estimate and it is quoted here as one. No confidence interval is published for it, so none is drawn anywhere on this page, and its complement is not published and has not been calculated. It is an estimate about judgements built from a single observed lesson as a class, not a measurement of any particular district, inspectorate, accreditation scheme or school.

The sample is thinner than the instrument is unstable. A tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. The two numbers are published together as one finding and are quoted together here. They are counted over different periods — observations per evaluation cycle, lessons per year — and the length of an evaluation cycle is not published with them on this page, so no per-year observation figure and no share-of-lessons-observed percentage appears anywhere on this page.

None of this is an argument against observation, and none of it is an argument against standards. It is an argument about resolution. A body that owns a standard can audit the documents a system produces about teaching; what it cannot presently do is see the teaching itself at a resolution the standard was written at. Written, adopted, inspected against, and unseen in the one place it was meant to live. That is the standard nobody sees.

How to read the chapters below. The findings, the figures and the citations on this page are final and citable today. The chapter prose is structural specimen text, marked as such wherever it appears, published in that form so the template can be reviewed before the full manuscript is typeset. Every chapter opens with this volume’s real finding, stated in full; everything marked specimen describes what that chapter will argue rather than arguing it.

The PDF edition

Get Volume 5 as a PDF#

Everything on this page is free to read and stays free. The PDF edition of The Standard Nobody Sees publishes with the full manuscript, not before it, and leaving three fields is how you get it on the day it does.

PDF edition

Three fields, and we disclose all three#

Work email, name, role. That is the entire form, and nothing you have just read sits behind it.

A work address reaches you more reliably than a personal one — but either will do, and we will not chase you for the other.

Three fields, disclosed in full: work email, name, role. Front-end demo — this form is not wired to anything yet, so nothing is sent and nothing is stored.

Specimen edition. The findings, figures and citations on this page are final. The chapter prose marked below is structural specimen text, shown so the template can be reviewed before the full manuscript is typeset.

1 · The question — how does a body that owns a teaching standard know it is enacted? #

Three published figures set the terms of the question. A single lesson observation scores between 0.14 and 0.37 on reliability. An estimated 63% of one-off lesson judgements are wrong. And a tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. Source: SURE research, Volume 5, August 2026. The full source note for each of the three is not reproduced on this page, so no base (n) and no period is shown for any of them here, and none is supplied in their place.

This section sets out the question the volume is written to answer: how a body that owns a teaching standard establishes that the standard is enacted in the rooms it was written for, rather than merely adopted by the organisations that answer to it.

The analysis then distinguishes the two things such a body can inspect — the documents a system produces about its teaching, and the teaching itself — and states which of the two its existing instruments actually reach.

The chapter closes by naming the reader it is written for, the body that writes or inspects against a standard rather than the school that delivers it, and the decision the rest of the volume is intended to inform.

2 · Finding one: a single lesson observation scores between 0.14 and 0.37 on reliability #

A single lesson observation scores between 0.14 and 0.37 on reliability. Source: SURE research, Volume 5, August 2026. The figure is published as a range rather than a point, so no midpoint, average or expected value is published and none is calculated here. The full source note is not reproduced on this page: the coefficient type, the base (n) and the period are not stated, and none is supplied. The range is a property of one observed lesson used as a measure of the teaching it is meant to represent. It is not a statement about any named inspectorate, framework or observation protocol, and it is not a figure this site has re-calculated.

Figure 1

The reliability of a single lesson observation is published as a range, not a point: 0.14 at the lower bound and 0.37 at the upper

Lower bound of the published range

0.14

Upper bound of the published range

0.37

01.0

The published reliability of a single lesson observation, shown as its two bounds. Source: SURE research, Volume 5, August 2026. Bars are drawn on a full 0 to 1.0 axis with no truncation and no break in the scale, and both values are printed on their bars, so nothing here depends on colour or on hovering. Only the two published bounds are drawn — no midpoint, average or expected value is shown, because none is published and none has been calculated here. No threshold, benchmark or line of acceptability is drawn on the chart, because this page publishes none. The full source note for this finding is not reproduced on this page: the coefficient type, the base (n) and the period are not stated, and none is supplied in their place.

The published reliability of a single lesson observation, lower and upper bounds, on a 0 to 1.0 axis. Source: SURE research, Volume 5, August 2026.
Value	Reliability of a single lesson observation
Lower bound of the published range	0.14
Upper bound of the published range	0.37
Axis maximum drawn on the chart	1.0
Midpoint or average of the range	Not published, and not calculated here
Coefficient type	Not stated on this page
Threshold or benchmark value	Not published, and none drawn
Base (n) for this figure	Not published on this page
Period for this figure	Not published on this page

This section examines what a coefficient of that size means for a judgement built on one observed lesson, and sets out the difference between an instrument that is inaccurate and one that is unstable.

The analysis then turns to the sources of instability the literature ordinarily separates out — the lesson, the observer and the occasion — and describes how each is handled in the design of an inspection visit or an evaluation cycle.

The chapter closes by stating what the range does not license. It is a statement about a single observation rather than about observation as a practice, and the volume does not extend it into the second.

3 · Finding two: an estimated 63% of one-off lesson judgements are wrong #

An estimated 63% of one-off lesson judgements are wrong. Source: SURE research, Volume 5, August 2026. It is published as an estimate and it is quoted here as one. The full source note is not reproduced on this page, so no base (n), no period and no definition of “wrong” beyond the published wording is stated here, and none is supplied. No confidence interval is published for it, and its complement is not published and has not been calculated on this page. It is an estimate about judgements built from a single observed lesson as a class — not a measurement of any particular district, inspectorate, accreditation scheme or school.

Figure 2

An estimated 63% of one-off lesson judgements are wrong — an estimate about judgements built from a single observed lesson as a class, not about any particular one

One-off lesson judgements that are wrong (estimated)

63%

0%100%

Estimated share of one-off lesson judgements that are wrong. Source: SURE research, Volume 5, August 2026. The bar is drawn on a full 0–100% scale with no truncated axis, and the value is printed on the bar, so nothing here depends on colour or on hovering. The figure is published as an estimate and is drawn as one; no confidence interval is published for it, so none is drawn. The remainder of the track is left blank rather than labelled, because the source publishes the 63% only and this site has not calculated its complement. The estimate concerns one-off lesson judgements as a class, and no direction of error is shown, because none is published: this is a figure for how often such a judgement misses, not for which way it misses. No base (n) and no period are published for this figure on this page, and none is supplied here.

Estimated share of one-off lesson judgements that are wrong. Source: SURE research, Volume 5, August 2026.
Measure	Value
One-off lesson judgements that are wrong, estimated	63%
Axis maximum drawn on the chart	100%
Complement of the estimate	Not published, and not calculated here
Confidence interval	Not published, and none drawn
Direction of error	Not published
Base (n) for this figure	Not published on this page
Period for this figure	Not published on this page

This section sets out what the estimate is an estimate of: the share of judgements formed from a single observed lesson that a fuller measurement of the same teaching would reverse.

The analysis then turns to what follows a judgement once it has been recorded — for the teacher who carries it, for the school that acts on it, and for the aggregate picture a system assembles by adding such judgements together.

The chapter closes by separating an error rate from a bias. The volume argues about how often a one-off judgement misses, not about which direction it misses in, and does not claim the second from the first.

4 · Finding three: 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year #

A tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. Source: SURE research, Volume 5, August 2026. Both halves of that sentence are published as one finding and are quoted together here. The figure is US-derived and describes the average large district, not any named one. The two quantities are counted over different periods — observations per evaluation cycle, lessons per year — and the length of an evaluation cycle is not published with them on this page, so no per-year observation figure and no share-of-lessons-observed percentage is stated anywhere on this page. The 1,000 is published as a round approximation and is not treated here as an exact count. No base (n) and no period are published for this finding on this page, and none is supplied.

Figure 3

A tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year

Lessons taught in a year

~1,000

Formal observations per evaluation cycle

2.4

0~1,000 lessons

Formal observations a tenured teacher receives per evaluation cycle in the average large US district, set against the lessons that teacher teaches in a year. Source: SURE research, Volume 5, August 2026. Both bars are drawn on one shared axis running from 0 to roughly 1,000 lessons, with no truncation and no break in the scale; drawn honestly to that axis the observation bar is a sliver a couple of pixels wide, which is the finding and not a rendering fault. Both values are printed beside their bars, so nothing here depends on colour or on hovering. The two quantities are counted over different periods, observations per evaluation cycle against lessons per year, and the length of an evaluation cycle is not published with this finding on this page — so no per-year observation count and no share-of-lessons-observed percentage is stated here, and none is calculated. The bar is drawn in proportion on the shared axis; the two numbers are printed exactly as published. The figure is US-derived and describes the average large district. No base (n) and no period are published for it on this page.

Formal observations per evaluation cycle against lessons taught in a year, for a tenured teacher in the average large US district. Source: SURE research, Volume 5, August 2026.
Measure	Value
Lessons that teacher teaches in a year	Roughly 1,000, published as an approximation
Formal observations in one evaluation cycle	2.4
Population	A tenured teacher in the average large US district
Axis maximum drawn on the chart	Roughly 1,000 lessons
Length of an evaluation cycle	Not published on this page
Observations expressed per year	Not published, and not calculated here
Share of lessons observed, as a percentage	Not published, and not calculated here
Geography	US-derived
Base (n) and period for this figure	Not published on this page

This section sets the observed sample against the taught population and describes the sampling problem that follows, without extending it into a claim about any particular district, inspectorate or school.

The analysis then turns to what an evaluation cycle is designed to do with a sample of that size, and to the assumptions about consistency a body has to make in order to treat such a sample as standing for a year of teaching.

The chapter closes by joining this finding to the two before it, and describes how the volume will treat the three together without quantifying their combination beyond what the published figures support.

5 · What this means for a system, an inspectorate or a standards body #

The three findings arrive at a system as one problem. A single lesson observation scores between 0.14 and 0.37 on reliability; an estimated 63% of one-off lesson judgements are wrong; and a tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year. Source: SURE research, Volume 5, August 2026, with no base (n) and no period published on this page for any of the three. A body that writes a teaching standard — a curriculum authority, an inspectorate, an accreditation scheme, a ministry, the owner of a framework such as Ofsted, KHDA, CIS, IB, CAEP or World Bank Teach — verifies that standard through that instrument and at that frequency.

This section turns from the evidence to its consequences for a body that owns a standard rather than a classroom, and separates what such a body can change from what it can only observe.

The analysis then sets out the levers ordinarily available to a system — the frequency of inspection, the training of observers, the design of the evaluation cycle — and describes what each can and cannot repair, given the findings stated above.

The chapter closes with what the volume proposes should be measured instead, written as a requirement on any instrument a system might adopt rather than as an endorsement of a particular one.

Methodology and limitations #

Two sources sit behind this page, and it is worth being exact about which claim comes from which. This volume’s own findings: the 0.14–0.37 reliability range, the 63% estimate, and the 2.4 formal observations against roughly 1,000 lessons taught a year are published by this series as SURE research, Volume 5, August 2026. The full source note for each is not reproduced on this page, so no base (n) and no period is shown for any of them, and none is supplied in their place. Our own competency data: 26,544 teachers assessed across 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, with pooled mean competency of 51% — 46.8% in 2022 and 53.4% in 2023. That dataset is not the source of the three findings above; it is described here because its standing limits apply to everything published under this masthead.

