DDI Stack Data Model

What it is

The DDI (data-driven instruction) stack turns internal Illuminate assessment results into the dashboards, report-card feeds and planning sheets KTAF's NJ regions run their assessment cycle on. Achievement directors enter assessments in Illuminate and tag them in an AppSheet app; teachers administer and score them; the stack fans the results out to:

• the DDI Suite Tableau workbook (two extracts: the assessment dashboard and the weekly DDI dashboard),
• the assessment entry audit (on the Data Quality Dashboard) and the sight words dashboard (on the Literacy Dashboard),
• four DeansList extracts that put scores on report cards and progress reports,
• two Google Sheets (the NJ DDI tier roster and the DeansList mod audit),
• the AppSheet catalog feed that closes the tagging loop, and
• the Illuminate branch of the assessment Cube (the star schema behind student_assessment_scores_view).

"Internal assessment" here means an Illuminate assessment someone tagged in the AppSheet — the tag is literally what makes it internal (see Terms). The stack covers the three NJ regions. Miami's final year on it is SY25-26; Miami returns when Focus assessment data is ingested, and there is no interim exception.

Owner: Anthony Walters, Director, Data. Marya Shukla owns assessment-tagging quality control in the AppSheet and the standard-domains lookup; achievement directors enter their own assessments.

How it fits together

Illuminate DnA (dlt sync)        AppSheet tagging app          Assessments
27 stg_illuminate__* tables      (Marya Shukla + ADs)          lookup sheet
+ 46 repository_* stubs               |    ^                   (5 named ranges)
       |                              v    |                        |
       |               stg_google_appsheet__illuminate_             |
       |               assessments_extension                        |
       |                              |        rpt_appsheet__       |
       v                              v        assessments ---------+--> AppSheet
int_illuminate__assessments --> int_assessments__assessments_members    (catalog
       |                              |                                  feed)
       |                              v
int_illuminate__agg_student_    int_assessments__assessments_canonical
responses (overall/standard/          |
group)                                v
       |                int_assessments__scaffold  <-- course enrollments,
       |                 ("expected to take")          Illuminate sessions
       v                              |
int_assessments__response_rollup  <---+  <-- performance bands, reporting terms
       |         |          |         |
       v         v          v         v
DDI Suite    DeansList   mod audit   assessment star (dim_assessments,
(2 Tableau   extracts    sheet       administrations, bridges,
extracts)    (4 CDO      + tier      fct_assessment_scores_*) --> Cube view
             feeds)      roster

int_illuminate__repository_data (custom student-data tables) feeds the sight words dashboard and the DeansList sight-words extract on a separate track from the response rollup. On the star side, int_assessments__score_anchors and int_assessments__resolved_section_enrollments sit between the scaffold and fct_assessment_scores_enrollment_scoped, tying each score to one course section.

Materialization and freshness

Every rpt_* extract in the family is a view. Under them, the staging layer is all tables (the 27 stg_illuminate__* models, the repository stubs, the sheet staging models, and the AppSheet staging table), as are int_assessments__response_rollup, __scaffold, __course_enrollments, __resolved_section_enrollments, __score_anchors, int_illuminate__root_standards, and the table marts of the assessment star. The remaining intermediates (int_illuminate__* apart from root_standards; int_assessments__assessments_members, __assessments_canonical, __performance_bands, __academic_goals) and dim_assessment_goals are views. A new score therefore reaches a consumer only after:

1. The Illuminate dlt sync lands it. The assessment tables (agg_student_responses*, students_assessments) sync at midnight and 5pm daily, plus Wednesday 10am/2pm and Friday 3pm. Reference tables (performance bands, reporting groups, repositories) sync at midnight only — the root of #5399: same-day scores on a brand-new assessment can sit unscored until the next midnight.
2. The assessment star rebuilds. int_assessments__response_rollup, the other table intermediates and the table marts share one cron tick, 0 0,10,13,15,17 * * * (5x/day, Eastern). The sync and the star both fire at midnight and 5pm, so data from the same-hour sync can miss that star tick and wait for the next one — check run timestamps before promising users a refresh time.
3. The consumer refreshes: the DDI Suite Tableau extracts at 1am and 6pm daily plus Friday 4pm (exposure ddi_suite cron, Dagster-owned); the DeansList SFTP extracts at 1:25am daily; Cube's proficiency_rollup pre-aggregation within about 30 minutes plus one partition build of each scores-fact rebuild, for the quarterly partitions from July 1 of the previous school year on. Older partitions are frozen and need a manual rebuild in Cube Cloud after a historical backfill.

Terms

• Internal assessment (is_internal_assessment; "normed" in the workbook, surfaced as is_normed_scope): an Illuminate assessment with a row in the AppSheet Illuminate Assessments Extension. The flag is if(iae.assessment_id is not null, ...) in int_assessments__assessments_members. Untagged assessments still flow through the scaffold's non-internal branch into rpt_tableau__assessment_dashboard, so the Assessment Dashboard worksheet can analyze every Illuminate assessment — but the Module Dashboard and DKI worksheets display only normed (tagged) assessments, and the canonical grouping, the report-card feeds and the star's internal branch use only tagged ones. An untagged assessment is otherwise invisible to every consumer below except the raw Illuminate intermediates.
• Canonical assessment (canonical_assessment_id): internal assessments are created once per region/variant in Illuminate, so members that share (academic_year, scope, subject_area, module_code, grade_level_id) are grouped, and the lowest member assessment_id names the group. Miami-only Florida copies (AppSheet regions_assessed exactly Miami) group separately, so New Jersey rows never inherit a Florida title or date. Non-internal assessments are their own canonical. Known defect: inconsistent SY25-26 tagging destabilizes the grouping (#5654).
• Scope: Illuminate's assessment category (decoded from dna_scopes) — Unit Assessment ("UA"), Cumulative Review Quizzes ("CRQ"), Cold Read Quizzes, Sight Words Quiz, and so on. K-1 cumulative-review and cold-read quizzes are relabeled Checkpoint in int_illuminate__assessments. WPP is a module_type value excluded from the DDI dashboard; its expansion is unwritten — owner to confirm.
• Module (module_type, module_sequence, module_code): the AppSheet tagging that sequences assessments within a scope; module_code is the concatenation (for example UA3). These have no Illuminate-native counterpart.
• academic_year vs academic_year_clean: Illuminate dates a school year by its spring (2026 for SY25-26); the warehouse dates it by its fall (2025). academic_year_clean = academic_year - 1 converts, and int_assessments__course_enrollments converts the other way (illuminate_academic_year = cc_academic_year + 1). Every consumer join crosses this boundary somewhere; when counts look shifted by exactly one year, check which convention each side uses.
• response_type: which aggregation a response row is — overall (the whole assessment), standard (one standard), or group (one reporting group). Minted in int_illuminate__agg_student_responses from the name of the unioned source relation, with response_type_id = -1 as the sentinel for overall rows. A null response_type on the scaffold-joined surfaces (int_assessments__response_rollup, the DDI dashboard) means assigned-but-not-taken; fct_assessment_scores_enrollment_scoped makes that explicit as not_taken.
• is_replacement: an internal-assessment sitting where the assessment's AppSheet-tagged grade level differs from the student's current Illuminate session grade — a retained or accelerated student sitting an off-grade assessment. Set in int_assessments__scaffold, with three limits that follow from its SQL: only Text Study, Mathematics, Social Studies and Science; only K-8 assessments; and only sittings that actually exist in Illuminate (the branch inner-joins students_assessments), so a not-taken off-grade assignment never appears. An assessment with no grade tag can never be a replacement. These rows bypass the course-enrollment join (null cc_dcid), are excluded from score anchoring and the enrollment-scoped bridge, and are carried by the student-scoped bridge instead. The sight words dashboard reuses the column name for the analogous off-grade-quiz case; the two are computed independently.
• Performance bands: Illuminate band sets attached per response type in int_assessments__performance_bands. int_illuminate__performance_band_sets computes each band's range as [minimum_value, next band's minimum - 0.1), top band capped at 9998.9. Each band also carries Illuminate's own is_mastery flag, so on band-driven surfaces the band set defines which labels count as mastery. The DeansList feeds are the exception: they hardcode their own ladders and cut points (see their section).
• DDI tiers (nj_student_tier), called "buckets" by every stakeholder ("Bucket 1 and 2" in tickets): the student intervention tier the tier roster publishes. The ladder is not in this family — it lives in the shared hub int_extracts__student_enrollments_subjects, and the roster passes it through.
• QBLs and Power Standards: retired programs. The lookup named range still exists and rpt_tableau__ddi_dashboard still computes is_qbl from it; finishing the retirement is #5656. rpt_tableau__assessment_dashboard already hardcodes its power-standards columns to null for workbook compatibility.
• Sight words: K-2 sight-word quizzes stored as an Illuminate repository (custom student-data table), not as assessment responses — they flow through int_illuminate__repository_data, not the response rollup.
• DKI, Module Dashboard, Mastery by Classroom: names users bring to support tickets. DKI View, Module Dashboard and Mastery by Classroom are worksheets inside the DDI Suite workbook; DKI is also the name of the recurring data meeting the dashboard preps, so "the DKI dashboard" in a ticket means the DDI Suite. (What DKI expands to is not yet written down — owner to confirm.)

