Learning Impact Measurement Dictionary · v1.1 External edition

WISE Testbed · Cycle 2

TA39 Learning Impact Measurement Dictionary

Define the smallest defensible set of records and measures needed to describe TA39 assessment, student revision and teacher instructional response—without exposing classroom content or overstating what product data can establish.

Prepared for

WISE and EdTech Impact

Version

1.1 External Edition — August 10, 2026

Purpose: Define the smallest defensible set of records and measures needed to describe TA39 assessment, student revision and teacher instructional response—without exposing classroom content or overstating what product data can establish.
Status: Proposed measurement and data-mapping contract for joint review. The educational constructs and calculations reflect TA39's current version 1.1 measurement model. Study sampling, schedules, thresholds, questionnaires and comparative methods remain subject to the jointly agreed Cycle 2 research protocol. The API examples in Appendix B are representative schema 1.1 payloads; items explicitly marked pending must not be treated as settled receiver behavior.

1.The educational model

TA39's ordinary assessment pathway and Revision Rounds share one formative-assessment foundation:

Student submits work → TA39 organizes rubric-aligned evidence of understanding → teacher reviews and adapts it → teacher releases personalised feedback → teacher and student can act on it

Revision Rounds extends that foundation:

Student responds and chooses a focus → revises → teacher examines the linked change → decides what should happen next

A second draft is not, by itself, evidence of learning. It shows that a student submitted again; it does not show that the student understood the feedback, made a relevant change or demonstrated stronger evidence on the intended criterion. Likewise, a teacher editing or rejecting an AI-prepared feedback draft is not automatically evidence of system failure. It may show the professional judgment that responsible use requires.

The measurement model therefore keeps two connected processes visible:

Student: Receive feedback → Understand it → Choose what to work on → Revise → Check the result → Use the learning again

Teacher: Review the evidence → Notice student needs → Decide what to do next → Give students another opportunity → See whether the learning continues

The central questions are:

Did the evidence and released feedback credibly represent the student's work? Did the student respond meaningfully and show stronger evidence on the targeted learning criterion? Did that evidence contribute to the teacher's next decision?

These are descriptive measures. Product records can show that events followed one another, were linked, or were associated. They do not prove that TA39 caused learning unless a separate comparative research design supports that claim.

2.Evidence rules

Keep the stages separate. A view is not understanding. A plan is not revision. A resubmission is not meaningful uptake. Uptake is not criterion-level evidence change, and criterion-level evidence change is not transfer.
Preserve the evidence chain. The strongest immediate account is: teacher-released feedback → student-selected priority and plan → relevant change in the next draft → teacher-reviewed outcome.
Keep teacher judgment visible. Teacher-confirmed, teacher-changed, sampled-review and unreviewed TA39 suggestions remain distinguishable.
Show denominators and coverage. Every percentage identifies who was eligible, how much evidence was usable and how much was teacher reviewed.
Do not create one impact score. The learning loop is reported as a short evidence profile so that researchers and educators can see where it worked or stopped.
Protect the student's cognitive work. TA39 may make the learning problem actionable; it does not do the student's reasoning or writing.
Keep educational use separate from discipline. These measures are not authorship detectors, misconduct findings, grades or proof of effort.

3.Outcome hierarchy

The complete model includes 22 computational measures. They do not all have the same evidential role.

Research role	Measures
Primary student outcome	R-08 targeted criterion-evidence change within supported revision, after independent method validation
Primary teacher outcome	T-03 recorded instructional decision, interpreted alongside eligible teacher report
Quality prerequisite	Independent human validation of the final teacher-released feedback for the relevant language and task context
Mechanisms	Feedback view, student clarity and planning, traceability, uptake and teacher review coverage
Implementation and context	Activity, opportunity, feedback pipeline, timing, rubric use, assignment context, revision participation and follow-up
Later learning	L-01, deferred until comparable later work exists

The primary outcome is not the number of drafts. For the student, it is whether independently rated evidence on the targeted criterion became stronger in the supported revision. For the teacher, it is whether the evidence informed a recorded instructional decision. Before the relevant methods are independently validated, TA39 may report clearly labelled product suggestions and implementation measures, not headline learning-impact findings.

For R-07 and R-08, every report must show the complete denominator cascade:

eligible → feedback available → viewed → revision opportunity → revised → traceable → sufficient evidence → classified outcome

4.The 22 computational measures

4.1Shared assessment foundation

These measures apply to ordinary one-submission assessment and Revision Rounds.

