Every source is a California Department of Education public file, downloaded from CDE's public data pages in a browser, the way the pages are meant to be used. The pipeline therefore treats source files as locally acquired inputs rather than fetching them at build time: record the access date beside each file when acquired, and the pipeline stamps it into coverage output. CI never touches the network.
| # | Source | What it provides | Acquisition | Status |
|---|---|---|---|---|
| D1 | Public Schools and Districts directory (pubschls.txt), https://www.cde.ca.gov/schooldirectory/ (download: text report) |
Canonical school list: CDS codes, names, districts, status, type, grades, coordinates, charter/virtual flags | Browser download, ~8.8 MB tab-delimited | Endpoint verified live 2026-08-07 (HEAD 200, text/csv, 8,788,547 bytes); parser built |
| D2 | Census Day Enrollment Data, https://www.cde.ca.gov/ds/ad/filesenrcensus.asp (file structure: https://www.cde.ca.gov/ds/ad/fsenrcensus.asp; the pre-2026 URL filesenr.asp now 404s, renamed upstream, observed 2026-08-07) | Enrollment by school, grade, demographic group | Browser download, annual. Acquire from the browser page and record which school year's file was taken | 2025-26 file acquired 2026-08-07, saved as cdenroll2526.txt (32,041,859 bytes, 269,090 data rows); parser built; ReportingCategory names reviewed against the file structure page 2026-08-07; the file's district and statewide AggregateLevel rows are read for the context columns on each school page, taking only Charter=ALL rows, because each entity is published three times (charter, non-charter, both) and in this file Davis Joint Unified's three district rows read 561 / 7,682 / 8,243 for the same category. Homeroom never sums school rows into a district or state figure |
| D3 | Chronic absenteeism files, https://www.cde.ca.gov/ds/ad/filesabd.asp, linking https://www3.cde.ca.gov/demo-downloads/attendance/chronicabsenteeism25-v2.txt (file structure: https://www.cde.ca.gov/ds/ad/fsabd.asp) | Chronic absenteeism eligible cumulative enrollment, count, and rate by school and group, with CDE small-cell masking | Browser download, annual | Acquired 2026-08-21. The 2024-25 file (chronicabsenteeism25.txt, 33,781,100 bytes, 341,490 data rows) was downloaded and its header read directly: identity columns are spaced (Academic Year, Charter School, ...) while the three measure columns are concatenated (ChronicAbsenteeismEligibleCumulativeEnrollment, ChronicAbsenteeismCount, ChronicAbsenteeismRate); Charter School and DASS are two independent All/Yes/No dimensions, so a genuine district or state figure needs both to read All, not the one dimension D2 has. 25 reporting-category codes were observed, none shared with D2's own codes for the same underlying groups (D2's RE_A is this file's RA); see src/homeroom/profiles.py ABSENTEEISM_CATEGORY_NAMES. Scanning every row found 104,469 of 341,490 (30.6%) carrying a mask on at least one of the three numeric cells, and masking is always all three together, never a subset. src/homeroom/absenteeism.py was written and verified against this file; docs/ROADMAP.md M3 records the measured coverage. Wired into make data and make site's default invocation (Makefile --absenteeism), rendered on every school page (ROADMAP M3) |
| D4 | CA School Dashboard research files, https://www.cde.ca.gov/ta/ac/cm/ | Dashboard indicator data behind the color bands | Browser download, annual | Planned |
| D5 | Teacher Assignment Monitoring Outcome files, https://www.cde.ca.gov/ds/ad/filestamo.asp, linking https://www3.cde.ca.gov/demo-downloads/tamo/tamo2324.txt (file structure: https://www.cde.ca.gov/ds/ad/fstamo.asp). CDE publishes these from the Commission on Teacher Credentialing's CalSAAS system | By school: teaching FTE (full-time-equivalent, not integer counts) held on a clear credential and appropriately assigned, and the other authorization states the state tracks, with CDE small-cell masking where it applies | Browser download, annual. Record which school year's file was taken; it reports on a different cycle than D2 | Acquired and schema-verified 2026-08-21; not joined to any artifact or page by default. The 2023-24 file (tamo2324.txt, 234,206,408 bytes, 1,528,796 data rows) was downloaded and its header read directly, which found the provisional contract this project carried before acquisition (issue #5) wrong in every particular: real column names are spaced (Total FTE, Clear FTE (count), ...) rather than concatenated; there are seven outcomes, not five (incomplete and na were missing entirely); values are fractional FTE, not integer assignment counts; Charter School/DASS are the words All/Yes/No, not D2's single-letter ALL/Y/N; the file's grain is one row per (school, charter, DASS, grade span, teacher experience level, teacher credential level, subject area) -- up to 150 rows per school, not one -- of which exactly one row per school (experience = credential = ALL, subject = TA) is CDE's own whole-school total, verified against all 10,064 schools in the file; and the real header carries CDE's own typo, Unknown FTE FTE (percent). Scanning every numeric cell in the file (1,528,796 rows) found zero instances of * or any other unrecognized sentinel: unlike D2 and D3, this file does not appear to mask at all in this vintage, and its file-structure page carries no small-cell suppression rule the way D3's does. src/homeroom/assignments.py and fixtures/tamo.sample.txt were rewritten to match; docs/ROADMAP.md D5a records the measured counts. Publishing a D5 figure on a school page, or wiring --assignments into make data's default invocation, is a separate decision this acquisition does not make; make data and make site are unchanged and no D5 number about a real school is published anywhere |
| D6 | Per-pupil expenditure (SACS/LCFF public files) | School-level spending where published | Browser download, annual | Planned |
Rules:
- CDS code (14-digit county-district-school) is the only join key. Names are display-only.
- CDE masks small cells (published as
*). A masked cell isnullin every artifact and is counted in coverage output. It is never zero. - No third-party or commercial data (no ratings sites, no real-estate feeds). If CDE does not publish it, Homeroom does not show it.
- A parser may be built ahead of its file, but a number may not. A source whose status
reads "awaiting acquisition" has an access date of
nulleverywhere, publishes nothing, and says so incoverage.json; the pipeline is tested to keep the code constant and this table in agreement, so a build cannot stamp a date nobody recorded. - Each source keeps its own school year. Sources report on different cycles, and a profile carries each year beside its own data rather than one label over data from two calendars.
The ask layer quotes CDE's own definitions rather than paraphrasing them. The
pages it quotes are retrieved by tools/corpus_fetch.py into corpus/, and
corpus/manifest.json is their provenance register: URL, retrieval date (UTC),
the page's own "Last Reviewed" date, SHA-256 of the HTML as received and of the
text as committed, and passage count. The same rules apply as to the data files:
a person runs the fetch, CI never does, and a file that no longer matches its
recorded hash is refused by the loader rather than quoted. See corpus/README.md.