Where the data comes from

• Illuminate DnA and Repositories, via the dlt sync (27 staging tables plus 46 enabled repository_<id> stubs; layout and the repository-model mechanics are in src/dbt/kipptaf/models/illuminate/CLAUDE.md). Only repositories 443-472 are on the dlt schedules; the 16 enabled stubs below 443 never sync, so their data is frozen (known issue). Owned by the data team; assessment content is owned by the achievement directors who build it.
• The AppSheet Illuminate Assessments Extension: the tagging app writes a BigQuery table read as stg_google_appsheet__illuminate_assessments_extension, keyed on source (illuminate or apex) plus assessment_id, since the two systems' ids collide. The app links the two tables on assessment_key, those two values joined with a hyphen (apex-123). The app's item list comes back out of the warehouse as rpt_appsheet__assessments, Miami's Focus Assessments included, so the loop is: catalog feed → staff tag in the app → tags land in BigQuery → tags drive everything. Marya Shukla runs QC; achievement directors tag their own assessments.
• The Assessments lookup spreadsheet (ask the data team for it): one spreadsheet, five named ranges, five staging models — standard_domains (Marya Shukla), academic_goals, course_subject_crosswalk, vendor_subject_crosswalk (data team), and the retired qbls_power_standards. All five externals share the spreadsheet's URI, so an edit to any tab re-triggers all five sources together.
• Shared hubs (out of family, one line each):
• int_extracts__student_enrollments / _weeks / _subjects: enrollment, demographics, week scaffold, and the subject-level supplement (tiers, iReady proficiency, state proficiency) the dashboards and tier roster read.
• base_powerschool__course_enrollments: course/section/teacher context, joined on illuminate_subject_area.
• stg_google_sheets__reporting__terms: RT-type reporting terms, date-range joined for term labels and term keys.
• stg_google_sheets__people__locations: school-to-region mapping in the scaffold.
• SchoolMint Grow + int_performance_management__observation_details: the walkthrough branch of the DDI dashboard.
• int_iready__instruction_by_lesson_union: weekly iReady lesson-pass counts on the DDI dashboard.
• Adjacent, not in family: rpt_ln__* (the roster feed into Illuminate: courses, enrollment, users, terms), and the non-Illuminate branches of the assessment star (Amplify/DIBELS, iReady, STAR, college, AP, state), which belong to the DIBELS, CARAT and STAT families.

