Research Agenda and Ethics Framework

Why the Council conducts research

Standards of evidence

• E1 Vendor testimonial

  A school or teacher says the tool helped. Unverified, unquantified, and selected by the party with an interest in it. A record of what someone believes. It carries no evidential weight and the Council does not cite it as evidence.

• E2 Vendor-reported usage metric

  Logins, questions attempted, minutes engaged, drawn from the vendor's own systems with no external check and no comparison group. Measures use. It answers whether a product was used, never whether it worked.

• E3 Documented case description

  An independent, structured account of what one school did, what changed, what it cost and what went wrong, against measures fixed before the account began. Honest description, and often practically useful. It is not evidence of effect.

• E4 Pre-post pilot

  The same pupils measured before and after on an instrument set in advance. Cannot separate the intervention from ordinary teaching, maturation, examination timing or the attention the pilot itself attracts. No causal claim may be drawn from this tier.

• E5 Matched comparison

  Comparable sections or schools, one with the intervention and one without, prior attainment accounted for, marking blind to group. Supports a cautious statement of association, provided the confounders that remain are named in the same paragraph as the finding.

• E6 Randomised or stepped-wedge study

  Allocation determined by design rather than by preference, pre-registered before collection, analysed as allocated. Supports a causal claim bounded strictly to the population, subject, tool and conditions studied.

• E7 Replication and synthesis

  The same question answered independently in different settings, with converging results. The only tier on which the Council would use the word established, and not a tier a first-year council reaches alone or reaches quickly.

Research ethics where children are involved

• Consent and assent both

  Written consent from a parent or guardian, and separately the child's own assent in language the child understands. A child's refusal ends that child's participation even where the parent has consented. Assent is revocable at any time and is not sought again after refusal.

• Opt-out without penalty

  Any pupil, parent or teacher may decline or withdraw at any point without giving a reason. Refusal has no effect on marks, class placement, teaching received, teacher appraisal or the school's standing with the Council. Non-participating pupils receive the ordinary lesson.

• Ethics review before pupil data

  No collection involving pupils begins before written clearance from the panel, covering protocol, consent materials, stated burden and data plan. A protocol changed after clearance returns for review, and collection stops while it does.

• No covert observation

  Pupils, parents and teachers are told what is being collected, by whom and for what purpose, before collection begins. The Council does not conduct covert observation, undisclosed recording, or any study that depends on deceiving a child.

• No education withheld

  Comparison groups receive normal teaching in full. Where an intervention is withheld for comparison, it is offered to that group at the close of the study if the school wishes. Instructional time the study will consume is stated in hours before consent is sought.

• The teacher may stop it

  The class teacher may suspend a study in their own classroom at any moment, without notice and without justification, and it resumes only if that teacher agrees. Suspensions are recorded on the study register and are not renegotiated with the teacher.

• Safeguarding overrides the study

  Anything observed or disclosed during research that suggests risk to a child is acted on the same day: reported to the school's designated authority, and, where the Protection of Children from Sexual Offences Act, 2012 is engaged, reported by the individual concerned to the Special Juvenile Police Unit or the local police under section 19, whether or not the school agrees, whether or not the Council agrees, and whether or not the school is itself the institution implicated. Research confidentiality does not extend to child safety, and every consent document says so in those terms.

• A disclosure made to a researcher

  A child who tells a researcher something is not handled as a data point. The researcher promises no confidentiality, does not question the child further or investigate, records the child's own words as soon as possible afterwards, and reports under the rule above the same day. The child is told what will happen next, in language the child understands, and is not told the matter will be kept secret.

• Vetted, and undertaken in writing

  Every adult who will have contact with children on Council fieldwork verifies identity to the school in advance and signs a safeguarding undertaking that names the section 19 duty and the two rules below. Access is refused without both. The school may refuse any individual without giving a reason, and is told in writing that it may.

• Two adults, and no private contact

  No researcher is alone with a child. A member of the school's staff is present at every observation, interview and assessment. No researcher contacts a child directly by telephone, message, e-mail or social media, at any time, for any purpose connected with a study or otherwise, and no study design may require it.