A-01assessment_activity_distribution

Question	When and where did teachers and students use the assessment workflow?
Unit	One deduplicated teacher or student assessment action. Unique actors and assignments are counted separately.
Population and window	Authorized actions for eligible participants and assignments; summarized by school-local day, week, month and study cycle.
Calculation	Counts and shares by action family, role and agreed school-hours/after-school band. Reach rates use eligible actors or assignments as the denominator.
Exclude or separate	Automated actions, retries, withdrawn records and unknown time-zone classifications.
Interpretation	Describes use and timing, not quality, engagement or learning.

A-02submission_opportunity_distribution

Question	Who had a genuine opportunity to submit, who submitted and who did not?
Unit	One learner × assignment opportunity; Revision Rounds also retains one learner × target-round opportunity.
Population and window	Learners in the frozen authorized roster who were assigned the work and had an open submission window.
Calculation	One terminal state per opportunity: submitted on time, submitted late, no submission, excused, withdrew before opportunity, joined too late or technically unavailable. Participation rates use only learners with a genuine opportunity.
Exclude or separate	Excused, withdrawn-before-opportunity, late-enrolment and technical-unavailability cases remain visible outside the rate.
Interpretation	No submission means no qualifying submission during a real opportunity; it does not establish disengagement or failure.

A-03initial_skill_evidence_distribution

Question	What did the selected source work show against each applicable criterion?
Unit	Learner × immutable source attempt × applicable skill or rubric criterion.
Population and window	Valid submissions for which the criterion applies and a versioned method can evaluate the frozen source attempt.
Calculation	Report demonstrated, partly demonstrated, not yet demonstrated and insufficient evidence. Outcome percentages use sufficient-evidence cases; insufficient evidence is also reported over all eligible cases.
Provenance	TA39 suggestion, individual teacher review, sampled review or unavailable are shown separately.
Interpretation	A criterion-level starting point, not a fixed learner trait, grade or diagnosis.

A-04feedback_pipeline_outcome_distribution

Question	What happened to each AI-prepared feedback case before student release?
Unit	Learner × source attempt × feedback case.
Population and window	Cases with an AI-prepared draft and a valid teacher review/release opportunity, through release, rejection or the agreed close/grace cutoff.
Calculation	One terminal disposition: released unchanged, edited, substantially rewritten, rejected or not released. Active unresolved cases remain separate.
Deduplication	One disposition per feedback case and immutable version; a later update is a new release version.
Interpretation	Editing, rewriting or rejection may demonstrate appropriate professional judgment. The distribution is not an AI-reliability score.

A-05feedback_timeliness_distribution

Question	How long did analysis, teacher review, release and student visibility take?
Unit	One valid submission-feedback case and each ordered interval within it.
Population and window	Cases with both valid endpoint timestamps for submission → analysis ready → review complete → release → student visibility.
Calculation	Median, quartiles and 90th percentile for each interval, with eligible and missing counts. Active review time is the union of foreground review intervals under a versioned idle rule.
Exclude or separate	Missing endpoints, clock errors, no release, integration delay and technical failure.
Interpretation	Speed is not feedback quality or teacher engagement.

A-06feedback_view_rate

Question	Did students open feedback that was actually available to them?
Unit	Learner × immutable feedback-release version.
Population and window	Releases that became visible and renderable, from visibility until the earlier of the target submission, cycle end or study cutoff.
Formula	Learners with at least one qualifying successful render ÷ learners for whom that release was visible and renderable.
Deduplication	First qualifying view per learner × release version.
Interpretation	Establishes an opportunity to engage. It does not prove reading, attention, understanding, trust or learning.

A-07teacher_feedback_intervention_distribution

Question	Why did teachers change or reject drafted feedback, and which interventions identified an AI issue?
Unit	One materially changed, rewritten or rejected feedback case with an optional structured reason response.
Population and window	Intervention cases for which the reason control was available, through terminal release or rejection.
Calculation	Reason-category counts and response coverage. AI-issue rate = cases with at least one approved AI-issue reason ÷ reviewed cases with an explicit reason response, including an explicit no-AI-issue response.
Provenance	Explicit teacher selection only; no inference from wording differences.
Interpretation	Voice adaptation, contextualisation or knowledge of a learner is not automatically an AI defect. Do not rank teachers by intervention frequency.

A-08rubric_and_template_activity_distribution

Question	How were rubrics and feedback templates created, adapted, reused and AI-optimized?
Unit	One teacher action on a stable instrument/version or AI-optimization request.
Population and window	Scoped teachers and instruments used for authorized assignments; summarized by assignment, month and study cycle.
Calculation	Counts and shares for create, edit, apply, reuse and retire; AI-optimization requested, applied, modified, dismissed or unresolved.
Deduplication	Transaction and resulting version; autosave and delivery retries do not count as edits.
Interpretation	Activity does not establish instrument quality or task fit.