The DDI Suite workbook

Tableau exposure ddi_suite, two extracts, both views. rpt_tableau__assessment_dashboard drives the Assessment Dashboard and Module Dashboard worksheets; rpt_tableau__ddi_dashboard drives the weekly worksheets (the DKI View; tickets also name a Mastery by Classroom worksheet whose extract the owner should confirm). The full worksheet-to-filter breakdown still needs the owner's walkthrough or a workbook download.

rpt_tableau__assessment_dashboard

• What it shows: Illuminate assessment response rows (overall, standard and group level) joined to enrollment, course/teacher and student-subject context. Carries every Illuminate assessment, tagged or not, with is_normed_scope marking the tagged ones; drives the Assessment Dashboard worksheet (all assessments) and the Module Dashboard worksheet (normed only).
• Grain: student x response record; the course join is pinned to one section per subject (rn_student_year_illuminate_subject_desc = 1), yet prod shows about 0.7% of keys duplicated exactly twice (cause untraced, measured 2026-10-02). No uniqueness test (known issue).
• Reads: int_assessments__response_rollup, int_extracts__student_enrollments (+_subjects), base_powerschool__course_enrollments.
• Worth knowing:
• Population: rn_year = 1, grade_level != 99, current and prior academic year only.
• Miami holds zero rows in prod in every year (measured 2026-10-02): the rollup still carries Miami AY2025 responses, but none survive this model's enrollment join after the Focus cutover.
• power_standard_goal, is_power_standard, standard_domain are hardcoded null — retired fields kept so the workbook's field list does not break.

rpt_tableau__ddi_dashboard

• What it shows: the weekly DDI cycle — students x school week x that week's tagged-assessment rows (expected or taken), plus iReady lesson completion (ES/MS) and a staff walkthrough branch, so instruction, assessment and coaching sit on one axis. It is not one row per enrolled student per week: the module_type != 'WPP' predicate sits in the WHERE clause on the LEFT-joined rollup, which turns the join inner — weeks with no tagged assessment produce no row, and untagged assessments (null module_type) never appear. Assigned-but-not-taken rows for tagged assessments do survive (tens of thousands per year). #4808 describes the same shape.
• Grain: three unioned branches — ES/MS (grades 0-8) and HS (9-12) at student x week x response row; walkthrough rows at observation-row grain with student columns nulled and staff overloaded into them (student_name is the staff member, subject_area is their microgoals, response_type is the literal walkthrough). No uniqueness test (known issue).
• Reads: int_extracts__student_enrollments_weeks, int_assessments__response_rollup, the standard-domains and QBLs sheets, int_assessments__academic_goals, courses, the iReady lesson union, and the SchoolMint Grow/performance-management chain.
• Worth knowing:
• Eligibility is enrollment as of the week end (is_enrolled_week_end, enroll_status in (0, 2, 3)), not current status — pinning to enroll_status = 0 retroactively erased withdrawn students' history (#4807).
• The module_type != 'WPP' filter carries a TODO: Remove SY26 marker, and it is also what makes the join inner (see the grain note above).
• A row with a null response_type and null date_taken is assigned-but-not-taken; is_complete encodes it, and the still-open PR #3576 proposes an is_completion_row flag to make completion-rate denominators explicit.
• The ES/MS and HS branches join the QBLs sheet and the goals sheet with different keys (ES/MS includes grade_level and requires qbl is not null; HS includes neither) — deliberate, but easy to misread.
• iReady lesson columns are ES/MS only; HS and walkthrough rows carry nulls.
• The iReady weekly metrics undercount when keyed to the assessment subject (#4808).