Summarised here, in full on the method pages

No base (n) and no period is published on this page for any of the three headline findings. Where we cannot show a base, we say so rather than supply one.
All three findings concern lesson observation as an instrument. None is a statement about any named inspectorate, curriculum body, accreditation scheme, ministry, district or school, and none should be read as one.
The reliability figure is published as a range, not a point. No midpoint, average or expected value has been calculated, the coefficient type is not stated on this page, and no threshold or line of acceptability is published or drawn.
The 63% is published as an estimate and is quoted as one. No confidence interval is published, so none is drawn; its complement is not published and has not been calculated; and no direction of error is published, so the figure describes how often a one-off judgement misses, not which way it misses.
The 2.4 and the roughly 1,000 are counted over different periods — per evaluation cycle and per year — and the length of an evaluation cycle is not published with them here. No per-year observation figure and no share-of-lessons-observed percentage appears anywhere on this page. The finding is US-derived and describes the average large district; the 1,000 is a published approximation, not an exact count.
This site publishes no inter-rater agreement figures of its own, and no error rates beyond the estimate above. Those studies have not been done. When they are, they will appear with their methods attached and their limits stated in the same place as their results.
Ofsted, KHDA, CIS, IB, CAEP and World Bank Teach are named on this page only as examples of bodies and frameworks that own a teaching standard. No finding here is a measurement of any of them.
Participation in our competency assessment was voluntary, so that population skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national workforce, and no figure on this page should be read as a national average.
The assessment measures knowledge-for-practice, not enacted classroom behaviour. A score describes what a teacher knows to do, not what an observer would see them do.
The 2022 and 2023 editions share no common items, so no growth claim can be made between them. The pooled 51% sits between 46.8% and 53.4% because the editions differ, not because teachers improved.
The chapter prose on this page is structural specimen text and carries no findings. It describes what each chapter will argue; it does not make the argument, and it contains no figures of its own.
We have no named school case studies, no product efficacy results and no adoption numbers, and we will not publish any until they exist.
The volumes in this series are published without an individual author byline, so there is no named author to attribute this volume to.

The full method sits on How we research, the dataset is described on The dataset, and the standing methodology note is at Methodology.

References #

SURE (2026). The Standard Nobody Sees. Research volume 5, August 2026 — the source of the three headline findings on this page. The full source note for each finding is not reproduced here.
Our own competency data, 2022 and 2023 editions — described on How we research and The dataset.

If you own a teaching standard that thousands of classrooms are meant to enact, the distance between the standard and its enactment is exactly what our platform is built to make visible — here is how it works.

Read next

Where this evidence goes next#

For systems

The same evidence, arranged around a standards body’s decisions

What a curriculum authority, an inspectorate or a ministry can do with findings about the one instrument its standard is verified through.

Volume 5 · Key findings

The findings of this volume on one page

A single lesson observation scores between 0.14 and 0.37 on reliability, an estimated 63% of one-off lesson judgements are wrong, and a tenured teacher in the average large US district receives 2.4 formal observations per evaluation cycle against roughly 1,000 lessons taught a year.

Volume 6 · August 2026

One Brand, a Thousand Classrooms

38% of the world’s 15,075 international schools now belong to a group (ISC Research, 2026) — the same visibility problem, multiplied by every campus.

Volume 4 · July 2026

Why Training Doesn’t Stick

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002).

Method

How we research

The design, the population and the limits behind 26,544 teachers assessed in 69 countries across the 2022 and 2023 editions.

All research volumes

Cite this

Quoting Volume 5#

Permalink
#/research/the-standard-nobody-sees/
Open this volume at its permanent path. Every chapter heading, figure and headline statistic on this page carries a frozen id, so you can link to a single finding rather than to the whole volume.

Plain citation
SURE (2026). The Standard Nobody Sees. Research volume 5, August 2026.

APA
SURE. (2026, August). The standard nobody sees (Research volume 5).

No individual author byline is assigned to the volumes in this series, so the corporate author is the correct one to cite. This series has no DOI; add the page address from your browser if your style requires a URL. If you are quoting one of the three findings, quote it whole — each is written as one self-contained sentence, and each is stated in that same form everywhere it appears on this site. Quote them with the caveat this page carries: no base (n) and no period is published for any of them here.

Licence
A reuse licence has not been published for this series yet, so we will not imply one. Quote the findings with attribution to SURE and a link back to this page, and ask us first if you need something broader. The findings, figures and citations here are the citable unit of this volume: the chapter prose is specimen text and should not be quoted as though it were the manuscript.

Subscribe

Volume 5’s full manuscript will appear at this address, with the findings exactly where they are now. One email per volume, with the findings and their bases in the email itself, so you can decide from the email whether the volume is worth your time.

Subscribe to the research series

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Research volume · July 2026

Why Training Doesn’t Stick

Training alone produces classroom implementation in roughly 0–5% of participants, and adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002) — the distance between those two figures is not a verdict on the content of the training, it is a description of what happens after the training stops.

Volume 4 · Written for teacher education and PD providers · ~20 min read when the full manuscript publishes · Read free · PDF needs email

The four figures

What this volume argues from#

The first two are the transfer findings this volume is built on. The third is external literature on the same question, carried with its authors’ own caveat attached. The fourth is our own competency data, included because the same argument lands on our instrument as hard as it lands on anyone else’s.

0–5%

of participants reach classroom implementation from training alone

Training alone produces classroom implementation in roughly 0–5% of participants. Source: Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD. External literature, cited as published and not re-analysed by us. The base (n) and the period behind this figure are not reproduced on this page, and none is supplied here in their place.

95%

of participants reach classroom implementation once coaching on their own lessons is added

Adding coaching on teachers’ own lessons raises implementation to approximately 95%. Source: Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD. External literature, cited as published and not re-analysed by us. The base (n) and the period behind this figure are not reproduced on this page, and none is supplied here in their place.

0.49 SD

is the pooled effect of teacher coaching on instruction across 60 causal studies

Across 60 causal studies of teacher coaching, pooled effects were 0.49 standard deviations on instruction and 0.18 standard deviations on achievement. The caveat travels with the figure and we do not separate them: those effects shrink as programmes scale, because coaching is capacity-bound. Source: Kraft, M. A., Blazar, D., & Hogan, D. (2018). Review of Educational Research, 88(4), 547–588. Not a finding of this volume.

51%

is pooled mean competency across 26,544 teachers assessed in 69 countries in the 2022 and 2023 editions

Across 26,544 teachers assessed in 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, pooled mean competency was 51% — 46.8% in 2022 and 53.4% in 2023. Source: SURE research. The assessment measures knowledge-for-practice, not enacted classroom behaviour, which places it on the near side of the gap this volume is about.

Executive summary

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises it to approximately 95% #

Teacher education and professional development are judged, almost everywhere, by what happens in the room where the training is delivered: attendance, engagement, a satisfaction score, a completion certificate, sometimes a post-session test. The finding this volume is built on is about a different room — the participant’s own classroom, on an ordinary Tuesday, several weeks later.

The two figures sit side by side and describe the same participants. Joyce and Showers report that training alone produces classroom implementation in roughly 0–5% of participants, and that adding coaching on teachers’ own lessons raises implementation to approximately 95%. The distance between those two figures is not a verdict on the content of the training. It is a description of what happens after the training stops.

That is a finding about transfer, and transfer is the part of professional learning nobody watches. A provider can evidence design, delivery, hours, coverage, reach and satisfaction, and can evidence them well. What a provider cannot currently evidence is enactment: whether the practice taught on Thursday shows up in a lesson the following month. Which leads somewhere uncomfortable for this sector. A programme that changed nothing and a programme that worked produce the same completion data, the same feedback forms and the same report to the funder. From outside the classroom they are indistinguishable — and the classroom is where the difference between them lives.

The published evidence on coaching points the same way, and carries its own limit. Across 60 causal studies of teacher coaching, pooled effects were 0.49 standard deviations on instruction and 0.18 standard deviations on achievement (Kraft, Blazar & Hogan, 2018). Their caveat is part of their finding and is reproduced with it here: those effects shrink as programmes scale, because coaching is capacity-bound. The thing that makes training transfer is precisely the thing that is hardest to supply to everybody at once.

The same argument lands on our own instrument. Across 26,544 teachers assessed in 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, pooled mean competency was 51% — 46.8% in 2022 and 53.4% in 2023. That assessment measures knowledge-for-practice, not enacted classroom behaviour: it is a measure of what a teacher knows to do. Joyce and Showers is a finding about the gap between knowing and doing, which puts our own headline number on the near side of the gap this volume is about. We would rather write that down than leave you to notice it.

None of this is an argument against training, and none of it is an argument for less of it. The 95% figure is itself a training figure: the coaching that produces it is coaching on the participant’s own lessons, which is more professional learning, not less, and it is more expensive per teacher rather than cheaper. The argument is about where the learning has to land, and about what anyone who designs or funds it can actually see once the session ends.

External literature — not a finding of this volume

Kraft, Blazar and Hogan (2018) reviewed 60 causal studies of teacher coaching and reported pooled effects of 0.49 standard deviations on instruction and 0.18 standard deviations on achievement. The caveat travels with the figure and we will not separate them: those effects shrink as programmes scale, because coaching is capacity-bound.

That is cited here as published literature bearing on the same question, with its own authors’ qualification attached. It is not a result of ours, not a claim about any product, and not one of this volume’s two headline findings. Full reference: Kraft, M. A., Blazar, D., & Hogan, D. (2018). The Effect of Teacher Coaching on Instruction and Achievement: A Meta-Analysis of the Causal Evidence. Review of Educational Research, 88(4), 547–588.

How to read the chapters below. The findings, the figures and the citations on this page are final and citable today. The chapter prose is structural specimen text, marked as such wherever it appears, published in that form so the template can be reviewed before the full manuscript is typeset. Every chapter opens with this volume’s real finding, stated in full; everything marked specimen describes what that chapter will argue rather than arguing it.

The PDF edition

Get Volume 4 as a PDF#

Everything on this page is free to read and stays free. The PDF edition of Why Training Doesn’t Stick publishes with the full manuscript, not before it, and leaving three fields is how you get it on the day it does.

PDF edition

Three fields, and we disclose all three#

Work email, name, role. That is the entire form, and nothing you have just read sits behind it.

A work address reaches you more reliably than a personal one — but either will do, and we will not chase you for the other.

Three fields, disclosed in full: work email, name, role. Front-end demo — this form is not wired to anything yet, so nothing is sent and nothing is stored.

Specimen edition. The findings, figures and citations on this page are final. The chapter prose marked below is structural specimen text, shown so the template can be reviewed before the full manuscript is typeset.

1 · The question — what happens between a training day and a Tuesday lesson? #

Two published figures set the terms of the question. Training alone produces classroom implementation in roughly 0–5% of participants, and adding coaching on teachers’ own lessons raises implementation to approximately 95% — both from Joyce, B., & Showers, B. (2002), Student Achievement Through Staff Development, 3rd ed. ASCD. They are that work’s findings about the transfer of training into classroom practice, cited here with attribution. The base (n) and the period behind each figure are not reproduced on this page, and neither is supplied here in their place.

This section sets out the terms of the question before any answer is attempted. It distinguishes the two rooms a training programme is judged in — the room where the session is delivered, and the classroom the participant returns to — and establishes why evidence gathered in the first is routinely read as evidence about the second.

The analysis then turns to the interval between them: the weeks in which a taught practice either becomes part of how a teacher runs a lesson or quietly does not. It names that interval as the object of the volume, and distinguishes it from the questions about programme quality with which it is usually confused.

The chapter closes by naming the reader it is written for — the institution preparing teachers, and the provider accountable for what participants do months after they leave — and the decision the rest of the volume is intended to inform.

2 · Finding one: training alone produces classroom implementation in roughly 0–5% of participants #

Training alone produces classroom implementation in roughly 0–5% of participants (Joyce & Showers, 2002). The figure describes implementation in the participant’s own classroom, not satisfaction with the training, not knowledge gained during it, and not intention to use it afterwards. It is published as a range rather than a point, and it is reproduced here as a range. The base (n) and the period behind it are not reproduced on this page, and none is supplied here in their place.

This section examines what the lower figure can and cannot support. It first establishes the distinction the figure rests on — between what a participant knows at the end of a session and what an observer would later see in that participant’s classroom — and why the two are routinely reported as though they were one measure.

The analysis then turns to the instruments a provider ordinarily holds at the point the session ends, and to which of them are capable of speaking to implementation at all.