A-09assessment_context_distribution

Question	In which grades, subjects, languages, task types and curriculum contexts was TA39 used?
Unit	One versioned assignment context.
Population and window	All scoped ordinary and revision-enabled assignments, frozen when active; material changes create prospective versions.
Calculation	Counts, shares and completeness by school, class, grade band, subject, language, task type, curriculum/framework, assessment mode, source-material state and rubric/template version.
Missingness	Unknown or unmapped context is an explicit category.
Interpretation	Context supports interpretation; it does not make unlike tasks, curricula or rubrics comparable.

4.2Revision Rounds extension

R-01revision_cycle_depth_distribution

Question	How deep was the supported revision opportunity?
Unit	Learner × assignment revision cycle, with assignment-level distribution.
Population and window	Learners assigned revision-enabled work, from first round opening through cycle end, withdrawal or study cutoff.
Calculation	Rounds offered/opened/completed, immutable drafts submitted and lifecycle outcome. Report the full 1, 2, 3 and 4+ draft distribution rather than only an average.
Exclude or separate	No revision offered, target not open, excused, withdrawn, integration unavailable and intentionally ended.
Interpretation	More rounds are not automatically better; repeated revision can become editing compliance.

R-02feedback_clarity_response_distribution

Question	Did the student report knowing what the feedback asked them to do?
Unit	Learner × selected feedback item × release version.
Population and window	Learners shown the checkpoint for a visible feedback item, before target submission, source-round/cycle end or study cutoff.
Calculation	Latest committed pre-target response: know what to do, not sure or ask teacher. Response coverage = committed responses ÷ eligible item-cases.
Deduplication	View, autosave, Back, close and timeout do not count; only an explicit commitment creates evidence.
Interpretation	Direct self-report, not a reading-comprehension score or proof. “I'm not sure” is useful evidence, not failure.

R-03clarification_request_and_resolution

Question	Did students seek clarification, and did a reply arrive in time for the intended revision?
Unit	One sent clarification thread bound to learner × feedback item × release version.
Population and window	Feedback viewers with a real authorized reply route, through target submission, cycle end or study cutoff.
Formulae	Request rate = learners sending at least one request ÷ learners with the ask path who viewed feedback. Response rate = threads with a teacher response ÷ sent threads. Timely response = threads answered before the relevant cutoff ÷ sent threads.
Exclude or separate	Composer opened but not sent, reply route unavailable and post-submission thread.
Interpretation	Fewer questions may reflect clarity, reluctance, access or another support route; silence is not automatically understanding.

R-04revision_priority_and_plan_rate

Question	Did the student choose a manageable focus and, where appropriate, make a deliberate plan?
Unit	Learner × source-target transition, with selected-item links beneath it.
Population and window	Learners with visible feedback and another draft explicitly intended, before target submission, pause/end or study cutoff.
Formulae	Priority-selection rate = learners selecting at least one item ÷ learners eligible to choose. Plan rate = learners explicitly saving a plan ÷ learners eligible to plan. Report valid no-plan and clarification dispositions separately.
Privacy	Plan prose is not included in the external interface.
Interpretation	Intention is not completion, correctness or learning. A valid choice not to plan is not failure.

R-05revision_participation_rate

Question	Did the student submit another draft during a genuine opportunity?
Unit	Learner × frozen source-target transition cohort.
Population and window	Learners with visible/renderable feedback and an opened target round, through close/grace cutoff.
Formula	Learners with at least one valid target submission ÷ eligible transition learners.
Exclude or separate	Pause/end, later round not yet opened, withdrawal, excusal, technical inability and unsupported submission.
Interpretation	Participation shows action, not the substance or cause of the revision.

R-06feedback_revision_traceability_coverage

Question	Can a change be connected to specific released feedback and, where present, the student's choice and plan?
Unit	Learner × actionable feedback item × source-target transition.
Population and window	Actionable items for learners with a valid target draft, through target analysis and linkage completion.
Calculation	Report coverage for source-target draft linkage, selected-priority linkage, plan/disposition linkage, locatable relevant evidence and complete chain. Each component uses its applicable denominator.
Provenance	Stable IDs are preferred; any inferred link is labelled with its method.
Interpretation	Traceability is a prerequisite for uptake, not evidence that uptake occurred.

R-07feedback_uptake_distribution

Question	How substantively did the target draft respond to a specific feedback item?
Unit	Learner × feedback item × source-target transition with stable traceability.
Population and window	Traceable item-transitions with inspectable source and target evidence, through the agreed review cutoff.
Calculation	Among sufficient-evidence cases, report no relevant change, surface change, partly addressed, substantively addressed and teacher-only different valid approach. Insufficient-evidence rate = insufficient cases ÷ all eligible traceable cases.
Required cascade	Eligible → feedback available → viewed → revision opportunity → revised → traceable → sufficient evidence → outcome.
Provenance	Individual teacher review, valid sampled review and unreviewed TA39 suggestion remain separate.
Interpretation	Immediate supported revision—not transfer or proof that TA39 caused change.