The audit dashboards

rpt_tableau__assessment_entry_audit (Data Quality Dashboard)

• What it shows: expected-to-take internal assessments for the current year against what has actually been entered in Illuminate — the "who has not entered scores" view.
• Grain: student x expected assessment, tested as (student_number, assessment_id).
• Reads: int_assessments__scaffold, int_extracts__student_enrollments (enroll_status = 0 only), reporting terms, int_illuminate__agg_student_responses (response_type = 'overall').
• Worth knowing: replacement sittings are excluded (not is_replacement) but the output still carries an is_replacement column hardcoded to null.

rpt_tableau__sight_words_dashboard (Literacy Dashboard)

• What it shows: Sight Words Quiz results per student x quiz x word, with untested on-grade students kept as null-value rows; shares a workbook with the DIBELS dashboard (see the DIBELS family doc for that side).
• Grain: (repository_id, student_number, sight_word), tested at severity: warn.
• Reads: int_illuminate__repositories/__repository_data, repository fields and grade levels, reporting terms, int_extracts__student_enrollments_subjects (Reading rows).
• Worth knowing: Paterson is excluded in both branches. The off-grade branch (is_replacement = true) keeps only scored rows, while the on-grade branch keeps every expected student — so completion rates only mean something on-grade, and even there the on-grade branch applies no enroll_status filter, so withdrawn students stay in the denominator.

DeansList report-card extracts

• What triggers it: the Dagster deanslist extract job, daily at 1:25am Eastern (deanslist-annual.yaml config; "annual" names the config rotation, not the cadence). Each asset queries its rpt_deanslist__* view and delivers json.gz to DeansList's SFTP as a custom data object (CDO).
• Inputs: int_assessments__response_rollup, the AppSheet region tags (regions_report_card, regions_progress_report), performance band sets, enrollments, the standard-domains sheet, reporting terms.
• The four feeds:
• rpt_deanslist__mod_assessment: K-4 enrichment (non-ELA/Math) subject averages per term, current year only; feeds only the Enrichment table on NJ ES report cards. Includes Unit Assessments always — tagged or not — and other scopes only when AppSheet-tagged report-card-eligible for the student's region. Output subject_area is the literal ENRICHMENT (a CDO schema placeholder).
• rpt_deanslist__mod_standards: ELA/Math/Writing reporting-group averages, all grades, current year only, with Writing folded into Text Study and a five-label mastery ladder (Advanced Mastery down to Far Below Mastery) from the band sets; feeds the "overall" course grades on ES report cards. Gated by tagging (is_internal_assessment) only — it has no region-tag gate.
• rpt_deanslist__mod_standards_domains: K-4 progress-report (overall) and report-card (standard-domain) performance, with no year filter (it carries every year the rollup holds). The cut-point case lists K-2 branches at 90/75/60 and then unconditional branches at 85/70/50/30/0, so the effective K-2 ladder is Exceeds 90+, Met 75-89, Approaching 50-74 (the 60 branch only keeps 70-74 out of the generic Met), and K-2 students fall into Below/Far Below under 50 like everyone else. Feeds the mastery pages on NJ ES report cards.
• rpt_deanslist__sight_words: raw sight-word mastery per student per word, current year forward, with no grade or region filter (Paterson included, unlike the dashboard); retested displays as its own status but counts as mastered (is_mastery = 1). The report cards use it for the K-1 sight-words table.
• Outputs: DeansList report cards and progress reports. The AppSheet region tags gate mod_standards_domains and the non-UA scopes of mod_assessment; untagged Unit Assessments still reach the enrichment feed, and mod_standards is gated by tagging alone, not region.
• Who runs it: nobody by hand; the schedule runs and the AppSheet tags steer it. When a score is missing from a report card, check the tag first, then the mod audit sheet.

The Google Sheets

rpt_gsheets__ddi_tier_roster

One row per currently-enrolled student x iReady subject with nj_student_tier, prior-year NJSLA and iReady EOY proficiency — the tier-planning roster. Pure passthrough of int_extracts__student_enrollments_subjects (current year, rn_year = 1, enroll_status = 0); the tier logic lives in that hub. Lands in the NJ DDI Roster - Source sheet (exposure nj_ddi_roster_source). The extract applies no region filter, so Miami rows are present and njsla_previous_year carries their FAST values — accepted (2026-10-03): the reporting sheets downstream alias the column correctly, so no SQL change.

rpt_gsheets__deanslist_mod_audit

The QA view behind the three mod feeds: the individual pre-aggregation response rows, with a windowed computed_avg_pct_correct reproducing each published feed's GROUP BY so a published average can be checked against its inputs. Scoped to the current and prior year — wider than mod_assessment and mod_standards (current-year) but narrower than mod_standards_domains, which has no year filter. There are no flag columns; the comparison is done by eye in the sheet. Known gap: its mod_standards slice does not apply the Writing-to-Text-Study remap the published feed applies, so Writing rows need manual reconciliation.