The chapter ends by stating the limits of the finding in the form in which the volume will carry them, so that no later section rests on a reading the figure does not bear.

3 · Finding two: coaching on teachers’ own lessons raises implementation to approximately 95% #

Adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002). The condition being described is not more training and not better training: it is follow-up on the participant’s own lessons after the session has ended. As with the lower figure, the base (n) and the period are not reproduced on this page, and none is supplied here in their place.

Figure 1

Training alone reaches roughly 0–5% of participants in their own classrooms; adding coaching on their own lessons reaches approximately 95%

Training alone, upper bound

Training alone

0–5%

Training plus coaching on teachers’ own lessons

About 95%

0%100%

Share of participants reaching classroom implementation, by training condition. Source: Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD. External literature, cited as published; we have not re-run the study and have not re-analysed its data. Bars are drawn on a full 0–100% scale with no truncated axis, and both values are printed beside or on their bars, so nothing here depends on colour or on hovering. The training-alone bar is drawn at 5%, the upper bound of the published range, so the chart cannot overstate the distance between the two conditions. No base (n) and no period are published for either figure on this page, and none is drawn here in their place.

Share of participants reaching classroom implementation, by training condition. Source: Joyce, B., & Showers, B. (2002). *Student Achievement Through Staff Development*, 3rd ed. ASCD.
Condition	Participants reaching classroom implementation
Training alone	Roughly 0–5%
Training alone, as drawn on the chart	5% (the upper bound of the published range)
Training plus coaching on teachers’ own lessons	Approximately 95%
Axis maximum drawn on the chart	100%
Base (n) for either figure	Not reproduced on this page
Period for either figure	Not reproduced on this page

This section examines what changes between the two conditions, and what does not. It establishes that the variable being altered sits after the session rather than inside it, and sets out why that placement matters to anyone designing a programme against a fixed budget of contact hours.

The analysis then turns to the cost structure implied by the higher figure: follow-up on a participant’s own lessons is more professional learning rather than less, and is more expensive per teacher rather than cheaper. The chapter is explicit that the finding is not an argument for reducing training.

The chapter closes on the constraint the rest of the volume depends on — that the condition producing the higher figure is bounded by the number of people available to observe lessons — and marks it before any recommendation is drawn.

4 · Why the proof stopped at the reaction form — and what that costs a provider whose programme actually worked #

Across 60 causal studies of teacher coaching, pooled effects were 0.49 standard deviations on instruction and 0.18 standard deviations on achievement (Kraft, Blazar & Hogan, 2018), and the authors’ caveat is reproduced with the figure rather than after it: those effects shrink as programmes scale, because coaching is capacity-bound. That is external literature bearing on the same question, cited as published; it is not a result of ours and not one of this volume’s two headline findings.

Figure 2

Pooled coaching effects are larger on what a teacher does in a lesson than on what students go on to score — 0.49 SD against 0.18 SD across 60 causal studies

Pooled effect on instruction

0.49 SD

Pooled effect on achievement

0.18 SD

0 SD0.50 SD

Pooled effect sizes from a meta-analysis of the causal evidence on teacher coaching, reported in standard deviations. Source: Kraft, M. A., Blazar, D., & Hogan, D. (2018). The Effect of Teacher Coaching on Instruction and Achievement: A Meta-Analysis of the Causal Evidence. Review of Educational Research, 88(4), 547–588 — 60 causal studies. Bars are drawn on a scale running from 0 to 0.50 standard deviations and every value is printed on its bar, so nothing here depends on colour or on hovering. The authors’ caveat is part of the finding and is carried with it: these effects shrink as programmes scale, because coaching is capacity-bound. External literature, cited as published; the study populations are not ours and we did not re-analyse the data. No effect size on this chart is a claim about any product, and none is a student-achievement percentage.

Pooled effect sizes on instruction and on achievement from a meta-analysis of 60 causal studies of teacher coaching. Source: Kraft, M. A., Blazar, D., & Hogan, D. (2018), *Review of Educational Research*, 88(4), 547–588.
Outcome	Pooled effect (standard deviations)
Instruction	0.49 SD
Achievement	0.18 SD
Number of causal studies pooled	60
Axis maximum drawn on the chart	0.50 SD
Caveat published with the finding	Effects shrink as programmes scale, because coaching is capacity-bound

This section sets out why the evidence a provider holds ordinarily stops at the end of the session. It describes the instruments that are available at that point, what each of them is genuinely good at, and the reason none of them reaches the classroom the training was meant to change.

The analysis then turns to the consequence for a provider whose programme did work: with the evidence stopping where it does, a programme that changed practice and a programme that did not are reported to a funder in the same terms. The chapter treats that symmetry as the commercial problem it is, and states who bears its cost.

The chapter closes by distinguishing the two claims that are easily conflated here — that the proof is missing, and that the programme failed — and is explicit that the evidence supports only the first.

5 · What this means for an institution preparing teachers, and for a PD provider #

Across 26,544 teachers assessed in 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, pooled mean competency was 51% — 46.8% in the 2022 edition and 53.4% in the 2023 edition. The two editions share no common items, so no growth claim can be made between them, and this site does not make one. That assessment measures knowledge-for-practice, not enacted classroom behaviour, which places our own headline number on the near side of the gap this volume is about.

This section translates the volume’s findings into the decisions the two audiences actually make: how an institution preparing teachers designs the final year and the first year that follows it, and how a provider structures what happens after a session ends.

The analysis then turns to what the evidence does not license — the promises neither audience should make on the strength of figures of this kind — and marks those limits before any practical suggestion is offered.

The chapter closes by separating the recommendations that rest on the transfer findings from those that rest on our own competency data, and states plainly which of the two is doing the work in each case.

Methodology and limitations #

This volume draws on two kinds of source, and it is worth being exact about which claim comes from which. External literature: the two transfer figures — roughly 0–5% of participants from training alone, approximately 95% once coaching on teachers’ own lessons is added — are reported from Joyce, B., & Showers, B. (2002), Student Achievement Through Staff Development, 3rd ed. ASCD; the pooled coaching effects of 0.49 standard deviations on instruction and 0.18 standard deviations on achievement across 60 causal studies are reported from Kraft, Blazar and Hogan (2018), with their scaling caveat attached. Our own competency data: 26,544 teachers assessed across 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, with pooled mean competency of 51% — 46.8% in 2022 and 53.4% in 2023.

Summarised here, in full on the method pages

Both transfer figures are Joyce and Showers’, not ours. We have not re-run the study, and nothing on this page should be read as a replication of it, an extension of it, or a measurement made by us.
No base (n) and no period are reproduced on this page for either transfer figure. Where we cannot show a base, we say so rather than supply one.
The lower figure is published as a range and the higher one as an approximation. Neither is a measurement of any particular programme, provider or institution, and neither names a cause.
The coaching meta-analysis is cited as published, and its caveat is part of the finding: the effects shrink as programmes scale, because coaching is capacity-bound. The study populations behind it are not ours and we did not re-analyse the data.
Participation in our competency assessment was voluntary, so that population skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national workforce, and no figure on this page should be read as a national average.
The assessment measures knowledge-for-practice, not enacted classroom behaviour — a limit this volume in particular has to state twice, because it is also this volume’s subject. A score describes what a teacher knows to do, not what an observer would see them do.
The 2022 and 2023 editions share no common items, so no growth claim can be made between them. The pooled 51% sits between 46.8% and 53.4% because the editions differ, not because teachers improved.
The chapter prose on this page is structural specimen text and carries no findings. It describes what each chapter will argue; it does not make the argument, and it contains no figures of its own.
We have no named school case studies, no product efficacy results and no adoption numbers, and we will not publish any until they exist.
The volumes in this series are published without an individual author byline, so there is no named author to attribute this volume to.

The full method sits on How we research, the dataset is described on The dataset, and the standing methodology note is at Methodology.

References #

Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD.
Kraft, M. A., Blazar, D., & Hogan, D. (2018). The Effect of Teacher Coaching on Instruction and Achievement: A Meta-Analysis of the Causal Evidence. Review of Educational Research, 88(4), 547–588.
Our own competency data, 2022 and 2023 editions — described on How we research and The dataset.

If you design teacher education or professional development, the evidence that your programme changed a classroom is currently held in classrooms nobody watches, and making what happens after the training visible is what our platform is built to do — here is how it works.

Read next

Where this evidence goes next#

For teacher education

The same evidence, arranged around a provider’s decisions

What this research means when you are accountable for what participants do months after they leave your programme — and what can be shown about the part that happens in their own classrooms.

Volume 4 · Key findings

The findings of this volume on one page

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002).

Volume 5 · August 2026

The Standard Nobody Sees

A single lesson observation scores between 0.14 and 0.37 on reliability, and an estimated 63% of one-off lesson judgements are wrong — the instrument most often used to check whether training landed.

Volume 2 · July 2026

The Gap Experience Doesn’t Close

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023.

Volume 6 · August 2026

One Brand, a Thousand Classrooms

38% of the world’s 15,075 international schools now belong to a group — where group professional development calendars meet the same transfer problem at network scale.

All research volumes

Cite this

Quoting Volume 4#

Permalink
#/research/why-training-doesnt-stick/
Open this volume at its permanent path. Every chapter heading, figure and headline statistic on this page carries a frozen id, so you can link to a single finding rather than to the whole volume. The findings alone sit at key findings.

Plain citation
SURE (2026). Why Training Doesn’t Stick. Research volume 4, July 2026.

APA
SURE. (2026, July). Why training doesn’t stick (Research volume 4).

No individual author byline is assigned to the volumes in this series, so the corporate author is the correct one to cite. This series has no DOI; add the page address from your browser if your style requires a URL. The transfer findings on this page are not ours to be credited with: cite them to Joyce, B., & Showers, B. (2002), Student Achievement Through Staff Development, 3rd ed. ASCD, and cite this page only for the reporting of them. The coaching effect sizes belong to Kraft, Blazar and Hogan (2018) and should be quoted with their scaling caveat attached.

Licence
A reuse licence has not been published for this series yet, so we will not imply one. Quote the findings with attribution to SURE and a link back to this page, and ask us first if you need something broader. The findings, figures and citations here are the citable unit of this volume: the chapter prose is specimen text and should not be quoted as though it were the manuscript.

Subscribe

Volume 4’s full manuscript will appear at this address, with the findings exactly where they are now. One email per volume, with the findings and their bases in the email itself, so you can decide from the email whether the volume is worth your time.

Subscribe to the research series

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Research volume · July 2026

Why Teachers Stop Showing Up

About 70% of teachers in their first five years have left the classroom or seriously considered it — and replacing one teacher costs $11,900–$24,900, a US-derived range published by Stronge & Associates / TIE Online.

Volume 3 · Written for schools · ~22 min read when the full manuscript publishes · Read free · PDF needs email

The four figures

What this volume argues from#

The first two are this volume’s own headline findings. The second two are cited from elsewhere in this series and carry their own sources, because the compounding cost this volume examines is not something Volume 3 measured on its own.

70%

About 70% of teachers in their first five years have left the classroom or seriously considered it

SURE research, Volume 3, July 2026. This finding combines two outcomes in one figure — having left, and having seriously considered leaving — so it is not a departure rate and is not used as one here. No base and no period are published for it on this page; where we cannot show a base, we say so rather than supply one.

$11,900–$24,900

is what replacing one teacher costs

Source: Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013). A US-derived figure, published as a range. No currency base year is attached to it as reproduced here, so none is stated. Read against a school outside the United States it is an order of magnitude, not a quotation.

r = 0.03

is the correlation between years of teaching experience and classroom competency

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023. Source: SURE research, Volume 2, July 2026. Cited here, not re-measured by this volume.

0–5%

of participants reach classroom implementation from training alone

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises implementation to approximately 95%. Source: Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD. Carried in this series by Volume 4; external literature, cited as published, not re-analysed by us.