R-08targeted_criterion_evidence_change_distribution

Question	Did the target work show stronger evidence on the criterion the feedback was intended to address?
Unit	Learner × targeted criterion × source-target transition.
Population and window	Cases with an explicit construct, valid task/rubric alignment, genuine revision opportunity and sufficient comparable source and target evidence.
Calculation	Among independently rated sufficient cases, report stronger evidence, similar evidence, weaker evidence and different valid evidence. Insufficient-comparable-evidence rate = insufficient cases ÷ all otherwise eligible targeted-criterion transitions.
Required cascade	Eligible → feedback available → viewed → revision opportunity → revised → traceable → sufficient evidence → outcome.
Provenance	The research outcome uses an approved human final rating; any TA39 proposal remains separately labelled.
Interpretation	Immediate criterion-level evidence change within supported revision—not obedience to feedback, independent transfer or AI-score subtraction.

4.3Teacher evidence and instructional response

T-01teacher_evidence_review_coverage

Question	Did the teacher explicitly review the evidence?
Unit	One reviewable initial-submission or revision-transition evidence case.
Population and window	Cases with reviewable evidence and an available review workflow, through the review cutoff.
Calculation	States: individually reviewed, valid sampled review, explicitly deferred/skipped, unreviewed or unavailable. Review coverage = individual + valid sampled cases ÷ cases with reviewable evidence.
Deduplication	Only explicit confirmation or adjustment counts; view, scroll, navigation and timeout write nothing.
Interpretation	Coverage is not teacher quality. Corrections may demonstrate professional judgment.

T-02student_and_class_need_prevalence

Question	What individual or shared needs became visible?
Unit	Learner × skill × evidence episode; aggregate by class × assignment × skill and, for revision, transition.
Population and window	Learners with sufficient evidence for the skill after the selected evidence episode and review cutoff.
Formula	Need-present learners ÷ learners with sufficient evidence for the skill. Each learner counts once per snapshot; insufficient evidence remains visible outside the denominator.
Provenance	Individual teacher review, sampled review and unreviewed TA39 suggestion remain separate.
Interpretation	Identifies a pattern, not its cause. Consider teaching, feedback clarity, time/support, task design and rubric fit.

T-03instructional_decision_distribution

Question	What next teaching decision followed the reviewed evidence?
Unit	One reviewed need or class pattern × teacher decision version.
Population and window	Reviewed cases with an available response workflow, through the assignment/term decision cutoff.
Calculation	Clarify/reply, reteach, practice, another revision opportunity, later comparable opportunity, adjust task/rubric/template, show-what-you-know, no action needed, decide later or other approved category. Decision-recording rate = explicit decisions ÷ eligible reviewed cases.
Provenance	Explicit teacher choice; TA39 may preserve the evidence link but does not author the decision.
Interpretation	Shows what the teacher chose. Teacher report is needed to establish how the evidence contributed to that choice.

T-04follow_up_fulfillment_rate

Question	Did an intended instructional response become a real, linked opportunity?
Unit	One teacher decision marked follow-up intended.
Population and window	Decisions requiring creation, assignment or delivery, through the stated fulfillment cutoff.
Formula	Created or delivered linked opportunities ÷ intended follow-ups whose cutoff has passed or state is terminal. Modified, cancelled, pending and expired remain separate.
Deduplication	Stable decision and follow-up IDs.
Interpretation	Shows teacher action, not student participation, improvement or transfer.

4.4Deferred later learning

L-01linked_later_learning_distribution

Question	Did the learning continue in a later, comparable task?
Unit	Learner × skill × source evidence episode × teacher-linked later task.
Availability	Deferred until TA39 can preserve a stable skill link, versioned task/rubric/support context, a teacher-planned or teacher-reviewed comparability link, sufficient reviewed later evidence and appropriate longitudinal privacy rules.
Calculation when activated	Among sufficient later-evidence cases, report recurring, improving, sustained and resolved for now. Report insufficient evidence and no comparable work separately.
Interpretation	This is not demonstrable through repeated drafts of the same assignment. Revisions show supported uptake; a later comparable task is needed to examine transfer.

5.Context required with every measure

Every result carries:

numerator, denominator and eligible population;
category counts, exclusions and missing states;
exact assignment, source draft, feedback release, target draft and revision transition where applicable;
school, class, grade band, subject, task type, language and curriculum/framework where used;
rubric, template, method and relevant model/prompt versions;
evidence coverage and teacher-review depth;
support or co-intervention context where collected;
reporting period, time zone and versioned school-calendar rule; and
small-cell suppression and access controls for school- or district-level reporting.

