Data Provenance and Processing Details

Official source inventory

The 2024-25 assessment release contract identifies 14 versioned workbook roles: six each for ELA and Math and two for Science. Each role records its official URL, expected sheet and columns, row and suppression counts, and cryptographic hashes. This allows a future ODE release to fail visibly when its structure changes instead of being silently forced through an old parser.

Assessment Group Reports landing page

• U.S. Census geographies and ACS 5-year neighborhood estimates, including income, education, poverty, and employment measures.
• NCES EDGE school identities and locale classifications: city, suburb, town, or rural.

Assessment assembly and suppression

• Workbook rows are joined and checked on the full assessment identity: academic year, district ID, school ID, subject, student group, and grade level. Names are retained for review, not substituted for those identifiers.
• Overlapping rows from workbook families must agree. A new student-group row is added only when its full key is absent; a shared row is a validation case, not an invitation to overwrite one source with another.
• ODE's suppressed, no-test, and otherwise nonnumeric cells are counted in source audits but are not converted to zero or imputed in dashboard data.
• ELA and Math All Grades values come from the official combined workbooks. The system does not average grade percentages or rebuild an All Grades result from only the visible grade rows. That reconstruction can be wrong when one or more grade components are suppressed.
• Science All Grades rows come from the official Science student-group workbooks; ODE did not publish a separate Science All Grades combined workbook analogous to ELA and Math.

Achievement calculations and student weighting

For each reported assessment row, the Scored Performance Denominator is the sum of students in performance Levels 1 through 4. The proficient count is Levels 3 plus 4, and percent proficient is that count divided by the scored denominator. These identities are checked row by row.

• Student-weighted summaries weight each school's achievement rate by its scored denominator. Only students represented in an ODE score distribution contribute to that achievement summary.
• Number of Participants can differ from the scored denominator and is used for participation, coverage, and source-audit questions.
• When the dashboard collapses schools into one district, city, or statewide point, counts are summed before the rate is calculated. It does not give a small school the same weight as a large school.
• The explicit-grade view preserves separate school-grade observations. The All Grades view uses one official combined observation for each available school and selected student group, avoiding repeated-school points and suppression-biased synthetic rollups.

Attendance is aligned to the tested population

Attendance is attached only when the ODE attendance population corresponds to the assessment population. The dashboard never substitutes a whole-school attendance rate merely because it is available.

• A Total Population grade row receives the exact school-and-grade attendance record.
• A Total Population All Grades row receives a count-weighted tested-grade rollup: grades 3-8 and 11 for ELA and Math; grades 5, 8, and 11 for Science.
• The rollup uses Attendance_Students_Included and the reported regular-attender and chronic-absence counts; its percentages are recomputed from those counts.
• If a required component is suppressed, missing, duplicated, or internally inconsistent, attendance is left blank. There is no whole-school fallback.
• The attendance source does not supply grade-matched attendance for assessment subgroups. Subgroup assessment rows therefore have blank attendance; those blanks protect the population definition and are not missing-data defects.
• Attendance_Group_Source records whether a value came from an exact grade match or a tested_grade_rollup.

Context, geography, and school-policy enrichment

• School addresses and reviewed coordinates are linked to Census tract and block-group geography, then to ACS neighborhood measures. These describe the school's surrounding area, not the family circumstances of each tested student.
• NCES identities and locales are reviewed against the complete ELA, Math, and Science school union. For virtual and some alternative programs, locale describes an administrative location and may not describe where enrolled students live.
• Class size, spending, and ODE report-card fields are joined on exact district-and-school identifiers. Ambiguous or unmatched identities fail closed; the current pipeline does not silently fall back to a school ID from another district.
• For dashboard display, an ODE interval such as “under 5%” may be shown at its midpoint and is marked as censored. Evidence Lab models omit censored or unavailable attendance rather than treating that midpoint as exact.
• The official All Grades sources add 19 newly represented schools: 13 physical schools with reviewed place context and six official virtual schools. Place-based context is deliberately blank for those virtual schools.
• Charter status, virtual status, and special-enrollment design remain separate dimensions. Effective-dated evidence for online, open-enrollment, immersion, lottery, and other non-neighborhood programs is maintained in data/lookup/special_enrollment_policy.csv and data/lookup/special_enrollment_policy_sources.csv; unresolved candidates remain in data/lookup/special_enrollment_review_candidates.csv. School-type and virtual corrections are separately recorded in school_type_overrides.csv and virtual_online_exclusions.csv.

Reviewed school-location corrections

A 2026 location audit compared processed records with ODE, NCES, district, geocoding, and official Census coordinate evidence. Reviewed corrections were applied consistently across subjects, and the NCES crosswalk was rebuilt against the complete school union. The resulting review found no unresolved material cross-subject location conflicts among ordinary physical schools. Location corrections change geography and linked ACS measures; they do not change assessment counts or performance levels.

Audited 2024-25 candidate inventory

These are row-level assessment records, not counts of unique students. A school appears on many rows because ODE reports grades and student groups separately. Attendance blanks mostly occur on student-group rows where the attendance population is not comparable.

The complete candidate audit found no duplicate assessment keys, no achievement-field mismatches, no denominator failures, and no attendance-source mismatches in these files.

Known limits and interpretation boundaries

• Suppression: ODE withholds some small-cell results. The dashboard cannot recover those values, and visible components may not reproduce an official combined result.
• Neighborhood context: ACS tract and block-group measures are contextual proxies, not student records and not proof of a causal mechanism.
• Timing: current spending metadata describes 2023-24 while primary outcomes describe 2024-25. Comparisons involving spending therefore have a one-year timing mismatch.
• Special programs: virtual, charter, magnet, immersion, open-enrollment, and alternative programs may serve students from outside the surrounding neighborhood. The dashboard provides separate controls rather than treating these categories as interchangeable.
• Older data: the Simulation Lab's 2018-19 Math option is a legacy comparison profile with fewer fields. Its 2,302 complete rows and 188,837 scored students passed the checks required for that simulation, but this is not a comprehensive certification of every historical data bundle or experiment.
• Inference: correlations, regression adjustments, residuals, and simulations are descriptive or model-based unless a page explicitly establishes a causal design. School-grade changes are not individual student growth.

Reproducibility and release gates

• The release contract pins source roles, workbook structure, counts, suppression patterns, and both file and semantic workbook-package hashes. This distinguishes a harmless spreadsheet-metadata change from a substantive source change.
• The transformation runs in a non-promoting mode first. It writes isolated candidates and comparison evidence rather than overwriting canonical or deployable data.
• The full 2024-25 dry run reproduced 33,960 ELA, 32,005 Math, and 14,916 Science rows, matched four independent evidence hashes, and left all 57 protected canonical and site files unchanged.
• Tests cover source lineage, schema and key uniqueness, achievement identities, attendance matching, dashboard calculations, activated controls, public wording, and selected rendered results.
• Promotion, site mirroring, and publication are separate steps. Candidate generation or a local commit does not by itself publish data.