Executive summary

About 70% of teachers in their first five years have left the classroom or seriously considered it, and replacing one teacher costs $11,900–$24,900 #

This volume asks what a school loses when a teacher leaves, and it argues from two published figures rather than from a theory of why teachers go. The two figures do different jobs: the first marks out a population, the second attaches a price to what happens when that population decides.

The population is the newest part of the staffroom. About 70% of teachers in their first five years have left the classroom or seriously considered it. That single figure combines two outcomes — having left, and having seriously considered leaving — so it is not a departure rate, and this site does not quote it as one. No base and no period are published for it on this page. Where we cannot show a base, we say so rather than supply one.

The price is paid per departure, not per year. Replacing one teacher costs $11,900–$24,900, a range published by Stronge & Associates and TIE Online, citing Mancuso, Roberts and White (2010) and Desroches (2013). It is a US-derived figure with no currency base year attached to it as reproduced here. Because the cost falls on each departure rather than on each year, a school can hold its headcount flat and pay it repeatedly: headcount is a net figure, replacement is a gross one.

The compounding cost is the part neither figure prices. Two findings from elsewhere in this series bear on what a replacement does and does not bring. First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023 (Volume 2). And training alone produces classroom implementation in roughly 0–5% of participants, while adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002, carried in Volume 4). Both are cited as published; neither was measured by this volume.

What this volume does not claim. Neither headline finding names a cause. Neither is a measurement of any particular school, group or country. A volume titled Why Teachers Stop Showing Up owes a why, and the why belongs in chapters with evidence attached to each step of the argument — not in a plausible-sounding explanation written to fill the space.

How to read the chapters below. The findings, the figures and the citations on this page are final and citable today. The chapter prose is structural specimen text, marked as such wherever it appears, published in that form so the template can be reviewed before the full manuscript is typeset. Every chapter opens with this volume’s real finding, stated in full; everything marked specimen describes what that chapter will argue rather than arguing it.

The PDF edition

Get Volume 3 as a PDF#

Everything on this page is free to read and stays free. The PDF edition of Why Teachers Stop Showing Up publishes with the full manuscript, not before it, and leaving three fields is how you get it on the day it does.

PDF edition

Three fields, and we disclose all three#

Work email, name, role. That is the entire form, and nothing you have just read sits behind it.

A work address reaches you more reliably than a personal one — but either will do, and we will not chase you for the other.

Three fields, disclosed in full: work email, name, role. Front-end demo — this form is not wired to anything yet, so nothing is sent and nothing is stored.

Specimen edition. The findings, figures and citations on this page are final. The chapter prose marked below is structural specimen text, shown so the template can be reviewed before the full manuscript is typeset.

1 · The question — what does a school lose when a teacher leaves? #

Two published figures set the terms of the question. About 70% of teachers in their first five years have left the classroom or seriously considered it, and replacing one teacher costs $11,900–$24,900 — a US-derived range published by Stronge & Associates / TIE Online, citing Mancuso, Roberts and White (2010) and Desroches (2013). No base, no period and no currency base year are published for either figure on this page, and none is supplied here in their place.

This section sets out the terms of the question before any answer is attempted. It distinguishes the two things a school is actually counting when a teacher resigns: the people who leave, and the transaction that follows.

The analysis then separates the accounting question from the causal one, and states plainly which of the two this volume is equipped to address with the evidence it holds.

The chapter closes by naming the reader it is written for — the school leader who has to plan a staffroom for the year after next — and the decision the rest of the volume is intended to inform.

2 · Finding one: about 70% of teachers in their first five years have left the classroom or seriously considered it #

About 70% of teachers in their first five years have left the classroom or seriously considered it. The figure combines two outcomes in one number — having left, and having seriously considered leaving — so it is not a departure rate, and it is not quoted as one anywhere on this site. No base and no period are published for it on this page; the two components are not published separately either, and neither is estimated here.

Figure 1

About 70% of teachers in their first five years have left the classroom or seriously considered it — roughly seven in ten of the newest people in a staffroom

Teachers in their first five years who have left the classroom or seriously considered it

About 70%

0%100%

Share of teachers in their first five years who have left the classroom or seriously considered it. Source: SURE research, Volume 3, July 2026. The bar is drawn on a full 0–100% scale with no truncated axis, and the value is printed on the bar, so nothing here depends on colour or on hovering. The figure is a combined one: leaving and seriously considering leaving are not published separately, so the chart shows the combined share only and no split is drawn. No base (n) and no period are published for this figure on this page, and none is drawn here in their place.

Share of teachers in their first five years who have left the classroom or seriously considered it. Source: SURE research, Volume 3, July 2026.
Group	Value
Teachers in their first five years who have left the classroom or seriously considered it	About 70%
Axis maximum drawn on the chart	100%
Teachers who have left, shown separately	Not published
Teachers who have seriously considered leaving, shown separately	Not published
Base (n) for this figure	Not published on this page
Period for this figure	Not published on this page

This section examines what the combined figure can and cannot support. It first establishes why a measure that folds departure together with seriously considering departure behaves differently from a turnover statistic, and what follows from that for a school reading it.

The analysis then turns to the population the figure describes — teachers in their first five years — and to why that group is the one a school has recruited most recently and needs for longest.

The chapter ends by stating the limits of the finding in the form in which the volume will carry them, so that no later section rests on a reading the figure does not bear.

3 · Finding two: replacing one teacher costs $11,900–$24,900 #

Replacing one teacher costs $11,900–$24,900. The range is US-derived and is reproduced here in the currency and the form its source publishes: Stronge & Associates / TIE Online, citing Mancuso, Roberts and White (2010) and Desroches (2013). It is not a global average, it has not been re-calculated by this site, it is not adjusted for any other market, and no currency base year is attached to it as reproduced here.

Figure 2

The published cost of replacing one teacher is a range, not a point: $11,900 at the lower bound and $24,900 at the upper

Lower bound of the published range

$11,900

Upper bound of the published range

$24,900

$0$24,900

The published cost of replacing one teacher, shown as its two bounds. Source: Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013). Bars are drawn on a scale running from $0 to $24,900, so the upper bound fills the track and the lower bound is drawn in proportion to it; both values are printed on their bars, so nothing here depends on colour or on hovering. Only the two published bounds are drawn — no midpoint, average or expected value is shown, because the source publishes none and this site has not calculated one. The range is US-derived and carries no currency base year as reproduced here.

The published cost of replacing one teacher, lower and upper bounds. A US-derived range. Source: Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013).
Value	Cost of replacing one teacher
Lower bound of the published range	$11,900
Upper bound of the published range	$24,900
Axis maximum drawn on the chart	$24,900
Midpoint or average of the range	Not published, and not calculated here
Currency base year	Not stated by the source as reproduced here
Geography	US-derived

This section sets out how a published cost range should and should not be used by a school reading it from outside the market it was derived in. It establishes the difference between quoting a figure and taking it as an order of magnitude.

The analysis then turns to the structure of the cost itself — what a replacement transaction consists of, and why the same event can sit at either end of a range this wide.

The chapter closes on the accounting point the rest of the volume depends on: that this cost attaches to each departure rather than to each year, and that a stable headcount therefore conceals rather than prevents it.

4 · The compounding cost — what a departure takes with it that a replacement does not bring #

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023. That finding belongs to Volume 2 and is cited here, not re-measured. Alongside it, training alone produces classroom implementation in roughly 0–5% of participants, while adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002), the transfer evidence carried in this series by Volume 4.

This section reads those two cited findings against the replacement transaction priced in the previous chapter. It sets out what each of them constrains about the school’s options once a departure has happened.

The analysis then turns to the difference between a cost that is paid once and a cost that recurs, and to the conditions under which the second is the more accurate description of what a school is carrying.

The chapter is explicit throughout about the boundary between the two cited findings and this volume’s own: nothing in it is offered as a measurement this volume performed.

5 · What this means for a school #

The research programme behind this series assessed 26,544 teachers across 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, and pooled mean competency across it is 51% — 46.8% in the 2022 edition and 53.4% in the 2023 edition. The two editions share no common items, so no growth claim can be made between them, and this site does not make one.

This section translates the volume’s findings into the decisions a school actually makes: how a staffroom is planned, how the newest teachers are supported, and where a leader’s attention is worth spending.

The analysis then turns to what the evidence does not license — the recommendations a school should refuse to draw from figures of this kind — and marks those limits before any practical suggestion is made.

The chapter closes by stating which of its recommendations rest on this volume’s own findings and which rest on evidence cited from elsewhere in the series.

Methodology and limitations #

This volume draws on three sources, and it is worth being exact about which claim comes from which. This volume’s own findings: the early-career figure of about 70% and the replacement-cost range of $11,900–$24,900, the latter sourced to Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013). Our own competency data: 26,544 teachers assessed across 69 countries and 10,562 schools, producing more than 1.2 million scored responses in the 2022 and 2023 editions, with pooled mean competency of 51% — 46.8% in 2022 and 53.4% in 2023 — and the experience comparison of 49.5% against 51.4% at r = 0.03 on a base of 20,429 scored attempts with experience data, published in Volume 2. External literature: the transfer evidence is cited as published from Joyce and Showers (2002).

Summarised here, in full on the method pages

The early-career figure combines having left the classroom with having seriously considered leaving. It is not a departure rate and must not be quoted as one, and the two components are not published separately.
No base (n) and no period are published on this page for the early-career figure. Where we cannot show a base, we say so rather than supply one.
The replacement-cost range is US-derived, published as a range, with no currency base year attached to it as reproduced here. It is not a global average, it is not adjusted for any other market, and this site has not re-calculated it. Read against a school outside the United States it is an order of magnitude, not a quotation.
Neither headline finding names a cause, and neither is a measurement of any particular school, group or country.
The experience comparison and the transfer evidence are cited from Volume 2 and from Joyce and Showers (2002) respectively. They are not findings this volume measured, and the populations behind them are not the population behind the early-career figure.
Participation in our competency assessment was voluntary, so that population skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national workforce, and no figure on this page should be read as a national average.
The assessment measures knowledge-for-practice, not enacted classroom behaviour. A score describes what a teacher knows to do, not what an observer would see them do.
The 2022 and 2023 editions share no common items, so no growth claim can be made between them. The pooled 51% sits between 46.8% and 53.4% because the editions differ, not because teachers improved.
The chapter prose on this page is structural specimen text and carries no findings. It describes what each chapter will argue; it does not make the argument, and it contains no figures of its own.
We have no named school case studies, no product efficacy results and no adoption numbers, and we will not publish any until they exist.
The volumes in this series are published without an individual author byline, so there is no named author to attribute this volume to.

The full method sits on How we research, the dataset is described on The dataset, and the standing methodology note is at Methodology.

References #

Desroches, S. (2013). Cited in Stronge & Associates / TIE Online, as the published source of the teacher replacement-cost range.
Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD.
Mancuso, S. V., Roberts, L., & White, G. P. (2010). Cited in Stronge & Associates / TIE Online, as the published source of the teacher replacement-cost range.
Stronge & Associates / TIE Online — the published source of the $11,900–$24,900 replacement-cost range.
SURE (2026). The Gap Experience Doesn’t Close. Research volume 2, July 2026 — read it here.
Our own competency data, 2022 and 2023 editions — described on How we research and The dataset.

If you lead a school, the development a teacher actually receives in those first five years is the part of this you can act on week by week, and that is the part our platform is built to change — here is how it works.

Read next

Where this evidence goes next#

For schools

The same evidence, arranged around a school’s decisions

What a head of school can do with findings about the teachers who are newest to the staffroom and costliest to replace.

Volume 3 · Key findings

The findings of this volume on one page

About 70% of teachers in their first five years have left the classroom or seriously considered it, and replacing one teacher costs $11,900–$24,900.

Volume 2 · July 2026

The Gap Experience Doesn’t Close

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023.

Volume 4 · July 2026

Why Training Doesn’t Stick

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002).