Unavailable is not zero. Not offered, not applicable, no submission, excused, withdrawn, technical failure, insufficient evidence and no comparable later work remain distinct.

6.Minimal atomic record set

TA39 may maintain richer internal telemetry. The proposed external interface uses only atomic records from which the measures can be reconstructed. A feature that is unavailable produces no misleading zero-valued outcome.

Event name	Minimum structured content	Measures supported
assignment_context_recorded	Versioned assignment, school/class, grade, subject, task/language, curriculum/framework where used, skill mapping and rubric/template versions	All filters; A-09
assessment_opportunity_recorded	Authorized learner/cohort, assignment/round opportunity, eligibility and structured exceptional state	A-01, A-02 and denominators
draft_submitted	Immutable attempt, assignment/round, initial or revision role and time	A-02, A-03, R-01, R-05–R-08
starting_skill_evidence_recorded	Attempt, criterion, evidence category, sufficiency, method and review provenance	A-03
feedback_case_disposed	Source attempt, disposition, review/release version and time	A-04, A-05
feedback_intervention_reason_recorded	Case, structured reason or explicit no-reason state and taxonomy version	A-07
feedback_made_available	Immutable release, source draft, release/visibility times, item count, origin/review state and disclosure version	A-04–A-06; chain start
feedback_viewed	Release, learner, first qualifying view time and view-rule version	A-06
rubric_or_template_activity_recorded	Instrument/version, action and AI-optimization outcome	A-08
revision_opportunity_opened	Source/target rounds, learner/cohort, open/close times and checkpoint mode	R-01, R-04, R-05
feedback_comprehension_recorded	Item/release, three-state clarity response, response mode and version	R-02
clarification_status_recorded	Item/release, sent/replied timestamps, status and before-revision state—without message text	R-03
revision_plan_submitted	Selected item references, target transition, committed disposition and time—without plan prose	R-04, R-06
uptake_outcome_recorded	Item and draft links, sufficiency, proposed/final category, review mode, changed-from-proposal and method version	R-06, R-07, T-01, T-02
criterion_evidence_change_recorded	Target criterion, source/target attempts, comparable-evidence state, approved category and human-rating provenance	R-08
teacher_evidence_reviewed	Evidence case/type, explicit review act, mode, final state and time	T-01 and review provenance
instructional_decision_recorded	Need/pattern reference, teacher decision, audience, target skill and follow-up intention	T-03, T-04
follow_up_opportunity_created	Originating decision, audience, skill, type and linked opportunity/context	T-04; future L-01

6.1Common event fields

Every event carries a stable event identifier where supported, schema and event name, original occurrence time, relevant pseudonymous actor and learner references, school/class/assignment context, provenance, method/version, missing/not-applicable state and reporting authorization.

Corrections are append-only and reference the earlier record; they do not silently overwrite the fact originally reported. The final correction mechanism remains subject to EdTech Impact confirmation.

6.2Privacy boundary

The usage-event interface carries references, categories, counts, timestamps, versions and provenance. It does not carry:

student writing or answers;
teacher feedback prose;
student plan or question text;
teacher replies or notes;
rubric, template, prompt or source-material text;
names, email addresses, UPNs or raw internal account identifiers;
raw browser user-agent, page URL/title or unrestricted internal payloads; or
disciplinary, authorship or misconduct scores.

Identifiers are pseudonymous and stable only for the authorized research purpose. Delivery is limited to the authorized cohort and assignments.

7.Reports this contract supports

7.1Individual student learning loop

For an ordinary assessment, the report stops truthfully at source work, criterion-linked evidence, teacher-reviewed released feedback and qualifying view where available.

For Revision Rounds, it may additionally show the student's committed clarity response and priority, revision participation, traceability, teacher-reviewed uptake and independently rated targeted-criterion evidence change. Transfer remains not measured until a comparable later task exists.

7.2Class assessment and learning loop

Shows the eligible roster, real opportunity, submission state, feedback pipeline and evidence/review coverage. For a revision transition it also shows the complete denominator cascade, clarity/planning distributions, revision participation, traceability, uptake, targeted evidence change where validated and class needs by skill.

7.3Teacher response loop

Shows evidence reviewed, needs noticed, decisions recorded and intended follow-ups that became opportunities. It does not rank teachers by acceptance of TA39 suggestions.

7.4School or district overview

Aggregates the same measures by school, grade, subject, task type, language and period with full distributions, denominators, review coverage, comparability cautions, minimum-cell suppression and appropriate access controls.

8.Complementary evidence outside product telemetry

Product events cannot establish perceived value, trust, teacher voice, why a student acted or whether the evidence changed a teacher's thinking. Those questions require routed student/teacher questionnaires, interviews or independent review.