• Adverse incidents are logged and counted

  Every safeguarding concern, near miss, breach of this section and complaint is entered in an adverse-incident log with its date, what was done and how it ended. The ethics review panel inspects the log at each meeting, and the number of entries and their disposal appear in the Council's annual public record. A breach of this section suspends the person from all Council work immediately, before and regardless of the removal procedure in the Charter.

• Burden stated, then honoured

  Consent materials state pupil time, teacher time and instructional time in hours. If actual burden exceeds stated burden, collection pauses and consent is sought afresh. Overruns are recorded on the register with the amount.

Data governance for research

• De-identify at source

  Direct identifiers are replaced with a study code at the earliest technically feasible point, normally within the school. The key linking code to child is held by the school or a named custodian, never by the analysis team, and is destroyed at the close of the study.

• Collect only what is listed

  A study collects only the fields named in its published protocol. Data outside that list is not collected, not derived and not inferred, whether or not it is available and whether or not it might later be interesting.

• Stored and processed in India

  Research data is stored and processed on infrastructure located in India. Any cross-border transfer requires the school's specific, separate written agreement and is disclosed in the resulting publication.

• Retention limits, then destruction

  Identifiable data is destroyed within thirty days of the close of collection, and the key linking code to child is destroyed at the close of the study. Withdrawal can be executed only while that key exists: afterwards nobody can find a particular child's records in the de-identified set, and the right to have them removed ends. That date is stated in the consent form in those words, so no parent is given a right that expires silently. De-identified datasets are retained three years from publication so findings can be checked, then destroyed. Consent records and the coordinator's verification certificates are kept separately for seven years, because they are the proof that consent was given at all; they are not research data and are not destroyed with it, and a deletion request does not reach them. Each destruction is recorded with its date on the study register and the record is given to the school.

• A named access register

  Every individual with access is named in a register kept with the protocol, with the date access was granted and withdrawn. Access is by name. Shared credentials and role-based access to research data are prohibited.

• No secondary use, no re-identification

  Research data is not used for product development, marketing, sales targeting, model training or any purpose beyond the study named in the protocol. No attempt to re-identify a de-identified dataset is permitted, by anyone, at any time.

• What survives the study

  At the close of a study the school receives its own data and the analysis. The Council itself holds nothing, because it has no legal personality and cannot. The de-identified set is held by PrepGraph as bare trustee for the Council under the Charter, segregated from PrepGraph's commercial systems, on the terms above and on no others, and PrepGraph may not use it for any purpose of its own. PrepGraph retains nothing else gathered for the study, and nothing beyond what its ordinary service agreement with that school already permits. That the funder is also the custodian is a conflict this framework contains rather than removes, and it is stated here so that a reader can weigh it instead of discovering it.

The convener conflict, stated plainly

• Disclosure on the first page

  Where PrepGraph has contributed funding, platform access, anonymised data or staff time, that role is stated on the first page in full sentences, with amounts or hours where they are known, not in an endnote or a footer.

• Pre-registration outside our control

  Questions, outcome measures, sample, exclusion rules and analysis plan are lodged on an independent public registry before collection begins. Deviations are reported with reasons. Analyses run after the fact are labelled exploratory and cannot support a conclusion.

• Factual review only, in writing

  PrepGraph has no editorial review. It receives one written factual-accuracy pass, limited to statements about its own systems, of ten working days. Silence publishes. Every comment it makes and every change that follows is logged and published with the output. It cannot delay, amend or withhold anything.

• Negative results, equal prominence

  A study finding no effect, or finding harm, is published on the same channel, at the same length, with the same announcement as one finding benefit. Studies abandoned before completion stay on the register with the reason for abandonment.

• Product evidence stays PrepGraph's own

  PrepGraph's internal results and product claims are never restated as Council findings. The Council does not validate, verify, certify or endorse them, and any Council document that refers to them says so in the same sentence.