Volume 5 · August 2026

The Standard Nobody Sees

A single lesson observation scores between 0.14 and 0.37 on reliability — the instrument most schools lean on hardest when they judge how a new teacher is doing.

All research volumes

Cite this

Quoting Volume 3#

Permalink
#/research/why-teachers-stop-showing-up/
Open this volume at its permanent path. Every chapter heading, figure and headline statistic on this page carries a frozen id, so you can link to a single finding rather than to the whole volume.

Plain citation
SURE (2026). Why Teachers Stop Showing Up. Research volume 3, July 2026.

APA
SURE. (2026, July). Why teachers stop showing up (Research volume 3).

No individual author byline is assigned to the volumes in this series, so the corporate author is the correct one to cite. This series has no DOI; add the page address from your browser if your style requires a URL. If you are quoting the replacement-cost range, quote it to its own source — Stronge & Associates / TIE Online, citing Mancuso, Roberts and White (2010) and Desroches (2013) — and not to us. If you are quoting the experience comparison, quote it to Volume 2 with its base of 20,429 scored attempts.

Licence
A reuse licence has not been published for this series yet, so we will not imply one. Quote the findings with attribution to SURE and a link back to this page, and ask us first if you need something broader. The findings, figures and citations here are the citable unit of this volume: the chapter prose is specimen text and should not be quoted as though it were the manuscript.

Subscribe

Volume 3’s full manuscript will appear at this address, with the findings exactly where they are now. One email per volume, with the findings and their bases in the email itself, so you can decide from the email whether the volume is worth your time.

Subscribe to the research series

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

All research volumes

Research volume 02 · July 2026

The Gap Experience Doesn’t Close

First-year teachers scored 49.5% on classroom competency. Teachers with ten or more years scored 51.4%. Across 20,429 scored attempts with experience data, pooled 2022–2023, the correlation between years in the classroom and competency score is r = 0.03 — which means knowing how long someone has taught tells you very close to nothing about this score.

For heads of school, department leads and school leadership teams Volume 2 of the SURE research series Planned length of the full edition: about 24 minutes Read free · PDF needs email

The four findings

What this volume found#

Each of the four is written as one self-contained sentence carrying its own metric, population, base and period, and each is stated in that same form everywhere it appears on this site. Where a base is not available to us, the card says so rather than supplying one.

r = 0.03

First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03.

Base: 20,429 scored attempts with experience data, pooled 2022–2023. SURE research, Volume 2, July 2026.

35 points

The median teacher carries a 35-percentage-point spread between their strongest and weakest domain.

SURE research, Volume 2, July 2026. The base and period behind this figure are not reproduced on this page.

24%

24% of teachers are simultaneously below 40% in one domain and above 70% in another.

SURE research, Volume 2, July 2026. The base and period behind this figure are not reproduced on this page.

51%

Across 26,544 teachers assessed in 69 countries in the 2022 and 2023 editions, pooled mean competency was 51%. By edition, mean competency was 46.8% in 2022 and 53.4% in 2023.

The two editions share no common items, so the distance between 46.8% and 53.4% is not growth and is not reported here as growth. SURE research programme, 2022 and 2023 editions.

Executive summary

The proxy every school reaches for carries almost none of the signal, and the variation it hides sits inside each teacher#

Almost every routine decision a school makes about a teacher runs on years of service. Who mentors whom. Who gets the difficult class and who gets the difficult parent. Who is trusted to run the department, who is put on a support plan, who is left alone because they have been doing this since before the head arrived. Experience is the cheapest proxy a school has for capability, and it is on the file already.

This volume tests that proxy against a score. Across 20,429 scored attempts with experience data, pooled across the 2022 and 2023 editions, first-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4%. The correlation between years of experience and competency score across that base is r = 0.03. A correlation of that size is, for practical purposes, the absence of a relationship: told only how long a teacher has taught, you would predict their score no better than by guessing.

What that finding is not. It is not a claim that experience is worthless, and we will not let it be read as one. This assessment measures knowledge-for-practice — what a teacher knows to do — and not enacted classroom behaviour. A teacher of fifteen years carries judgement, relationships, timing and a memory of what has already failed in this school, and none of that is in these items. What the finding says is narrower and harder to dismiss: on the part of teaching this instrument can see, time in the classroom does not accumulate on its own. Whatever a decade builds, it is not showing up here.

It should also be read against the programme it comes from. Across 26,544 teachers assessed in 69 countries in the 2022 and 2023 editions, pooled mean competency was 51%. Both experience groups sit near that pooled mean, at 49.5% and 51.4%. The flatness is the finding; the level is the context for it.

The gap is not between cohorts. It is inside each teacher.#

If experience does not separate teachers, something else has to account for the variation, and it does. The median teacher carries a 35-percentage-point spread between their strongest and weakest domain. 24% of teachers are simultaneously below 40% in one domain and above 70% in another — the same teacher, at the same moment, genuinely strong at one part of the job and genuinely weak at another.

That is a different shape of problem from the one most staff development is built for. A programme aimed at a cohort — this is what our early-career teachers need, this is what our experienced staff need — is aimed at a variable that explains almost nothing, and it hands the same content to a teacher who is at 70% in that domain and a teacher who is at 40% in it. The average in the staffroom is real; the average teacher is not.

The weakness also has a shape worth knowing. In the 2023 edition, the profession’s lowest correct-response rates fall in the parts of practice an ordinary lesson actually runs on: learner behaviour management at 39.9% (n = 15,047), curriculum planning and progression at 42.5% (n = 30,165) and use of assessment data in achieving learner outcomes at 45.1% (n = 30,205), from 754,971 attempted responses by 15,548 teachers across 25 sub-domains.

The pooled mean itself conceals something. Mean competency was 46.8% in the 2022 edition and 53.4% in the 2023 edition. The two editions share no common items, so the distance between them is not growth, not decline and not change over time — and the pooled 51% is therefore an average of two different measurements rather than a level any cohort reached. The experience comparison is reported pooled across both editions for the same reason: it is a comparison between groups within a base, not a comparison between years.

For a school, the practical consequence is a targeting problem rather than a training problem. You cannot aim at a 35-point spread you cannot see, and the field on the personnel file that everyone reaches for — years of service — is the field that turns out to carry r = 0.03 of the signal.

The chapters below take the argument in order: the question a school is really asking when it reaches for experience; the flat comparison itself; the jagged within-teacher shape that replaces it; what the pooled mean conceals once it is split by edition; and what a school can do about a proxy that carries almost none of the signal. The findings, figures and citations in those chapters are final. The connecting prose is marked as specimen text, and is shown in that form so the page can be reviewed before the full manuscript is typeset.

The PDF edition

Get Volume 2 as a PDF#

The page you are reading stays free, with or without the form. Nothing above this point is behind it, and nothing below it is either.

PDF edition

Three fields, and we disclose all three#

The PDF carries the same four findings and the same figures with the full manuscript typeset around them, in a form you can circulate to a leadership team or a governing board. Leave a work email, a name and a role, and it goes to you when it is ready to send. Those three fields are the whole form.

A work address reaches you more reliably than a personal one — but either will do, and we will not chase you for the other.

Three fields, disclosed in full: work email, name, role. Front-end demo — this form is not wired to anything yet, so nothing is sent and nothing is stored.

Specimen edition. The findings, figures and citations on this page are final. The chapter prose marked below is structural specimen text, shown so the template can be reviewed before the full manuscript is typeset.

1 · The question — can experience be relied on to develop teaching practice? #

Schools treat time in the classroom as the thing that turns a new teacher into a good one, so this volume tests that assumption against a score: across 20,429 scored attempts with experience data, pooled across the 2022 and 2023 editions, first-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4%, a correlation between years in the classroom and competency score of r = 0.03.

This section sets out the question the volume is built to answer, and the reason a school should care what the answer is: whether time in the classroom can be relied on, by itself, to develop teaching practice.

The chapter then locates the assumption inside ordinary school decisions — who mentors, who is trusted with the difficult class, who is left alone, who is put on a support plan — and sets out why a proxy that costs nothing to apply is the hardest kind for an institution to retire.

It closes by fixing the boundary of the test that follows: what the assessment behind this volume can see, what it cannot see, and what a finding drawn from it is therefore entitled to claim.

How this volume relates to the series #

This volume opens the series; there is no Volume 1. Volume 3 follows the teachers who leave, of whom about 70% in their first five years have left the classroom or seriously considered it, at a replacement cost of $11,900–$24,900 per teacher (Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013)). Volume 4 asks why the standard remedy does not transfer: training alone produces classroom implementation in roughly 0–5% of participants, and adding coaching on teachers’ own lessons raises it to approximately 95% (Joyce & Showers, 2002). Volume 5 examines the instrument schools use to look for the gap this volume describes, where a single lesson observation scores between 0.14 and 0.37 on reliability. Volume 6 reads the same problem at network scale, where 38% of the world’s 15,075 international schools now belong to a group (ISC Research, 2026).

2 · Finding one: a decade of classroom experience moves competency under two points #

Across 20,429 scored attempts with experience data, pooled across the 2022 and 2023 editions, first-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a distance of under two percentage points across a decade of practice, and a correlation between years in the classroom and competency score of r = 0.03.

Figure 1

Ten or more years in the classroom leaves competency almost where it started — 49.5% in the first year, 51.4% at ten years or more, a correlation of r = 0.03#

First-year teachers

49.5%

Ten or more years’ experience

51.4%

Mean classroom competency score by years of teaching experience, pooled across the 2022 and 2023 editions. Base: 20,429 scored attempts with experience data. The correlation between years of experience and competency score across that base is r = 0.03. Source: SURE research, Volume 2, July 2026. Both bars are drawn on a full 0–100% scale with no truncated axis, and every value is printed on its own bar, so nothing here depends on colour or on hovering. No mastery threshold is drawn on this chart, because this research programme does not publish one and we will not imply one exists.

Mean classroom competency score by years of teaching experience, pooled 2022 and 2023 editions. Base: 20,429 scored attempts with experience data. Correlation between years of experience and competency score: r = 0.03. Source: SURE research, Volume 2, July 2026.
Group	Mean classroom competency score
First-year teachers	49.5%
Teachers with ten or more years of teaching experience	51.4%
Pooled mean competency across the full programme, for context (26,544 teachers, 69 countries, 2022 and 2023 editions)	51%

Where these numbers come from

This section presents the comparison itself and the base it rests on, then works through what a correlation of this size does and does not entitle a reader to conclude.

The analysis then turns to the objections a school leader raises first — that the two groups are not comparable, that the instrument is the wrong one, that a real effect is being flattened by the pooling — and answers each of them against the base named in the source line above.

The chapter ends by holding apart the two claims that are easiest to collapse into one: that experience does not register on this measure, and that experience does not matter.

3 · Finding two: the gap is jagged, not general #

If years of service do not separate teachers, the variation has to sit somewhere, and it sits inside the individual: the median teacher carries a 35-percentage-point spread between their strongest and weakest domain, and 24% of teachers are simultaneously below 40% in one domain and above 70% in another — the same teacher, at the same moment, genuinely strong at one part of the job and genuinely weak at another. Both figures are from SURE research, Volume 2, July 2026; the base and period behind them are not reproduced on this page, and this page will not supply one it does not have.

