← All proof trees · BUILDING 21

1. 1
   What you seeWithheld by the publisher

   The publisher suppressed this cell (too few students to report). The site stores it as suppressed and never as zero.

   vs. peers

   —

   vs. citywide

   —

   unit

   percent

   direction

   higher is better

2. 2
   How it was computedvalue = published 4-year cohort graduation rate (district file, indexed by cohort entry year; state file, indexed by report year)

   1. Read the district's graduation CSV (all cohorts, 4/5/6-year rates, by subgroup) and store by cohort year.
   2. The state's Future Ready file also carries a 4-year rate for every school including charters; it is stored under the same key but indexed by report year.

   Note: stored percentiles are computed by a batch script after loads; a difference from the live reconstruction means a newer load changed the comparison set and the batch has not been re-run.

   Evidence: metric tree: 4-year cohort graduation rate

3. 3
   The stored cellThe stored cell

   table

   philly_school_year_metrics

   value

   null

   denominator

   57

   suppressed

   true

   citywide_percentile

   null

   comparison_group_percentile

   null

   source_load_id

   8ecdf1ea-7780-4200-bd21-8e8d089c99f1

   Evidence: data model overview

4. 4
   The load that wrote itWritten on 2026-06-04 by sdp_graduation.ts

   source id

   sdp_graduation_school_2014-2025

   file URL

   https://cdn.philasd.org/offices/performance/Open_Data/School_Performance/Graduation_Rates/OD_SDP_Graduation_Rates_School_S_2026-02-12.csv

   SHA-256 of the file read

   8ae3969118c3a9f01e4d8d04aa634855571ab71e914f3d2b689ad811d5639225Copy hash

   bytes

   3,919,459

   year covered

   2010-11..2021-22

   year of this cell

   2016-17 — this measure is stored under a different year than the file's report year; see the metric page

   code commit

   f75e33137e8166eec3f8b5a4b39463d306ef710c

   To confirm you have the same file: download it from the URL above, then run shasum -a 256 <downloaded file> Copy command — the output should equal the hash above. If the publisher has since replaced the file, the hashes will differ, which is itself useful to know.

   Evidence: loader script · first-fetch findings for this file · source registry

5. 5
   The publisher's fileGraduation rates by cohort, 4/5/6-year

   School District of Philadelphia (Performance Office). Four-, five-, and six-year cohort graduation rates by school and subgroup, indexed by the cohort's 9th-grade entry year.

   format

   CSV

   join key

   ULCS Code

   landing page

   https://opendataphilly.org/datasets/sdp-school-performance/

   • Indexed by cohort start year, not graduation year; the state's Future Ready file indexes the same rate by report year.
6. 6
   How it was checkedChecks: 2 passed

   • ✓ passedBase case (district file, 2021-22 cohort): 75 of 75 unsuppressed cells match
   • ✓ passedCompleteness

   Checked at ±0.5: 0 mismatches, 0 cells served but absent from the file, 1 in the file but not served (ALL, n/a, latest year).

   Evidence: Base case (district file, 2021-22 cohort) · Completeness

7. 7
   Assumptions & caveatsAssumptions and caveats

   • The two publishers' rows never collide on (school, year, subgroup) because they use different year conventions; readers must check which convention a given year follows.

   • Two year conventions coexist under one key — a known v1 wart flagged for normalization.

   Algorithms involved: Citywide and peer-group percentiles, Peer groups: 40 nearest neighbors on demographics