The AppSheet catalog feed

rpt_appsheet__assessments sends the entire assessment catalog (int_assessments__assessments_members, no filter) to the tagging app, including the three region-tag fields staff edit and a 0-indexed grade_level. This is the outbound half of the tagging loop; the inbound half is the AppSheet staging table. Because assessments_members already left-joins the extension, staff see their own prior tags when they open the app.

The assessment Cube (Illuminate branch)

The assessment star — dim_assessments, dim_assessment_administrations, dim_assessment_goals, bridge_assessment_expectations_enrollment_scoped / _student_scoped, fct_assessment_scores_enrollment_scoped / _student_scoped — is multi-source. This page covers only its Illuminate branch; Amplify/DIBELS, iReady, STAR, college and state branches belong to their own families, and the Cube-side reference (measure semantics, view structure) ships with PR #5495 under src/cube/mcp/project_knowledge/.

What the Illuminate branch contributes:

• dim_assessments is member-grained for Illuminate (one row per assessment_id, type = 'illuminate'); other sources are canonical-grained.
• dim_assessment_administrations fans the canonical catalog out per assessed region; source_assessment_id is populated only for Illuminate.
• The two expectation bridges split the scaffold on resolvability: sittings with a real course enrollment go enrollment-scoped; replacement sittings and the synthesized K-4 Newark/Camden ES-Writing rows go student-scoped.
• fct_assessment_scores_enrollment_scoped's internal branch keeps assigned-but-not-taken rows (response_type = 'not_taken') and drops any score int_assessments__resolved_section_enrollments cannot tie to a section (the resolver — subject section first, then a discipline match for state scores, then homeroom — is the scope of record).
• Neither fact exposes a source column. The only reliable way to isolate Illuminate rows is the join path assessment_administration_key -> dim_assessment_administrations.assessment_key -> dim_assessments.type = 'illuminate'.
• The public surface is student_assessment_scores_view, row-level-secured per viewer; its proficiency_rollup pre-aggregation picks up each star rebuild within about 30 minutes plus one partition build, for the quarterly partitions from July 1 of the previous school year on. Filter assessment_date rather than date_day for date ranges: it prunes the month-partitioned scores fact.

Supporting models

In-family intermediates and their outside readers ("also read by" — a change here moves those consumers too):

• int_assessments__response_rollup — the family workhorse: scaffold x responses x bands, one row per expected student-assessment-response. Also read by rpt_gsheets__assessment_roster, rpt_gsheets__school_metrics_extract, rpt_tableau__miami_fast, and int_assessments__college_assessment_practice (CARAT family), so a rollup change also moves the practice-SAT chain.
• int_assessments__assessments_members — also read by bridge_assessment_administration_members (disabled).
• int_assessments__scaffold — also read by the entry audit (in family).
• int_assessments__academic_goals — also read by rpt_tableau__academic_goals_rollup (mid-retirement) and rpt_tableau__state_assessments_dashboard (STAT family).
• int_illuminate__assessments, __agg_student_responses_standard, stg_illuminate__public__students, and the overall-responses staging — also read by int_act__test_prep_scores (ACT family).
• int_illuminate__repositories / __repository_data — also read by the sight words dashboard (in family).
• stg_google_sheets__assessments__course_subject_crosswalk — also read by dim_courses.

