UCT ← Back to the explorer's documentation page  ·  source: resolutions/README.md

Adjudication register

One CSV per faculty (com.csv … hum.csv — all six exist) holding per-case adjudications of handbook print errors and ambiguities. Commerce's is the only seeded register so far (44 provisional entries); EBE/LAW/FHS passes are queued in DEV-TODO.md, and SCI/HUM are empty by design (major-years carry no_anchor — nothing to adjudicate per year). Consumed by build_final_dataset.py (rule R3, which pre-empts all auto-rules). The as-printed tables are never touched — every adjudication is applied only in the final-clean layer, with the original values retained alongside.

Adding an entry

  1. Confirm the discrepancy against the PDF (the source_page column in the as-printed tables points at the page).
  2. Append a row (columns below). res_id must be unique and stable (<FAC>-<year>-<seq>, e.g. COM-2025-008); one row per (year, plan_code, study_year) — a case recurring across editions gets one row per year, sharing the rationale.
  3. Re-run python build_final_dataset.py --year <year> (or the batch runner). validation/validate_final.py fails if a register row goes unconsumed (typo'd plan code / year), so mistakes surface immediately.

Columns

Column Values / meaning
res_id COM-2025-001 — stable, referenced by resolution_ref in outputs
year, faculty, plan_code, study_year which specialisation-year
scope spec_year (default), row, published_fee
row_selector course_code (or seq) when scope=row
issue stated_misprint, unlabelled_choice_menu, or_double_count, missing_block, fee_label_mismatch, other
action accept_computed, accept_stated, set_stated_corrected, set_final_credits, pin_choice, include_row, exclude_row, set_final_fee, set_published_ref
value number for set_* actions; ;-separated course codes for pin_choice
rationale the argued justification, in full sentences
evidence PDF page refs, cross-edition comparisons, arithmetic
decided_by, decided_date provenance of the judgment

decided_by="provisional (Claude), pending review" marks machine-proposed adjudications awaiting human sign-off; amend the row (and decided_by) when reviewed.