• No comparative product evaluation

  The Council does not test vendors against one another, publish shortlists, rate products or advise on procurement. "The Council does not endorse individual procurement decisions or technology vendors. Its role is to advance responsible AI readiness, educator capability and evidence-based innovation in schools."

• Publication decided outside the secretariat

  Publication is governed by Charter Article 12: no study is published in the Council's name without the approval of the Founding Chair and a majority of the voting members, and none while fewer than seven voting members are in office. The ethics review panel holds a veto over publication of a study, exercisable on grounds of ethics or participant protection and overridable by nobody; it holds no share in the approval. The secretariat prepares, formats and distributes. It does not decide what is published, when, or in what words.

• Authors may publish regardless

  If the Council has not published a completed study within sixty days of its registered reporting date, the authors may publish it themselves, in full, under the same open licence. The Council signs nothing: it has no legal personality and enters into no contracts. Research agreements are made by PrepGraph as bare trustee for the Council, on the Council's terms, under the Charter's article on legal personality. Every such agreement contains that right, no agreement may remove it, and it is not waivable on the authors' behalf by the Council, the Chair or PrepGraph.

• No evaluation of a PrepGraph product on PrepGraph's money

  The Charter prohibits the Council from publishing an evaluation of a PrepGraph product funded by PrepGraph, and this framework supplies no way of packaging one. An earlier draft gave that study a front-page label, which was a way of formatting the thing the constitution forbids, and it has been removed. Until a funder independent of PrepGraph is found, the Council cannot examine PrepGraph's products at all, and the absence of a finding about them is not a favourable one.

• The convener's product cannot be studied sideways

  Where the intervention studied is the convener's product, the study is an evaluation of that product whatever it is titled, and the prohibition applies to it. Every protocol names the product that delivered the intervention, by vendor and version, and that name is published with the finding. A study of AI-assisted practice run on PrepGraph's platform is a study of PrepGraph's platform; reporting it as a general finding about AI in Indian classrooms is the evasion this rule exists to stop.

• Independence is claimed narrowly

  The labelling rule survives only for the case the Charter does contemplate: a study funded by a funder independent of PrepGraph in which PrepGraph supplied data, platform access or staff time. That study states on its front page who funded it, what PrepGraph supplied, what PrepGraph was not permitted to do, and that it is not independent of PrepGraph in the respects in which PrepGraph contributed.

• A duty to find the funder that would allow it

  A restraint that happens to suit the party restrained is not left open-ended. The Council seeks a funder independent of PrepGraph for an evaluation of PrepGraph's products within the inaugural term, and again in each term after it. Whom it approached, when, and what came of it are recorded in the annual public record whether or not the attempt succeeded. Until such a study is published, every Council research output states on its first page that no product of the convener has been examined and why.

• The same right of reply for anyone criticised

  A third party identified and criticised in an output receives the passages that concern it on the same terms and the same ten working days that PrepGraph receives, may publish its objection alongside in its own words, and can no more delay, amend or withhold the output than PrepGraph can. A pre-publication right held by the funder and not by the subject of the criticism is not a control.

• Legal risk is not managed by quiet deletion

  Any output naming a third party adversely is read before publication by a lawyer instructed for that purpose, and the fact and date of that reading are recorded with the study. On receipt of a legal notice nothing is silently unpublished: the notice, the response and the decision are minuted, the document stays at its address, and any change is made as a dated correction under section seven. PrepGraph is the publisher of record and carries the risk of publication, as bare trustee for the Council under the Charter.

Open publication

• Free and ungated

  Every report, instrument and dataset summary downloads without registration, e-mail address or form of any kind. No Council research output is used as a lead-generation mechanism, for PrepGraph or for anyone else.

• Licensed for school reuse

  Published under Creative Commons Attribution 4.0. Copyright is held by PrepGraph as bare trustee for the Council under the Charter, because the Council can hold nothing in its own name, and PrepGraph cannot alter the licence the Council has settled. Schools may copy, translate, adapt and teach from Council material, including in paid staff training of their own, provided the Council is credited. The licence covers the text. It does not license the Council's name or marks, whose use is governed by the Charter, and no reuse may present the Council as endorsing a product.

