Methodology: mapping the bio_oge bank to official OGE exam task numbers

Question this answers: the open FIPI task bank tags every biology question with a КЭС (content codifier, e.g. 7.9 Обмен веществ...) and an answer_type_code (see scraper/answer_types.py) — but not the official exam task number (№1 through №26). This document derives that mapping from FIPI's own official documents, sourced and read directly (not guessed), and is honest about where the derivation is exact vs. a best-effort group. It mirrors METHODOLOGY.md's approach for math (math_ege_profil) — same philosophy, separate document because OGE biology's numbering, structure, and answer-type taxonomy are entirely different from EGE math's.

Primary sources (official, read directly — not summaries of summaries)

All fetched from FIPI's own document server (doc.fipi.ru), the authoritative source — not a third-party test-prep site. PDFs are re-hosted here so you can open them directly: Методология on the dashboard links every one of these inline.

This document uses the 2026 spec as primary (it's the current, ratified — not draft — version), unlike math's METHODOLOGY.md which deliberately targets 2027 because math's profile-level exam is being restructured (19→20 tasks) next year. Biology has no such restructuring: the 2026 spec's own §12 states "Изменения структуры и содержания КИМ отсутствуют" (no structural changes vs. 2025), and a first-page skim of the 2027 kodifikator draft shows the same content-section numbering (1–7) already present in 2026's codifier and in our own scraped codifiers column — so there's no forward-looking reason to prefer the draft here. The 2027 documents are still hosted above for reference/comparison.

Exam structure (from the 2026 specification, §6–7)

26 tasks, two parts, 47 max points, 150 minutes: - Part 1 (tasks 1–21): short answer (21 tasks, 34 points, 72% of the total). - Part 2 (tasks 22–26): extended answer (5 tasks, 13 points, 28% of the total) — this is answer_type_code = 'extended_answer' in our schema, and, exactly like math's Развёрнутый ответ, has no FIPI answer-checker at all. It is deliberately untouched by this project, same reasoning as math's open-answer part (see OPEN_ANSWER_METHODOLOGY.md).

Task-by-task answer type — verified against the real demo variant, not the abstract spec table alone

The specification's own appendix ("Обобщённый план") states the counts and content-codes per task, but its two-column PDF layout doesn't extract cleanly enough to trust as an exact table (see "What's exact vs. best-effort" below). Instead, every one of the 26 tasks below was read directly off the actual 2026 demo variant — a real example question, in the real format, for every task number — which is unambiguous: you can see exactly what shape of question and answer each number is.

Counts match the specification's declared totals exactly: 5×single_choice (6,8,12,14,15), 6×multi_choice (4,7,9,16,17,19), 5×matching (2,11,13,18,21), 3×sequence (3,5,20), 1×term_placement (10), 1×short_answer (1), 5×extended_answer (22–26) = 26.

Answer-submission format per type (separately reverse-engineered, see BIO_OGE_ANALYSIS.md and WORKLOG.md 2026-09-15)

Not part of the task-numbering question above, but the other half of "understanding this exam": single_choice = bare digit; multi_choice = positional binary string over every checkbox in order (not the selected numbers concatenated — that was tried first and FIPI's checker rejected it); sequence/matching/term_placement = concatenated digits in position order; short_answer = plain text. All confirmed empirically against the real FIPI checker, not assumed.

What's exact vs. best-effort — read this before trusting a number

Same principle as math's METHODOLOGY.md: where the bank's own metadata (codifier + answer type) uniquely determines a task number, it's exact; where several task numbers share the same signature and can't be told apart, they're grouped honestly rather than guessed.

scraper/assign_oge_task_numbers.py encodes these rules:

• Unique by answer type alone (exact, no further work needed): short_answer→1, term_placement→10 — each is the only task using that answer type, so every row of that type gets that number with full confidence.
• sequence (3 task numbers, split by content): task 20 detected by codifier section 2 (ecosystems) or food-chain keywords, OR (as a confirming signal on already-solved rows) a letter-containing correct_answer; task 3 by codifier sections 5/6 (plant/animal systematics) or taxonomy keywords; task 5 by elimination (whatever's left). All three came out exact — 149/149 sequence questions classified, no group needed.
• single_choice (5 task numbers): task 12 detected via FIPI's own fixed phrase "Верны ли следующие суждения" in the question text (a format signal, not content — codifier keywords can't find this one, since task 12 can be about any topic); task 6 by scientific-method/instrument keywords; tasks 14 and 15 are grouped — both draw on the same broad "human body" codifier section (7.x) with nothing in our scraped metadata to tell "identify this organ in a picture" (14) apart from "name this life process" (15); task 8 is the residual.
• multi_choice (6 task numbers): tasks 4 (codifier section 1) and 19 (section 2) came out exact; tasks 7 and 9 are grouped (both organism-classification content, section 4/5/6) and tasks 16 and 17 are grouped (both human-body content, section 7) — same fundamental limitation as math's grouped buckets: the bank simply doesn't record which specific task template a question was drawn from, only its topic and answer shape.
• matching (5 task numbers): tasks 2 (kingdoms/systematics), 18 (skeleton, "скелет"/"кост" keywords), and 21 (ecosystem section) came out exact; task 11 (general human-process matching) is the residual for section-7 matching questions; task 13 (the rare photo-comparison-to-models format) is left unclassified — nothing in our scraped codifier/text distinguishes it from task 11's ordinary matching, and it's genuinely rare in the source exam (one distinct sub-format among five matching tasks), so grouping it in with 11 would silently misrepresent items that may not actually be task 13 at all. Better to leave it NULL than assign a number with no real signal behind it.
• extended_answer: not classified at all (matches math's precedent) — no verification methodology exists for this type yet regardless of task number, so a task-number label would be pure decoration with no functional use.

Result on the real 1979-question bank (2026-09-15 run): 978 exact, 510 grouped, 491 unclassified (355 of those are extended_answer, the deliberately-untouched part — only 136 rows out of 1979 have genuinely no task-number signal).

Implementation

scraper/assign_oge_task_numbers.py applies the rules above over codifiers + answer_type_code + question_text (+ correct_answer as a secondary signal for task 20), writing: - questions.oge_task_number — a single int when exact, NULL when grouped or unclassified - questions.oge_task_group — e.g. '14,15' when grouped, NULL otherwise - questions.oge_task_confidence — 'exact' | 'grouped' | NULL

These are separate columns from ege_task_number/ege_task_group/ege_task_confidence (math's columns) — not reused, since OGE's numbering is an entirely independent scheme from EGE's (see schema.sql's comment on why). Re-run assign_oge_task_numbers.py any time this rule table is revised or the bank is re-scraped — it's idempotent, just re-derives and overwrites these three columns for every bio_oge row.

The dashboard's "Номер задания ОГЭ" filter (fipi.lungu.ru) and the tutor's task-selection (fipi_bank.js problem --oge-task N) both read these columns directly — see TUTOR_INTEGRATION.md and AGENTS.md in the ioiva/tutor repo for how Varya can ask for a specific exam task number and have it threaded through.