Figure 2

Competency is uneven by area rather than uniformly low: across the 25 sub-domains measured in the 2023 edition, correct-response rates run from 39.9% to 71.9%#

Learner behaviour management

39.9%

Professional ethics and values

41.7%

Curriculum planning and progression

42.5%

Learner needs identification

44.8%

Use of assessment data in achieving learner outcomes

45.1%

Integrating technology, pedagogy and content

45.8%

Purpose of assessments before, during and after a lesson

46.5%

Inter-disciplinary lesson planning

47.0%

Types of assessment

48.4%

Learning theories in teaching practice

49.3%

Differentiated lesson plans

49.3%

Theories of child development

51.4%

Activities for professional development plan

53.4%

Responsible use of technology

56.5%

Accountability towards learners

57.1%

Classroom resources and safety

57.3%

Effective lesson planning

59.4%

Use of reflective thinking skills for effective teaching and learning

59.7%

Types of curriculum

60.4%

Delivery of a discipline-specific lesson

62.1%

Teaching strategies and methods

62.7%

Purpose of curriculum design and development

63.1%

Use of technology to support teaching and learning

63.9%

Approaches to discipline-specific teaching

66.2%

Effective classroom communication

71.9%

All 25 sub-domains measured in the 2023 edition, ordered weakest to strongest by correct-response rate. Each estimate is a simple proportion of correct answers among attempted responses. Base overall: 754,971 attempted responses from 15,548 teachers, 2023 edition; per-sub-domain bases range from 15,047 to 60,385 attempted responses and are listed row by row, with each row’s core domain, in the table behind this chart. Eleven of the 25 sit below 50%, and the accent bars mark those rows. Read this chart for what it is: it describes how the profession’s performance varies between areas of practice across this population, and it is not a picture of the 35-percentage-point spread inside an individual teacher quoted above, for which no base is reproduced on this page. Bars are drawn on a 0–100% axis and every value is printed beside its bar, so nothing here depends on colour or on hovering.

All 25 sub-domains, 2023 edition: core domain, correct-response rate and base n, ordered from weakest to strongest. Base overall: 754,971 attempted responses from 15,548 teachers.
Sub-domain	Core domain	Correct (%)	n (attempted responses)
Learner behaviour management	Learning Environment	39.9	15,047
Professional ethics and values	Ethics and Values	41.7	15,127
Curriculum planning and progression	Curriculum, Pedagogy and Technology	42.5	30,165
Learner needs identification	Curriculum, Pedagogy and Technology	44.8	30,185
Use of assessment data in achieving learner outcomes	Learning Assessment & Data	45.1	30,205
Integrating technology, pedagogy and content	Curriculum, Pedagogy and Technology	45.8	30,242
Purpose of assessments before, during and after a lesson	Learning Assessment & Data	46.5	30,187
Inter-disciplinary lesson planning	Student Engagement	47.0	30,204
Types of assessment	Learning Assessment & Data	48.4	45,295
Learning theories in teaching practice	Curriculum, Pedagogy and Technology	49.3	30,210
Differentiated lesson plans	Student Engagement	49.3	45,297
Theories of child development	Curriculum, Pedagogy and Technology	51.4	30,025
Activities for professional development plan	Teacher Growth & Professional Development	53.4	30,189
Responsible use of technology	Ethics and Values	56.5	15,095
Accountability towards learners	Ethics and Values	57.1	15,109
Classroom resources and safety	Learning Environment	57.3	30,211
Effective lesson planning	Student Engagement	59.4	30,209
Use of reflective thinking skills for effective teaching and learning	Teacher Growth & Professional Development	59.7	30,213
Types of curriculum	Curriculum, Pedagogy and Technology	60.4	30,248
Delivery of a discipline-specific lesson	Student Engagement	62.1	30,215
Teaching strategies and methods	Curriculum, Pedagogy and Technology	62.7	30,263
Purpose of curriculum design and development	Curriculum, Pedagogy and Technology	63.1	30,212
Use of technology to support teaching and learning	Curriculum, Pedagogy and Technology	63.9	30,231
Approaches to discipline-specific teaching	Curriculum, Pedagogy and Technology	66.2	30,202
Effective classroom communication	Learning Environment	71.9	60,385

Where these numbers come from

This section describes the shape of the gap rather than its size: not a profession that is evenly short of the standard, but individual practitioners who are strong in some parts of the job and weak in others at the same time.

The analysis then turns to what that shape does to a staff development calendar built around cohorts, and to the difference between an average that is true of a staffroom and an average that is true of nobody in it.

The chapter closes on the practical consequence for a school leader: a gap of this shape has to be located before it can be addressed, and locating it is a different task from delivering training.

4 · Finding three: what the pooled mean conceals #

Across 26,544 teachers assessed in 69 countries in the 2022 and 2023 editions, pooled mean competency was 51%; split by edition, mean competency was 46.8% in the 2022 edition and 53.4% in the 2023 edition, and because the two editions share no common items, the distance between them cannot be read as growth, decline or change over time.

Figure 3

The pooled 51% is not a level any cohort reached: it sits between a 46.8% edition and a 53.4% edition that share no common items#

2022 edition

46.8%

2023 edition

53.4%

Pooled, both editions

51%

Mean competency by edition and pooled across both, from the SURE competency research programme: 26,544 teachers assessed in 69 countries across 10,562 schools, from more than 1.2 million scored responses, 2022 and 2023 editions. The two editions share no common items, so the distance between 46.8% and 53.4% is not growth, not decline and not change over time; the rows are deliberately not drawn as a time series, and the third bar is shown to make the point that the pooled figure is the result of combining two different measurements rather than a level any cohort reached. Bars are drawn on a full 0–100% scale with no truncated axis and every value is printed on its own bar, so nothing here depends on colour or on hovering.

Mean competency by edition and pooled across both editions, SURE competency research programme, 2022 and 2023 editions. Base: 26,544 teachers assessed in 69 countries across 10,562 schools, from more than 1.2 million scored responses. The two editions share no common items, so the difference between them is not a change over time.
Measurement	Mean competency
2022 edition	46.8%
2023 edition	53.4%
Pooled across both editions	51%

How we research

This section sets out what is lost when two editions of an assessment are reported as one number, and why this volume nevertheless reports its central comparison pooled across both.

The analysis then turns to the rule the series holds itself to when two measurements are not on a common scale, and to the reading such a pair of figures cannot support however tempting the arrangement of them looks.

The chapter closes by restating the boundary in the plainest available terms, so that no later section of this volume, and no summary of it elsewhere, can be read as a claim about improvement.

5 · What this means for a school #

For a school the consequence is a targeting problem rather than a training problem: across 20,429 scored attempts with experience data, pooled 2022–2023, years in the classroom correlate with competency score at r = 0.03, so the field every school reaches for carries almost none of the signal, while the 35-percentage-point spread that does carry it sits inside a teacher’s own profile, where no personnel file records it.

This section translates the three findings into the decisions a school actually makes: how teachers are paired for mentoring, how development time is allocated, and what a leadership team looks at when it decides who needs support.

The analysis then sets out what a school would have to be able to see for those decisions to rest on something better than years of service, and what it would take to see it often enough to matter within a school year.

The chapter ends where the volume began, with the proxy on the personnel file, and states plainly what this evidence does and does not license a school to do with it.

Methodology and limitations #

This volume draws on one source, and it is worth being exact about it. Our own competency research: 26,544 teachers assessed across 69 countries and 10,562 schools, producing more than 1.2 million scored responses, in the 2022 and 2023 editions, against six CPAT domains, with 25 sub-domains measured in the 2023 edition. The experience comparison in Figure 1 is drawn from the 20,429 scored attempts that carried experience data, pooled across both editions. The sub-domain rates in Figure 2 come from the 2023 edition, on a base of 754,971 attempted responses from 15,548 teachers. The external findings named in this volume — Joyce and Showers (2002) on transfer, and the observation-reliability and turnover figures cited in the sibling volumes — belong to those sources and to those volumes, not to this one.

Limits

What the findings cover#

The experience comparison is drawn from 20,429 scored attempts that carried experience data, pooled across the 2022 and 2023 editions. It compares two groups — teachers in their first year and teachers with ten or more years — and reports the correlation across that base. It is not a statement about any named school, group, district or country, and it should not be read as one.

Participation was voluntary, and that shapes who is in the base#

This is the limitation that matters most to this volume, so it goes first. Teachers took the assessment because they chose to. The resulting population skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national workforce, and no figure on this page should be read as the score of a country’s teachers, a system’s teachers or your own. It bears directly on the headline finding: a self-selected sample is, if anything, a friendly test of the proxy, and the proxy still returned r = 0.03.

The assessment measures knowledge-for-practice, not enacted behaviour#

Every score on this page is a measure of what a teacher knows to do, not of what a teacher does in a room with thirty children in it. That boundary is why this volume says experience does not show up in this score rather than saying experience does not matter. Nothing here is a classroom observation, a performance rating or a judgement of any individual teacher, and no percentage on this page describes student achievement.

No growth claim between the editions#

Mean competency was 46.8% in the 2022 edition and 53.4% in the 2023 edition. The two editions share no common items, so that distance cannot be interpreted as improvement, decline or change over time, and this site does not interpret it that way anywhere, including in Figure 3. The experience comparison is reported pooled across both editions for the same reason: it is a comparison between groups within a base, not a comparison between years.

Two figures whose base and period are not reproduced here#

The source notes behind the 35-percentage-point median spread and the 24% figure are not reproduced on this page, so the base and period for those two are not shown, and neither figure is charted here for that reason. Where this page cannot show a base, it says so rather than supplying one. Figure 2 shows how performance varies between areas of practice across the population; it is not a measurement of the spread inside an individual teacher.

The chapter prose marked specimen is not the manuscript#

Each chapter above opens with a finding that is final and citable, and then continues in blocks marked as specimen text. Those blocks describe what the chapter will argue; they are not the argument, they contain no findings, and they should not be quoted as though they were. They are published in that form so the structure of the page can be reviewed before the full manuscript is typeset, and they will be replaced by the manuscript at this same address, with the findings and figures exactly where they are now.

What we do not publish at all#

We publish no product efficacy results, no adoption numbers, no named school case studies, no anonymised ones, no inter-rater agreement figures of our own and no error rates. Those studies have not been done. When they are, they will appear with their methods attached and their limits stated in the same place as their results, not in a later footnote.

The standing limits of the research programme#

These volumes sit inside a competency research programme whose limits apply to everything published under this masthead: 26,544 teachers assessed in 69 countries across 10,562 schools in the 2022 and 2023 editions, from more than 1.2 million scored responses, against six CPAT domains, with 25 sub-domains measured in the 2023 edition.

No author byline#

The volumes in this series are published without an individual author byline. There is no named author to attribute this volume to, so please cite it to SURE. The series begins at Volume 2; there is no Volume 1 to refer back to.

Full method: How we research · The dataset · Methodology

If years of service carry r = 0.03 of the signal, a school still needs to know which of a teacher’s own domains is the weak one, and returning exactly that — per teacher, after a lesson — is what our platform is built to do: how it works.

Read next

Where this evidence goes next#

For schools

Heads of school and leadership teams

What this research means when the teacher in front of you has fifteen years on file and a 35-percentage-point spread inside their own practice — and what a school can actually see between annual reviews.

Volume 2 · Key findings

The findings of this volume on one page

r = 0.03 across 20,429 scored attempts, the 35-percentage-point median spread, the 24% who are below 40% in one domain and above 70% in another, and the pooled 51%.

Research volume 03 · July 2026

Why Teachers Stop Showing Up

About 70% of teachers in their first five years have left the classroom or seriously considered it, and replacing one teacher costs $11,900–$24,900 (Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013)).

Research volume 04 · July 2026

Why Training Doesn’t Stick

Training alone produces classroom implementation in roughly 0–5% of participants; adding coaching on teachers’ own lessons raises implementation to approximately 95% (Joyce & Showers, 2002) — which is what closing a within-teacher gap would take.

Research volume 05 · August 2026

The Standard Nobody Sees

A single lesson observation scores between 0.14 and 0.37 on reliability, and an estimated 63% of one-off lesson judgements are wrong — the instrument schools currently use to find the gap this volume describes.

