Helping school systems make measurement and assessment work for everyone they serve.
August 2026
Responsible AI & Assessment
Measure the Capabilities Students Will Need
When a Score Shapes Opportunity, Show the Work
A claim about responsible AI becomes actionable when a school leader can inspect the evidence behind it. The Duolingo English Test Responsible AI Transparency Report states the standard, reports performance against it over time, and leaves room for human review. The companion flipbook lays out a path through validity, reliability, fairness, security, quality assurance, and reporting. Take it into procurement and ask: What does the system claim? What supports that claim? Where can people intervene or appeal? Read the report and the paper trail here.
AI changes the assessment question. Alongside knowledge, what human capabilities should a school system recognize, develop, and report? A new Stanford Accelerator for Learning and ETS white paper explores richer evidence of learning while keeping validity, fairness, transparency, and trust in view. Use it with teams deciding which capabilities belong in the system and what evidence would let them recognize those capabilities. Read the flipbook. Download the report.
A District AI Policy Has to Work Before Monday
A district policy becomes real when a teacher, student, or family can use it in an ordinary week. The Study Group guide brings EDSAFE AI Alliance resources together so teams can define boundaries, explain expectations, protect data, and name the human who remains accountable when questions or safety concerns surface. It carries policy into the places leaders manage every week: assignments, disclosure, procurement, staff practice, and review. Read the flipbook.
Innovating FOR WHOM we measure
A measure can collect more information and still miss the people who need its evidence. Innovating for Whom We Measure asks teams to start with the people who will use it: learners, educators, families, communities, and public decision-makers. First clarify the decision. Then define the evidence that would help. Only then choose or build an instrument. That order gives teams an assessment they can use and defend.
Design Assessment for Its Users
In this webinar, Linda Darling-Hammond, John Hattie, Paul LeMahieu, and Elizabeth Mokyr Horner discuss multimodal AI alongside the sciences of learning, development, measurement, and improvement. The recording gives teams a way to judge new possibilities by the decisions and feedback they make possible. Access the webinar.
Ten Mind Frames for Better Assessment Practices
Assessment can satisfy an accountability requirement and still leave a teacher unsure what to do next. John Hattie, Stephen G. Sireci, and Eva L. Baker offer 10 "mind frames" for changing how leaders interpret and use evidence. The case study points assessment toward student growth and the next meaningful action. Reporting stays connected to learning and the action that follows. Download it here.
Turn Evidence Into Instruction
Compliance can end with a record. A learning-centered assessment system carries evidence into the work of teaching. This Achievement Network case study uses district examples to show how aligned assessment, clear evidence, and professional learning help educators adjust instruction while students can still benefit. It gives system leaders a way to see what happens between the assessment calendar and the decisions made in classrooms. Download it here.
Disability, Learner Variation, and Assessment
Make the Route to Evidence Part of the Design
An assessment can miss a learner when the task adds a barrier unrelated to the learning claim. Access by Design gathers 10 resources for district leaders on accommodations, universal design, procurement, and technology decisions. It helps teams name the learning claim, locate the barrier, and protect reliable ways for every learner to show what they know. The learning claim stays clear; the route to evidence becomes more usable. Read the flipbook.
Prioritizing Students with Disabilities in District AI Policy
Students with disabilities already use AI for school-related work, while safeguards around those systems remain unsettled. The brief "Prioritizing Students with Disabilities in AI Policy V2" brings disability rights, bias, accessibility, and human accountability into AI policy. The Study Group contributed to and signed the paper. It asks school leaders to test whether an AI policy protects the learners most affected by a system's errors and omissions. Read the full brief here.
Reading Support Cannot Stop in Third Grade
Older students can carry foundational reading gaps into classes built on the assumption that the problem has been solved. This case study pairs Stanford's ROAR with Achievement Network's implementation expertise and investments from Magpie Literacy and AERDF to support a validated foundational-skills assessment for older readers. Identify the need, then connect the signal to support while the student is still in secondary school. Download it here.
High-Value Learning, High-Quality Assessment
Dear Colleague Letter: The Screentime Debate Needs a Child at the Center
Before a district counts minutes, it should start with the learner: a rural student may reach advanced coursework; a student with dyslexia may practice decoding; a homebound student may stay connected to class. The Department of Education's Dear Colleague Letter asks whether technology improves learning and academic outcomes. The question is what the screen makes possible for the student. This flipbook helps leaders carry that question through selection, implementation, evaluation, and renewal. Read the flipbook. Check out the Department’s Dear Colleague Letter.
Managing Technology-Based Assessment
AI can expand what an assessment observes. The meaning still depends on the inference a human team can defend. The ITC/ATP Guidelines for Technology-Based Assessment help practitioners examine validity, fairness, privacy, transparency, and appropriate use across design, scoring, reporting, and policy. Use them to test new tools: more visible student work helps only when the inference drawn from it is defensible. Read more here.