The complementary study model therefore includes:

student feedback experience among students who actually opened the feedback;
student-reported feedback influence among students with a genuine opportunity to act;
teacher-reported credibility, sufficiency, usefulness and contribution to the next decision;
reasons for selective or low use;
professional-judgment development through paired human-coded evidence; and
independent human validation of the final feedback students actually received.

These are research instruments, not ordinary usage-event fields. Raw qualitative responses and classroom content are not sent through the EdTech Impact usage API.

Appendix A

EdTech Impact transport position

A.1 Confirmed as of August 7, 2026

Item	Agreed position
Schema	Direct API schema version 1.1 replaces the earlier xAPI direction for this integration
Product	TA39 product_id is 4739; the guide types this field as a string, so examples use "4739"
Sender	TA39's backend is the only component that communicates with EdTech Impact
Delivery quantum	One atomic event per relevant occurrence; weekly/monthly measures may be derived from those events
Required envelope	schema_version, event_name, product_id, school_id, actor_id and occurred_at
Enrichment	Structured actor, activity, result, context and attributes fields may be supplied
Raw payload	TA39 may omit raw
Timing	Events retain the original occurred_at and may arrive late or out of order
Identity	Teacher/student IDs are pseudonymous and stable; no names, email addresses, UPNs or raw account IDs
School mapping	TA39 supplies consistent school IDs during onboarding
Linkage	Privacy-safe assignment, submission, feedback and revision identifiers may be sent under attributes
Content boundary	No writing, feedback, rubric prose, plan text, student question, other free text, result.response, client.user_agent or page information
Research boundary	Only authorized Cycle 2 cohorts and assignments
Activation	External delivery remains disabled until cohort and connection confirmation
Validation	Internal mock receiver first, then EdTech Impact sandbox, before production

A.2 Pending EdTech Impact implementation or confirmation

These items must not be inferred from the guide or treated as production-ready:

sandbox/production endpoints and authentication;
final representation when the actor performs an action concerning another learner;
acceptance and mapping of class_roster_observed events;
event_id as the receiver's idempotency key;
response body, accepted/rejected event identification and retry semantics;
correction, supersession and deletion-request handling; and
reconciliation visibility and the final mapping of TA39 event names to EdTech Impact measures.

The examples below show TA39's proposed representation for joint sandbox validation. They do not resolve the pending items.

Appendix B

Representative schema 1.1 API output

B.1 Design of the payloads

Each example:

uses the required schema 1.1 envelope;
uses product_id as the string "4739";
omits raw, result.response, client.user_agent and page fields;
uses pseudonymous IDs;
sends categories and references rather than classroom prose; and
preserves the original occurred_at.

event_id appears because it is present in the guide's enriched example and is TA39's proposed stable retry identifier. Its idempotency semantics remain pending.

B.2 Ordinary assessment: student submits work

{
  "schema_version": "1.1",
  "event_id": "6e81e4a0-9ac7-4e3a-8d6a-0be9ca2ab101",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_48f1d2",
  "event_name": "draft_submitted",
  "occurred_at": "2026-09-15T07:42:18Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "student",
    "country": "QA",
    "grade_index": 8,
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "sub_psn_b472",
    "type": "assessment_submission",
    "subject": "english",
    "attempt": 1,
    "parent_id": "asg_psn_a913",
    "assigned_by": "usr_psn_t221"
  },
  "result": {
    "success": true
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913"
  },
  "attributes": {
    "assessment_mode": "ordinary",
    "submission_role": "initial",
    "language": "en",
    "task_type": "argument_writing",
    "submission_id": "sub_psn_b472",
    "attempt_id": "att_psn_001",
    "rubric_version": "rub_psn_09_v3"
  }
}

This supports submission opportunity, activity and assessment-context reporting. The student's work is not transmitted.

B.3 Teacher reviews and edits AI-prepared feedback

{
  "schema_version": "1.1",
  "event_id": "2a238b66-5803-45d8-99d6-3c29ee462202",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_t221",
  "event_name": "feedback_case_disposed",
  "occurred_at": "2026-09-15T10:18:43Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "teacher",
    "country": "QA",
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "fbc_psn_772a",
    "type": "feedback_review",
    "subject": "english",
    "ai": true,
    "parent_id": "asg_psn_a913"
  },
  "result": {
    "success": true,
    "duration_ms": 428000
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913"
  },
  "attributes": {
    "learner_id": "usr_psn_48f1d2",
    "source_attempt_id": "att_psn_001",
    "feedback_case_id": "fbc_psn_772a",
    "feedback_version": 1,
    "disposition": "edited",
    "reason_response_state": "provided",
    "reason_categories": [
      "actionability_or_clarity",
      "teacher_voice"
    ],
    "reason_taxonomy_version": "1.1",
    "rubric_version": "rub_psn_09_v3",
    "template_version": "tpl_psn_04_v2"
  }
}

learner_id under attributes represents the learner concerned by a teacher action. This mapping remains subject to EdTech Impact confirmation. No feedback prose or free-text edit reason is sent.