• Instruments published with findings

  Question wording, rubrics, marking schemes, consent forms and analysis plans are published with each report, so another party can repeat the work and reach its own conclusion without asking the Council for anything.

• A public register of studies

  Every study is listed when it starts, not when it succeeds, with a registration number and a scheduled reporting date. Entries record studies running, completed, overdue and abandoned, with reasons. Entries are never removed.

• A two-page summary in plain language

  Each report carries a summary a principal can read in ten minutes: what was found, in how many schools, and what the finding does not support. Limitations appear in the summary itself, not only in the full text.

• Corrections made in the open

  Errors are corrected publicly, in a dated correction log. The superseded version stays accessible at its original address, and the correction states what changed and why. Silent revision of a published document is prohibited.

How schools and researchers take part

• What a school provides

  A named coordinator, agreed access to specified classes, teacher time stated in hours in advance, and permission for the data named in the protocol. Nothing beyond that is asked, and no commercial conversation is attached to participation at any stage.

• No fee in either direction

  Schools pay nothing to take part and are paid nothing to take part. No discount, licence, product, preference, priority or recognition is offered for participating, for continuing, or for a favourable result.

• Its own data, read honestly

  Each school receives its own results before publication, in plain language, including results that are unflattering, with sample size and the limits of interpretation stated. The school is told what its data cannot support as well as what it can.

• Anonymous by default

  Schools are not named unless they ask in writing to be named. Reporting is aggregate by default, and no combination of published details is permitted that would identify a school, a class or a teacher indirectly.

• Right to withdraw, unconditionally

  A school may withdraw at any point without a reason, and chooses whether its data is destroyed or retained de-identified. A withdrawing school is not named, not characterised in the report, and not approached to reconsider.

• Researcher obligations

  Researchers declare funding and conflicts on the same terms that bind PrepGraph, submit to the same ethics review, register on the same external registry, and publish under the same open licence. There is no separate track for anyone.

• Authorship reflects work

  Authorship reflects contribution to the study. Office-holders of the Council and employees of PrepGraph are not added to papers they did not work on. Honorary and courtesy authorship are prohibited, including for the Founding Chair.

What the Council will not claim

• No causal claim from a pilot

  Improved, raised, increased and reduced are not used of a pre-post pilot. Association is described as association, with the alternative explanations named in the same paragraph, not in a later section.

• No proven, validated or guaranteed

  These words do not appear in Council outputs about AI in education, in any tense, including in summaries and spoken statements. Nor does the Council permit its material to be quoted as though it had used them.

• No proprietary metric as a standard

  Any index or framework the Council builds is published with its full construction, its scoring and its known weaknesses. It is never described as an industry, national or recognised standard, or as a benchmark a school must meet.

• No number without its denominator

  No percentage is published without the sample size, the comparison it is measured against, the attrition, and a statement of uncertainty. A figure that cannot carry those four things is not published at all.

• No system claim from a few schools

  Findings are stated for the population actually studied. The Council does not extrapolate from participating schools to Indian schools, to a board, to a state, or to a phase of education.

• No verdict on a vendor

  No product, including PrepGraph's, is described as effective or ineffective on the Council's authority. The Council does not rank, rate, shortlist or advise against named products, in publication or in private correspondence.

• No representativeness from volunteers

  Schools that volunteer for research are not a sample of Indian schools. Every publication drawn from volunteer participation states that limitation in the summary, each time, without softening.

• No announcement before the report

  No press statement, briefing, presentation or public claim precedes publication of the document it describes. The document comes first, and the claim may not exceed it.

• No status conferred by participation

  "The Council is a voluntary, non-statutory initiative. It is not a government, regulatory or accrediting body, and it does not certify, rate or approve schools, curricula or products." Taking part in research confers no rating, recognition or approval on a school.