All research volumes

Cite this

Quoting Volume 2#

Citation

Permalink: #/research/the-gap-experience-doesnt-close/ — every chapter heading, figure and headline statistic on this page carries a frozen id, so you can link to a single finding rather than the whole volume. The findings alone sit at #/research/the-gap-experience-doesnt-close/.
Plain citation: SURE, The Gap Experience Doesn’t Close, Research volume 2, July 2026.
APA: SURE. (2026, July). The gap experience doesn’t close (Research volume 2).
Citing the headline finding: First-year teachers scored 49.5% on classroom competency and teachers with ten or more years scored 51.4% — a correlation of r = 0.03, across 20,429 scored attempts with experience data, pooled 2022–2023. Please carry the base with the number; it is what makes the figure checkable.
Citing the figures: Figure 1 (the experience comparison), Figure 2 (the 25 sub-domains, 2023 edition, base 754,971 attempted responses from 15,548 teachers) and Figure 3 (the two editions), The Gap Experience Doesn’t Close, SURE, July 2026. Figure 3 must not be reproduced as a time series: the two editions share no common items.
Author: The volumes carry no individual author byline, so SURE is the correct author. This series has no DOI; add the page address from your browser if your style requires a URL.
Licence: A reuse licence has not been published for this series yet, so we will not imply one. Quote the four findings and the three figures with attribution and a link back to this page. The connecting chapter prose marked as specimen text is not the manuscript and carries no findings — please do not quote from it, and please do not cite this page as though it did.

Stay with the series

If you would rather be told than keep checking, one email per volume is the whole arrangement — this one when its manuscript replaces the specimen prose, and each new volume on the day it goes up.

One email per volume, with the findings and their bases in the email itself. Front-end demo — not yet wired to an email service.

More about the series and what subscribing sends you

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Method · standing note

How we research

One assessment, two editions, 26,544 teachers in 69 countries — and a short list of things this series will not say. Every volume links here, because a finding is only worth what the method behind it is worth, and a method repeated in five places drifts. It is written once, in this place, and each volume is read against it.

Standing note · current as of August 2026 Applies to all five published volumes No author byline — cite to SURE

01

The dataset#

Every figure this series draws from its own data comes from one source: a scored teacher competency assessment, published in two editions, 2022 and 2023.

26,544 teachers assessed across the two editions
69 countries represented among those teachers
10,562 schools represented among those teachers
more than 1.2 million scored responses
six CPAT domains, resolving into 25 sub-domains
pooled mean competency 51% — 46.8% in the 2022 edition, 53.4% in the 2023 edition

Sub-domains were measured in the 2023 edition only, on a base of 754,971 attempted responses from 15,548 teachers. All 25 rows, with the correct-response rate and the base for each, are published as aggregate data on The dataset.

Results are reported as correct-response rates: the proportion of correct answers among attempted responses, at domain and sub-domain level. There is no index, no weighting and no composite score of our own construction sitting between the responses and the percentages we print. When a volume shows a figure, the arithmetic behind it is a division you could do yourself from the table on the dataset page.

02

Two editions, and no growth claim between them#

The 2022 and 2023 editions share no common items. That single fact governs how their two numbers may be used. Mean competency was 46.8% in the 2022 edition and 53.4% in the 2023 edition; those are two separate measurements taken on two different item sets, not two points on a line.

The difference between them is not growth, not improvement and not a trend, and this site does not report it as one. Where a single figure across both editions is useful, we publish the pooled mean of 51% and say in the same sentence that it is pooled.

Why we state this so loudly

46.8% to 53.4% inside a year is the shape of a good headline. It is also the shape of a false one. Because the item sets differ, a reader who put those two numbers on the same axis would be reading a change in the questions as a change in the profession. That is the single most tempting misuse of this dataset, so the correction is printed here, on the hub, on every volume that touches either figure, and on the dataset page itself.

03

What the assessment measures — and what it does not#

It measures knowledge-for-practice. A teacher sitting the assessment demonstrates what they know a teaching situation calls for. It does not measure enacted classroom behaviour. Nobody watched them teach.

So a sentence like “teachers scored 39.9% correct on learner behaviour management, on a base of 15,047 attempted responses in the 2023 edition” is a statement about knowledge. It is not a statement that four lessons in ten are well managed, and it must never be written as one. Read the other way round, a high score is not a guarantee that anything happens on Monday morning.

The distance between knowing and doing is a real research question. This dataset does not answer it and neither do we. Volume 4 is about that distance, and it argues from published literature — Joyce and Showers (2002) — rather than from our data, precisely because our data cannot reach it.

04

Who sat it, and who did not#

Participation was voluntary. Teachers chose to take the assessment; nobody was sampled into it.

The consequence is a population that skews motivated, graduate-qualified, toward India and toward women. It is not a probability sample of any national teaching workforce. No figure from it should be read as a national statistic, a country ranking or a system comparison, and this site publishes no country-level breakdowns for that reason. The 69 countries are where the assessed teachers came from; they are not 69 sampled frames.

This is the limitation we would most like to be able to remove and cannot. A voluntary assessment recruits the teachers most willing to sit an assessment. If self-selection moves these scores, the direction we would expect it to move them is upward — but we have not measured that, so it stays reasoning rather than a finding, and nothing in this series rests on it. What we can say is that the bias exists, that it is not quantified, and that it is a reason to read 51% as a description of the teachers who sat this assessment rather than a portrait of a workforce.

05

Whose finding is whose#

Two kinds of number appear in these volumes, and they are never merged into one.

Ours

Figures from the competency dataset described above. They are labelled as our own research and carry their base and their period in the same sentence as the number.

Somebody else’s

Figures from published literature. They are attributed to their original authors, on the page where they appear, in the form those authors published them. The external sources this series cites are:

Nye, B., Konstantopoulos, S., & Hedges, L. V. (2004). How Large Are Teacher Effects? Educational Evaluation and Policy Analysis, 26(3), 237–257. — used in Volume 6.
Joyce, B., & Showers, B. (2002). Student Achievement Through Staff Development, 3rd ed. ASCD. — used in Volume 4.
Kraft, M. A., Blazar, D., & Hogan, D. (2018). The Effect of Teacher Coaching on Instruction and Achievement: A Meta-Analysis of the Causal Evidence. Review of Educational Research, 88(4), 547–588. — 60 causal studies; 0.49 SD on instruction; 0.18 SD on achievement. We carry the authors’ caveat wherever we cite it: effects shrink as programmes scale, because coaching is capacity-bound.
Stronge & Associates / TIE Online, citing Mancuso, Roberts & White (2010) and Desroches (2013). — used in Volume 3.
ISC Research (2026). Global International Schools Snapshot 2026. — used in Volume 6.
Hanushek, E. A., & Rivkin, S. G. (2012). The Distribution of Teacher Quality and Implications for Policy. Annual Review of Economics, 4, 131–157.
Allen, J. P., Pianta, R. C., Gregory, A., Mikami, A. Y., & Lun, J. (2011). Science, 333(6045), 1034–1037.

We did not re-analyse any of these studies, pool them with our data, or restate their findings on a base other than the one their authors published. Where a volume sets an external number beside one of ours — as Volume 6 does with teacher effects and Volume 4 does with training transfer — both are labelled and the sentence says which is which.

The negative rule matters as much as the positive one. No external finding is ever presented as though it were ours, and no figure of ours is ever laundered through a citation to make it look external.

06

How we write a number#

One statistic, one self-contained sentence, carrying the metric, the population, the base and the period. “Across 26,544 teachers assessed in 69 countries in 2022–2023, mean competency was 51%” — not “most teachers score poorly”.
One number, one phrasing, everywhere. If a figure is 51% in a summary it is 51% in the body, never “just over half”, and never rounded into a different number on a different page.
Absolute dates only. “July 2026”, never “recently” and never “three months ago”.
No base, no number. If we cannot show what a figure was computed on, we write the sentence without it or leave the section out.

07

Product outcomes are projections, not results#

There is a hard line on this site between the research and the product, and this is where it is drawn.

Everything we say about what our platform would do for a school, a group, a system or a training provider is an evidence-supported projection: reasoning outward from the published findings above to an expected outcome. It is not a measured result from a deployment. Where the reasoning leans on somebody else’s evidence — for instance Kraft, Blazar and Hogan’s 2018 meta-analysis of 60 causal studies, reporting 0.49 SD on instruction and 0.18 SD on achievement — that evidence is cited, and so is the caveat its authors attached to it.

We have published no product efficacy results, no adoption numbers and no student-achievement claim at any ESSA tier, because those studies have not been done. Nothing on this site is described as proven, and no case study is attributed to an unnamed school. When there is an efficacy study, it will publish with its method and its limits in the same place as its result, and the difference between projected and measured will be legible on the page rather than buried in a footnote.

08

What we have not published#

Two of these sit inside the method itself. Naming them is the point of this page — a method note that only lists its strengths is marketing.

The two real gaps

No author bylines#

The volumes in this series carry no individual author byline. In a research series that is a genuine gap, not a stylistic choice we can defend: a byline is how a reader knows whose judgement they are reading and who to hold responsible for it. Until that changes, the volumes should be cited to SURE, and this page is the closest thing the series has to an accountable method statement.

No inter-rater agreement figures and no error rates#

We publish no inter-rater agreement statistics of our own and no error rate for this assessment. Those analyses have not been published, so the numbers do not exist for us to show you, and we will not estimate them in public to fill the hole. This matters most for a site that spends Volume 5 arguing about the reliability of somebody else’s instrument: the estimate reported there — that an estimated 63% of one-off lesson judgements are wrong — is a finding about lesson observation, published in that volume, and it is not an error rate for this assessment. We do not have that number.

Also absent, and for the same reason

Named school case studies — no school in the dataset is identified anywhere in this series.
Product efficacy results and adoption numbers.
Published price bands.
Volume 1. The series begins at Volume 2; there is no public Volume 1 and this site does not reference one.
Full HTML editions of Volumes 2, 3, 4 and 5 — their headline findings are final and citable, and the chapters are in preparation. Each of those pages says so on itself.

09

Where the data came from#

Provenance

The competency dataset behind this series was produced by Suraasa’s teacher competency research programme (2022 and 2023 editions); the published PDFs carry that origin.

That is stated here, once, in full, and it is not repeated elsewhere on this site. Every other page that needs it links back to this note rather than restating it.

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

Open data · CC BY 4.0

The dataset

The aggregate tables every volume in this series argues from, published in full on one page so that any figure you meet in a volume can be checked against its source without asking us for anything. Base overall: 754,971 attempted responses from 15,548 teachers, 2023 edition.

Licence: CC BY 4.0 25 sub-domains · six CPAT domains 2023 edition Aggregate only — no individual records

The base

What the programme holds#

Across both editions the competency research programme covers 26,544 teachers in 69 countries, from 10,562 schools, with more than 1.2 million scored responses. Pooled mean competency is 51%: 46.8% in the 2022 edition and 53.4% in the 2023 edition.

The 25 sub-domains beneath the six CPAT domains were measured in the 2023 edition only, on a base of 754,971 attempted responses from 15,548 teachers. That is the base for the whole of the table below, and the 25 row bases sum to it exactly.

Because the 2022 and 2023 editions share no common items, the two edition means are two separate measurements and not two points on a trend. Anyone reusing these tables inherits that constraint: no growth claim can be made between the editions, and none is made here. The full method note is How we research.

Table 1 · the full 25

All 25 sub-domains, with base#

Ordered weakest to strongest by correct-response rate. Eleven of the 25 sub-domains scored below 50% correct; the value is printed in every row, so nothing here depends on colour.