B.4 Teacher releases an immutable feedback version

{
  "schema_version": "1.1",
  "event_id": "3af43820-226d-44ba-a8d8-4c6aa7e1f303",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_t221",
  "event_name": "feedback_made_available",
  "occurred_at": "2026-09-15T10:20:11Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "teacher",
    "country": "QA",
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "rel_psn_88c3",
    "type": "feedback_release",
    "subject": "english",
    "ai": true,
    "parent_id": "asg_psn_a913"
  },
  "result": {
    "success": true
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913"
  },
  "attributes": {
    "learner_id": "usr_psn_48f1d2",
    "source_attempt_id": "att_psn_001",
    "feedback_case_id": "fbc_psn_772a",
    "feedback_release_id": "rel_psn_88c3",
    "release_version": 1,
    "item_count": 3,
    "origin_state": "ai_assisted",
    "review_state": "reviewed_and_released",
    "student_visible_at": "2026-09-15T10:20:11Z",
    "role_disclosure_version": "1.0"
  }
}

This creates the start of the feedback-to-action evidence chain. It says what state was released and when; it does not transmit what the feedback said.

B.5 Revision Rounds: student reports clarity

{
  "schema_version": "1.1",
  "event_id": "f9355796-efc8-4f37-9fc2-c9a31c9bd404",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_48f1d2",
  "event_name": "feedback_comprehension_recorded",
  "occurred_at": "2026-09-15T14:07:32Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "student",
    "country": "QA",
    "grade_index": 8,
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "fbi_psn_31",
    "type": "feedback_clarity_checkpoint",
    "subject": "english",
    "parent_id": "rel_psn_88c3"
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913",
    "learning_path_id": "rvc_psn_201"
  },
  "attributes": {
    "feedback_release_id": "rel_psn_88c3",
    "feedback_item_id": "fbi_psn_31",
    "source_round_id": "rnd_psn_01",
    "target_round_id": "rnd_psn_02",
    "clarity_response": "not_sure",
    "checkpoint_mode": "recommended",
    "response_mode": "typed",
    "response_version": 1
  }
}

The category is a direct student report. No student question or plan prose is sent, and not_sure is not coded as failure.

B.6 Revision Rounds: student commits a priority and plan

{
  "schema_version": "1.1",
  "event_id": "4b8911b2-3618-4cbc-84a2-4f23c913d505",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_48f1d2",
  "event_name": "revision_plan_submitted",
  "occurred_at": "2026-09-16T06:58:09Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "student",
    "country": "QA",
    "grade_index": 8,
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "pln_psn_641",
    "type": "revision_planning",
    "subject": "english",
    "parent_id": "rvc_psn_201"
  },
  "result": {
    "success": true
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913",
    "learning_path_id": "rvc_psn_201"
  },
  "attributes": {
    "feedback_release_id": "rel_psn_88c3",
    "selected_feedback_item_ids": [
      "fbi_psn_31"
    ],
    "priority_count": 1,
    "source_round_id": "rnd_psn_01",
    "target_round_id": "rnd_psn_02",
    "plan_state": "saved",
    "plan_version": 1,
    "plan_text_transmitted": false
  }
}

This establishes intention and traceability. It does not reveal the plan and does not establish that the student completed or benefited from it.

B.7 Teacher confirms a feedback-linked revision outcome

{
  "schema_version": "1.1",
  "event_id": "de4aa960-f445-45f8-886b-93815025c606",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_t221",
  "event_name": "uptake_outcome_recorded",
  "occurred_at": "2026-09-20T11:36:55Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "teacher",
    "country": "QA",
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "upk_psn_591",
    "type": "revision_evidence_review",
    "subject": "english",
    "parent_id": "asg_psn_a913"
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913",
    "learning_path_id": "rvc_psn_201"
  },
  "attributes": {
    "learner_id": "usr_psn_48f1d2",
    "feedback_release_id": "rel_psn_88c3",
    "feedback_item_id": "fbi_psn_31",
    "source_attempt_id": "att_psn_001",
    "target_attempt_id": "att_psn_002",
    "selected_priority_linked": true,
    "plan_linked": true,
    "evidence_sufficiency": "sufficient",
    "ta39_suggested_outcome": "partly_addressed",
    "teacher_final_outcome": "partly_addressed",
    "review_mode": "individual",
    "changed_from_suggestion": false,
    "method_version": "uptake_method_1.1"
  }
}

The review provenance is explicit. The category is about the work's response to a feedback item; it is not a grade, a transfer claim or proof of causal impact.