Inputs (hand-maintained)

Input	Maintainer	Feeds	Update rhythm
AppSheet assessment tagging	ADs; QC: Marya Shukla	everything (is_internal_assessment)	frequent; turned over every year
Standard domains named range	Marya Shukla	report-card domains, DDI dashboard	largely static
Academic goals named range	data team	goals on the DDI dashboard, dim_assessment_goals	once a year in theory, several each fall
Course subject crosswalk	data team	int_assessments__course_enrollments, dim_courses	annual audit (below)
Vendor subject crosswalk	data team	state/vendor branches of the star	rare
QBLs / Power Standards named range	nobody (retired)	rpt_tableau__ddi_dashboard.is_qbl (#5656)	none

The course subject crosswalk's annual audit is worth systematizing (the family skill should carry it as a procedure): list new courses with current-year enrollments that are missing from the sheet, then hand the list to c3/academic ops to confirm which are tested subjects — connected to Illuminate results, state testing results, both, or eventually Focus Assessments.

Decisions

• Tag-driven scope. The AppSheet extension, not Illuminate metadata, decides what is a DDI assessment. This keeps the stack robust to Illuminate's messy catalog (duplicates, "Copy of" titles) at the cost of making the tags a single point of failure.
• Canonical grouping by attributes, not region. Network policy is one shared internal assessment per module, so regional copies share a canonical id via their shared attributes. The one exception is Miami's own Florida versions, which stay in a separate group.
• Replacement sittings are visible but unscored. Off-grade sittings show on rosters and the student-scoped bridge but never anchor scores — the section resolver cannot place them.
• The resolver is the scope of record for the fact. An internal score with no resolvable section is dropped from fct_assessment_scores_enrollment_scoped rather than carried with a null FK.
• NJ state history is capped at 7 years in the fact; FL is not.
• 5x/day star, daily Cube pre-agg. Intraday rebuilds serve Tableau; Cube users see daily data.
• Miami exit. Miami's Illuminate feed ended with SY25-26. Its historical rows stay; nothing new lands until Focus assessment data is ingested, at which point the internal branch grows a Focus source (decided 2026-09-30).

Support themes

What people actually ask about, from a keyword pass (assessment dashboard, module dashboard, DDI, DKI) over the Zendesk warehouse copy (17 relevant tickets) and the two Slack channels where this work is discussed (22 relevant threads), 2026-04-02 through 2026-10-02, measured 2026-10-02. The themes, most frequent first, and where each one points:

1. "My assessment is not on the dashboard." The assessment director's own triage rule: anything not rolling up is untagged or mistagged — wrong term date, wrong grade's tag, or no tag at all. When the tag is right, follow the freshness chain above: did the Illuminate sync land it; has the assessment star ticked since; has the Tableau extract refreshed. Occasionally the cause is genuinely on the data side (an overnight pipeline bug has reproduced this with tagging fully correct), so confirm the rows in the extract before and after a refresh rather than re-arguing the tag. After a tag fix, users expect a manual refresh push rather than waiting for the next tick, and they know the refresh times.
2. Two worksheets disagree, or the dashboard disagrees with Illuminate. Module Dashboard vs DKI View discrepancies are usually denominator questions: completion rows vs mastery rows (null response_type is assigned-but-not-taken), and course-enrollment vs grade-level population (an Algebra 1 student is in grade 8 on one cut and in the Algebra course on the other). A multi-assessment DKI View cut also confuses when quizzes are tagged to different dates per grade level — viewing by grade level is the documented workaround. Dashboard-vs-Illuminate discrepancies get reconciled against the extract rows and the sync times.
3. Wrong or missing teacher/section. Two distinct causes. In the model: the course join keeps one section per student, year and subject (rn_student_year_illuminate_subject_desc = 1), so a student's honors or second section is dropped from classroom rollups, and a co-teacher or interventionist can appear as the section of record. In the source: a wrong course assignment in PowerSchool rolls students up under the wrong course on the Module Dashboard — the fix is in PowerSchool, and it populates on the next morning's refresh.
4. Access. A blank DDI Suite page from the Launch page, or a login failure, is Tableau licensing or permissions — not a data defect. Route to the Tableau admin path before reading any SQL.
5. Scores entered but banded wrong. Entered answers showing a too-low mastery band is the shape of #5399 (band tables sync at midnight only) — check the band-set join before suspecting the scores.

