# Results record for one corpus (PB-RAG-10-03, story RAG-10-S3).
# Every field must be filled before the decision is shared.
# Part 1 is written by snippets/eval/score_validation.py, in this order.
# Part 2 is added by hand after the script has run.

# --- Part 1: written by the script ---
corpus: "${CORPUS_ID}"            # --corpus
sample:
  documents: null                 # --docs: documents in the sample (team default: at least 50)
  pages: null                     # scored pages, counted from pages.csv
  types: {}                       # scored pages for each document type, for example digital_pdf: 20
extraction:
  path: "content_understanding"   # --path: content_understanding or document_layout
  api_version: "2026-04-01"       # --api-version
  preview_features: []            # --preview-features, for example image descriptions (item D9)
  failed_documents_pct: null      # documents with an error or a warning (items D3 and G6)
quality:                          # percentages from the hand scores (part E)
  table_fidelity: null
  figure_coverage: null
  number_accuracy: null
  answer_accuracy: null
  holdout_gap_points: null        # sample minus hold-out, table fidelity (item B7)
  scorer_gap_points: null         # --scorer-gap (item E13)
answers:                          # counts from questions.csv (part F)
  total: null
  right: null
  partly: null
  wrong: null
  no_answer: null
  kinds: {}                       # questions of each kind: table, figure, text, not_in_docs
  answer_accuracy: null
cost:                             # from the real bill (items D2 and G1)
  pages_billed: null              # --pages-billed
  run_cost: null                  # --cost, in your currency
  per_1000_pages: null            # run_cost / pages_billed x 1,000
  budget_per_1000_pages: null     # --budget-per-1000 (item A7)
  processing_minutes: null        # --minutes
decision: null                    # go, go_with_conditions, no_go or not_decided
reasons: []                       # why the script chose the decision
reviewer: null                    # fill in by hand: who approved the decision (item G8)

# --- Part 2: add by hand ---
run:
  search_tier: "${SEARCH_TIER}"   # item A5
  region: "${REGION}"             # items A5 and A6
  started_utc: null               # item D2
  finished_utc: null              # item D2
conditions: []                    # item G7, for example "Scans under 200 dpi are out of scope"
revalidate_when: []               # item G9, for example "A new scanner or document type"