{"resourceId":"psu-gates-benchmark-ai-feedback-student-reasoning-claude-20261109","versions":[{"version":"external-23ada14a25958fac79d7ed037512748613e7a32a2b3d4884db88fdab838f4411","resource":{"id":"psu-gates-benchmark-ai-feedback-student-reasoning-claude-20261109","title":"Portland State $2M Gates/Learning Commons project to benchmark how Claude and other models give feedback on grades 6-12 data reasoning","organization":"Portland State University","sector":"Higher education research / K-12","geography":"Oregon, United States","publishedAt":"November 9th 2026","publicationDate":"2026-11-09","eventDate":"2026-11-09","sourceName":"Portland State University News","sourceLabel":"University news release on funded research project","sourceUrl":"https://www.pdx.edu/news/how-good-ai-guiding-student-thinking-2m-grant-helps-psu-researchers-find-out","evidenceClass":"academic-research","outcomeClass":"emerging","topics":["knowledge-work","accessibility-workforce","governance-procurement"],"finding":"Portland State University researchers received a two-year, $2 million Gates Foundation and Learning Commons grant to build a benchmark for how AI systems, explicitly including Claude and GPT-6 Astra, respond to grades 6-12 students' data-reasoning work. The benchmark will use 1,200 de-identified iterative samples from partner districts and score whether models interpret student reasoning, choose an appropriate instructional move and preserve the student's thinking. Researchers say it will give districts an independent way to examine capabilities before adoption. No results yet.","sledRelevance":"Interpretation: K-12 districts considering Claude for Teachers or student-facing tutoring will eventually have an independent, classroom-grounded benchmark rather than vendor claims; until then, this is a signal of evaluation capacity, not evidence of effectiveness. The focus on linguistically diverse students matters for equity reviews.","evidence":"University release describing funded design, data volume and evaluation dimensions; no outcomes.","architectureImplications":"Interpretation: none yet; districts may later integrate the benchmark into procurement evaluation.","governanceImplications":"Interpretation: district AI committees can cite this as an emerging independent evaluation source and ask vendors to participate.","securityPrivacyImplications":"Student data will be stripped of identity per the release; districts sharing data need agreements.","caveats":"Announcement of planned research; no results; partner districts and model versions not named; release is from the university itself. Cross-tagged k12 (benchmark built from district student work) and research (university-led evaluation infrastructure).","streamIds":["anthropic","k12","research"],"roles":{"sales":"Frame this as a future evidence source for district buyers. Portland State is building a $2M Gates-funded benchmark that will test how Claude and other models scaffold student reasoning rather than give answers. Ask districts how they currently evaluate tutoring or feedback tools and whether equity for English learners is part of review. Offer to help define evaluation criteria. Do not claim Claude performs well; there are no results yet.","engineering":"Engineering relevance is evaluation design. The benchmark dimensions (interpreting reasoning, choosing next move, preserving student work, adapting to new understanding) can inform district pilot rubrics now. Prerequisites: de-identified student work samples and teacher-scored feedback. Constraint: no published results or model versions. Proof of value: a small district rubric pilot comparing model feedback against teacher feedback on the same student work.","delivery":"Delivery teams supporting districts can align pilot evaluation plans with these dimensions and plan to re-test when the benchmark is released. Dependencies are teacher time and data-sharing agreements. Ownership sits with district curriculum and instructional technology leads. Risks include adopting tools before independent evidence exists and evaluations that ignore English learners. Acceptance would be rubric scores collected for every pilot tool, with subgroup results for English learners reported."},"roleDetails":{"sales":{"customerProblem":"Districts lack independent evidence on whether AI feedback supports student thinking or just gives answers.","stakeholders":["District curriculum director","Instructional technology lead","EL/multilingual program lead","Teachers"],"discoveryQuestions":["How do you evaluate AI feedback tools today?","Are English learners part of your evaluation?","Would you participate in independent benchmarks?"],"valueHypothesis":"Aligning pilots to independent evaluation dimensions improves adoption decisions.","potentialEngagement":"Evaluation-rubric design for a district AI feedback pilot.","unsupportedClaims":["Claude scaffolds better than other models","Any learning gains","Benchmark availability date"]},"engineering":{"fit":"Evaluation framework input; no product capability evidence.","architectureIntegration":"Future benchmark could plug into district procurement testing.","prerequisites":["De-identified student work","Teacher-scored feedback"],"constraints":["No results yet","Model versions unnamed"],"security":"De-identification of student data required.","proofOfValue":["Rubric pilot comparing AI and teacher feedback","Subgroup analysis for English learners"]},"delivery":{"work":"Align district pilots with benchmark dimensions; re-test on release.","dependencies":["Teacher time","Data-sharing agreements"],"ownership":"District curriculum and instructional technology leads.","skillsAdoption":"Teacher training on evaluating AI feedback quality.","governanceCheckpoints":["Student data privacy review","Equity review"],"acceptanceCriteria":["Proposed: rubric scores collected for each pilot tool","Proposed: subgroup results reported"],"risks":["Premature adoption","Ignoring English learners"]}},"retrievedAt":"2026-11-10T04:09:34Z","enrichedAt":"2026-11-10T04:18:20Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"The literacy component asks whether AI feedback recognizes sound reasoning not expressed in polished English, relevant to English learners.","procurementImplications":"Interpretation: future RFPs could require benchmark results on student-reasoning feedback.","operatingModelImplications":"Interpretation: teachers remain central; the project frames AI as support, not replacement.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://www.pdx.edu/news/how-good-ai-guiding-student-thinking-2m-grant-helps-psu-researchers-find-out","referenceExcerpt":"How well does Claude give scaffolding versus blurting out the answer?","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}