Table 1 · 2023 edition

Eleven of the 25 sub-domains measured in the 2023 edition scored below 50% correct, from learner behaviour management at 39.9% up to effective classroom communication at 71.9%

All 25 CPAT sub-domains measured in the 2023 edition of the competency research, ordered from the lowest correct-response rate to the highest. Columns: core domain, sub-domain, correct-response rate as a percentage, and n as the number of attempted responses. Base overall: 754,971 attempted responses from 15,548 teachers.
Core domain	Sub-domain	Correct (%)	n (attempted responses)
Learning Environment	Learner behaviour management	39.9	15,047
Ethics and Values	Professional ethics and values	41.7	15,127
Curriculum, Pedagogy and Technology	Curriculum planning and progression	42.5	30,165
Curriculum, Pedagogy and Technology	Learner needs identification	44.8	30,185
Learning Assessment & Data	Use of assessment data in achieving learner outcomes	45.1	30,205
Curriculum, Pedagogy and Technology	Integrating technology, pedagogy and content	45.8	30,242
Learning Assessment & Data	Purpose of assessments before, during and after a lesson	46.5	30,187
Student Engagement	Inter-disciplinary lesson planning	47.0	30,204
Learning Assessment & Data	Types of assessment	48.4	45,295
Curriculum, Pedagogy and Technology	Learning theories in teaching practice	49.3	30,210
Student Engagement	Differentiated lesson plans	49.3	45,297
Curriculum, Pedagogy and Technology	Theories of child development	51.4	30,025
Teacher Growth & Professional Development	Activities for professional development plan	53.4	30,189
Ethics and Values	Responsible use of technology	56.5	15,095
Ethics and Values	Accountability towards learners	57.1	15,109
Learning Environment	Classroom resources and safety	57.3	30,211
Student Engagement	Effective lesson planning	59.4	30,209
Teacher Growth & Professional Development	Use of reflective thinking skills for effective teaching and learning	59.7	30,213
Curriculum, Pedagogy and Technology	Types of curriculum	60.4	30,248
Student Engagement	Delivery of a discipline-specific lesson	62.1	30,215
Curriculum, Pedagogy and Technology	Teaching strategies and methods	62.7	30,263
Curriculum, Pedagogy and Technology	Purpose of curriculum design and development	63.1	30,212
Curriculum, Pedagogy and Technology	Use of technology to support teaching and learning	63.9	30,231
Curriculum, Pedagogy and Technology	Approaches to discipline-specific teaching	66.2	30,202
Learning Environment	Effective classroom communication	71.9	60,385

See the same 25 rows as a ranked chart How these figures were produced

All 25 sub-domains measured in the 2023 edition of our competency research, ordered by correct-response rate. Each value is a simple proportion of correct answers among attempted responses. Base overall: 754,971 attempted responses from 15,548 teachers; the 25 row bases sum to that overall base. Per-sub-domain bases range from 15,047 to 60,385 attempted responses, which is why a row with a low base is not a smaller finding but a differently-based one. Two sub-domains tie at 49.3%: learning theories in teaching practice, and differentiated lesson plans. Sub-domains were measured in the 2023 edition only, so this table is not a pooled 2022–2023 figure and no change over time can be read from it.

Licence

CC BY 4.0 — take it, use it, credit it#

The aggregate tables on this page are published under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). You may copy them, redistribute them, chart them, put them in a slide deck, quote them in a paper, and build on them commercially. The single condition is attribution: credit the source, and say if you changed anything.

You do not need to ask us, and you do not need to tell us. There is no field to fill in above the table for a reason.

Cite this dataset

Attribution string — paste this: The 25 CPAT sub-domains, 2023 edition: correct-response rates. SURE research. Base: 754,971 attempted responses from 15,548 teachers. Licensed CC BY 4.0.
Permalink: #/research/data-set/
APA: SURE. (2026). The 25 CPAT sub-domains, 2023 edition: correct-response rates [Data set]. CC BY 4.0.
A note on the author field: The volumes and these tables carry no individual author byline, so SURE is the correct author. The series has no DOI; add the page address from your browser if your style requires a URL.

Scope

What these tables contain, and what they do not#

What is here

The correct-response rate for each of the 25 sub-domains, 2023 edition.
The base n for each sub-domain, as attempted responses.
The core CPAT domain each sub-domain sits under.
The programme-level counts: 26,544 teachers, 69 countries, 10,562 schools, more than 1.2 million scored responses.
The pooled mean of 51%, and the two edition means of 46.8% (2022) and 53.4% (2023).

What is not here, and will not be

Individual teacher records, scores or identifiers of any kind.
School names or school-level results — no school among the 10,562 is identified anywhere in this series.
Country-level results or rankings. Participation was voluntary, so a country figure from this dataset would not mean what a reader would take it to mean.
Item text, item-level statistics or free-text answers.
Anything at all about students.
2022-edition sub-domain figures — sub-domains were measured in the 2023 edition only.
Inter-rater agreement figures and error rates. Those analyses have not been published.

Three constraints that travel with the data

The n column counts responses, not people. A teacher contributes attempted responses to several sub-domains, so the column describes the base of each measurement and is not a headcount. The headcount for this table is 15,548 teachers.

These are scores for knowledge-for-practice, not enacted classroom behaviour. No row in this table describes what any teacher does in a classroom, and a reuse that implies otherwise is a misreading of the source, not of the licence.

The population is self-selected. Participation was voluntary, so it skews motivated, graduate-qualified, toward India and toward women, and it is not a probability sample of any national workforce. Please carry that sentence with the numbers.

Access

If you need it as a file#

Everything the volumes cite from the 2023 sub-domain analysis is on this page already, in a real table you can select, copy and paste into a spreadsheet. For most reuses that is the whole job, and it costs you nothing.

If you need a machine-readable copy instead, a request takes two things and nothing else: an email address to send the file to, and the organisation asking. No role, no phone number, no project description, no call. We ask for the organisation only so that we know who is using the tables, and it is not passed to anything else.

A public request address has not been published on this site yet — saying so is more useful to you than an address that bounces. Until it goes up, subscribing to the series is how you will hear that the file is available; it is one email per volume and nothing else.

Subscribe to the research series.

One email when a volume publishes. No product marketing in it — that is the deal, and it is the reason people stay subscribed across a ten-month decision cycle.

Stay up to date

One field. Unsubscribe in one click, any time. By subscribing you agree to our Privacy Policy.

← Back to the homepage

Not built yet

This page is scheduled, not written.

The blueprint specifies — it is in the build order but has not been built. Rather than bounce you to the homepage and pretend the link worked, here is where it sits.

Build order · blueprint §12

Phase 2 — the conversion spine

Platform, trust, pricing, pilot and demo.

Phase 1, the citable core, is live now: the research index, five volume pages, the process statement and the data set.

Go to the research Back to the homepage

← SURE Terms of Use →

Legal

Privacy Policy

Effective 26 August 2026 · Applies to this website and describes the commitments the SURE platform is built on.

SURE ("we", "us") helps schools develop teaching quality from ordinary lesson recordings. This policy explains what we collect through this website and how the platform treats classroom data. We built the product privacy-first, and this page is written to be read, not skimmed past.

What we collect on this site

Only what you type into our forms: your name, work email, role, and school or organisation when you request a walkthrough, subscribe to the research series, or request a PDF. We use it to respond to you, send the research you asked for, and arrange the conversations you request. The research mailing list receives research and nothing else. We do not sell personal information.

Classroom recordings on the SURE platform

The platform's rules are architectural, not just policy, and they are the same ones published on our homepage:

Recording is teacher-initiated. Nothing records without the teacher starting it.
Student data is never captured and is always masked. The system observes teaching, not children.
The individual Class Insight Report reaches the teacher and no one else. Leadership sees aggregate patterns only — never a ranking, never an individual.
If something wasn't in the recording, nobody is asked to supply it.

Legal frameworks

Recording, consent, and retention were designed around India's Digital Personal Data Protection Act (2023), FERPA (United States), and the GDPR — researched before the recording feature was built, not retrofitted. Platform deployments operate under a written agreement with each school, including a data processing agreement that governs roles, retention, and deletion.

Retention and deletion

Website form data is kept only as long as it is relevant to the conversation you started; ask us and we will delete it. Platform data retention is governed by the school's agreement, and deletion requests are honoured under it.

Your rights

You can ask what we hold about you, correct it, or have it deleted. Reach us through any form on this site — — and we will act on it.

Changes

If this policy changes, the effective date above changes with it, and material changes will be noted plainly on this page.

← SURE Privacy Policy →

Legal

Terms of Use

Effective 26 August 2026 · These terms cover this website. Platform deployments are governed by a separate written agreement.

By using this website you agree to these terms. They are short on purpose.

What this site is

This site presents the SURE platform and publishes the SURE Research series. Content is provided for information; it is not professional advice for any particular school's circumstances.

Using the research

Full research reports are free to read here. Figures and findings may be quoted and reused with attribution to SURE Research. The aggregate dataset is licensed CC BY 4.0. Research volumes themselves may not be resold or passed off as your own work. External findings we cite belong to their original authors, and we attribute them.

Your submissions

When you submit a form, you agree the information is accurate and yours to share. We use it as described in the Privacy Policy.

Acceptable use

Don't misrepresent your affiliation, interfere with the site's operation, or use automated scraping in ways that burden the service or bypass access gates.

Intellectual property

The SURE name, this site's design, and its content belong to us. The reuse rights granted above are the exception, and we like people using them.

The platform

Access to the SURE platform — recordings, reports, and leadership views — is governed by a written agreement between SURE and your school, including its data processing terms. Those terms, not these, govern platform use.

No warranties; limitation of liability

The site is provided "as is". To the extent the law allows, we are not liable for indirect or consequential losses arising from use of this website. Nothing here limits liability that cannot lawfully be limited.

Changes

We may update these terms; the effective date above will change when we do. Continued use of the site after a change means you accept it.

Ten years in the classroom. Under two points of improvement.

One recording. Two views. Nothing else asked of anyone.

From one lesson to a compounding loop.

A teacher records

SURE listens

One thing comes back

The pattern emerges

Scored on CPAT, not on vibes.

The observation cycle, side by side.

Four kinds of buyer. Four different arguments.

Most products cite research.This one is downstream of it.

Rules that make recording feel safe.

Teachers own their reports

Never a ranking

Students are not the subject

Nothing beyond the recording

Privacy law came first

Procurement answers, ready

Coaching that finally feels like trust.

Asked in every first conversation.

Decide with the evidence in hand.

Subscribe to the research series.

See it on your own classrooms.

One chart no single volume contains.

Find the volume written for you.

By volume

By finding

By who you are

Five volumes.

One Brand, a Thousand Classrooms

The Standard Nobody Sees

Why Training Doesn't Stick

Why Teachers Stop Showing Up

The Gap Experience Doesn't Close

What's free, what costs an email, and why.

Subscribe to the series.

Method, independence, and what the series does not claim.

Subscribe to the research series.

What this volume found

The promise a group makes is made at network level, experienced at classroom level, and managed at the one level in between where the signal cancels out #

Get Volume 6 as a PDF#

Three fields, and we disclose all three#

1 · The promise a network makes #

2 · Finding one: the network era has arrived, and it concentrates the risk #

3 · Finding two: the spread that breaks the promise lives inside each campus #

4 · Finding three: oversight thins exactly as the group grows #

5 · Finding four: the group’s instruments report too late and train too broadly #

Annexure: all 25 sub-domains in our research data #

Methodology and limitations #

What this evidence cannot tell you #

References #

Where this evidence goes next

The same evidence, arranged around a head office’s decisions

The four findings of this volume on one page

The Standard Nobody Sees

Why Training Doesn’t Stick

The Gap Experience Doesn’t Close

Quoting Volume 6

Get the next volume when it is published

Subscribe to the research series.

What this volume argues from#

A standard written for thousands of classrooms is verified through a sample that thin, with an instrument that unstable #

Get Volume 5 as a PDF#

Three fields, and we disclose all three#

1 · The question — how does a body that owns a teaching standard know it is enacted? #

2 · Finding one: a single lesson observation scores between 0.14 and 0.37 on reliability #

3 · Finding two: an estimated 63% of one-off lesson judgements are wrong #

4 · Finding three: 2.4 formal observations per evaluation cycle, against roughly 1,000 lessons taught a year #

5 · What this means for a system, an inspectorate or a standards body #

Methodology and limitations #

References #

Where this evidence goes next#

The same evidence, arranged around a standards body’s decisions

The findings of this volume on one page

One Brand, a Thousand Classrooms

Why Training Doesn’t Stick

How we research

Quoting Volume 5#

Be told when the manuscript publishes#

Subscribe to the research series.