One recurring feature ask: averaging across module sequence numbers (for honor roll). That is a workbook/extract change, not a data defect.

Known issues, need to fix

Tracked elsewhere:

• #5654 — inconsistent SY25-26 scope/module tagging destabilizes canonical groups.
• #5399 — performance band tables sync only at midnight; same-day scores stay unscored intraday.
• #4808 — DDI dashboard iReady metrics undercount.
• #3797 — Illuminate catalog defects (ops-tracked).
• #5349 — fragmented Illuminate standards codes.
• #4173 / #4174 — mart orphan rows (college-practice module codes; 5 pre-existing expectation orphans).
• #5656 — finish retiring QBLs/Power Standards (the sheet staging model and is_qbl).
• #4446 — Illuminate dlt sync refactor.
• #5700 — the sight-words dashboard's warn-level uniqueness test fails in prod (about 5,500 duplicate keys, measured 2026-10-03): 6 SY21-22 Sight Words quizzes carry placeholder TBD word fields in Illuminate. Fixed in Illuminate, not dbt.

Found during the documentation run, not yet tracked separately:

• Uniqueness tests cover every family model with a prod-verified clean grain; two still lack one: rpt_tableau__assessment_dashboard (no clean key exists — the ~0.7% duplicated keys above) and rpt_tableau__ddi_dashboard (its properties YAML is in flight on PR #3576).
• The academic-goals sheet holds 2 duplicated key rows (one AY2024, one AY2025 subject-level HS goal, each entered twice); they flow through int_assessments__academic_goals and are masked downstream by dim_assessment_goals' highest-school-goal-wins dedupe. Check: group by the 5 goal keys having count(*) > 1.
• rpt_tableau__assessment_dashboard's description carries a stale Miami row count and says Miami AY2026 rows have "not landed upstream yet"; prod holds zero Miami rows in any year and none will land — Miami left the stack.
• dim_assessments documentation points at bridge_assessment_administration_members as the member drill-down, but that bridge is disabled with zero consumers.
• int_illuminate__performance_band_sets carries a commented-out materialized: table override — decide and either enable or delete it.
• The 16 enabled repository stubs below 443 are absent from the dlt schedules, so their repository data is frozen (see Where the data comes from).
• The mod audit's Writing remap gap and its year-scope mismatch with mod_standards_domains (see its section).
• The NJ DDI Roster - Source sheet sits in the Reports drive folder despite its name, and no Reports-layer IMPORTRANGE consumer of either extract sheet was found — confirm how people actually read them.
• rpt_tableau__ddi_dashboard's TODO: Remove SY26 filter on module_type != 'WPP', which is also what makes its rollup join inner.

Yearly upkeep

1. July: the current_academic_year var rollover (repo-wide, not family-specific) moves every current-year filter in the family at once.
2. Before the first administration: achievement directors create the year's assessments in Illuminate and tag them in the AppSheet; Marya Shukla QCs the tags. Nothing shows anywhere until this happens.
3. Goals: add the new year's rows to the academic-goals named range.
4. Standard domains: update the named range when standards or domains change.
5. Reporting terms: confirm the year's RT rows exist (shared hub, but this family breaks visibly when they lag).
6. August: turn the DDI Suite Tableau refresh schedule back on and review its cron list with the assessment director — anything to add or remove for the year.
7. Fall: bump the DDI Suite workbook tabs' default year to the new school year (a workbook edit, requested every year).
8. Fall: run the course subject crosswalk audit (see Inputs) — new courses with current-year enrollments that the sheet is missing, confirmed with c3/academic ops.
9. Check the DDI Suite after the first assessment window: the first real rows exercise the whole chain.