B.8 Teacher records the next instructional decision

{
  "schema_version": "1.1",
  "event_id": "aa8bb82f-b346-424f-a602-56f7571eb707",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "usr_psn_t221",
  "event_name": "instructional_decision_recorded",
  "occurred_at": "2026-09-20T11:44:16Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "teacher",
    "country": "QA",
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "dec_psn_812",
    "type": "instructional_decision",
    "subject": "english",
    "parent_id": "asg_psn_a913"
  },
  "context": {
    "course_id": "crs_psn_31e2",
    "assignment_id": "asg_psn_a913"
  },
  "attributes": {
    "origin_evidence_id": "pat_psn_117",
    "criterion_id": "crt_psn_evidence_explanation",
    "decision_category": "provide_practice",
    "audience_category": "selected_students",
    "audience_count": 5,
    "follow_up_intended": true,
    "decision_taxonomy_version": "1.1",
    "teacher_authored": true
  }
}

This records that the teacher chose a response. It does not send the teaching instructions, assert that the evidence caused the choice or claim that the follow-up was effective.

B.9 Pending receiver mapping: roster observation

The following proposed event is required to construct an honest zero-submission denominator, but EdTech Impact's acceptance and actor/learner mapping remain pending:

{
  "schema_version": "1.1",
  "event_id": "12da8c67-af4a-48a7-98bf-3fbaecf1b808",
  "product_id": "4739",
  "school_id": "sch_qf_001",
  "actor_id": "sys_ta39_roster_sync",
  "event_name": "class_roster_observed",
  "occurred_at": "2026-09-10T05:10:00Z",
  "event_source": "ta39_api",
  "product_version": "2026.09",
  "actor": {
    "role": "system",
    "country": "QA",
    "class_id": "cls_psn_7b31",
    "cohort_id": "wise_cycle_2"
  },
  "activity": {
    "id": "rst_psn_91c",
    "type": "roster_observation",
    "parent_id": "cls_psn_7b31"
  },
  "result": {
    "success": true
  },
  "context": {
    "course_id": "crs_psn_31e2"
  },
  "attributes": {
    "learner_id": "usr_psn_48f1d2",
    "membership_status": "active",
    "roster_source": "google_classroom",
    "observation_version": 1
  }
}

This event contains no name or email address. It must remain disabled until EdTech Impact confirms the representation and mapping.

Appendix C

How atomic events become measures

TA39 proposes sending atomic events rather than pre-calculated weekly or monthly aggregates. A measure is calculated only after applying its eligibility, window, deduplication, exclusions, filters and provenance rules.

C.1 Example: revision participation

For one class × assignment × Draft 1 → Draft 2 transition:

Start from the frozen eligible roster/opportunity records.
Retain learners whose feedback became visible and renderable.
Retain learners for whom Draft 2 was genuinely opened long enough to submit.
Exclude excused, withdrawn and verified technical-inability cases while reporting them separately.
Count each learner with at least one valid Draft 2 submission once.

If 30 learners were assigned, 28 received visible feedback, 27 had a genuine Draft 2 opportunity, one was excused and 23 submitted:

R-05 revision participation = 23 ÷ 26 = 88.5%

The report also states that one learner was excused, one did not receive visible feedback and one never received an open Draft 2 opportunity. Those cases are not silently coded as non-participation.

C.2 Example: uptake and targeted evidence change

For the same cohort, the report does not begin with the final analyzable cases. It shows:

Cascade stage	Learners
Eligible for the assignment	30
Feedback became visible	28
Feedback viewed	26
Genuine revision opportunity	26
Submitted the revision	23
Feedback-to-revision chain traceable	21
Sufficient evidence for uptake classification	19
Teacher-reviewed uptake outcome	15
Independently rated comparable criterion evidence	12

R-07 distributions use the 19 sufficient uptake cases and disclose review depth. R-08 uses only the 12 independently rated comparable criterion cases. Neither percentage is presented as an outcome for all 30 students.

C.3 Reconciliation

During sandbox validation, TA39 and EdTech Impact should confirm that:

every accepted event is counted once;
late and out-of-order delivery preserves occurred_at;
the same school, class, assignment and pseudonymous actor mappings are used;
exclusions and missing states remain distinct;
category totals reconcile to eligible denominators; and
any receiver-derived metric reproduces this dictionary's calculation and version.

If EdTech Impact requires a pre-calculated measure snapshot, TA39 can provide one for reconciliation with metric version, category counts, numerator, denominator, eligible/excluded counts, coverage, period, filters and provenance. Atomic events remain the primary source of record.