config.yaml

# Fallback LLM for text generation and visual analysis. Image generation and
# speech use their own task-specific defaults; step-level `model` values still
# override this one. This can also be changed in Studio under Settings > Models.
default_model: "openai:gpt-5.4"
default_image_generation_model: "openai:gpt-image-2"
default_speech_generation_model: "gpt-4o-mini-tts"

structure_types:
  group: >
    Generic container used to bind sibling text leaves that form one visual
    unit (e.g. multi-sentence paragraphs, groups of labels). Use this whenever
    two or more text leaves belong together visually but none of the more
    specific containers applies.
  heading: A heading block — contains one or more leaf nodes that together form a single heading.
  paragraph: >
    A prose paragraph. Children are text leaves (usually role "text") in reading order.
  list: An ordered or unordered list. Children are list_item containers.
  list_item: One item of a list. Children are the leaf(s) making up that item.
  table: A data table. Children are table_row containers, in top-to-bottom order.
  table_row: One row of a table. Children are table_cell containers, in left-to-right order.
  table_cell: >
    One cell of a table. May be empty (no children) to preserve column alignment
    for visually blank cells. When non-empty, children are the leaves or
    containers inside the cell.
  sidebar: >
    A visually separated callout, tip box, or aside. Children are the leaves
    and containers inside the sidebar.
  panel: >
    A generic bordered or visually separated panel that is not a sidebar
    (e.g. a decorated info panel or a full-page boxed layout).
  activity: >
    A complete activity or exercise block. Children include any introductory
    leaves (activity_number / activity_instruction / activity_question) and
    one or more activity_option containers.
  activity_option: >
    One option or answer choice within an activity. Always a container, even
    when it holds a single text leaf — never use a bare leaf as an option.
  image_group: >
    A container that binds an image to its associated content — a caption,
    label, or overlay text. MUST have as its first child a leaf with role
    `image` whose `image_id` references one of the listed extracted images,
    followed by one or more caption/label/overlay leaves. Only use this
    when the image has associated content; a standalone image with no
    associated content is emitted as a bare `role: "image"` leaf instead.
  preformatted: >
    A block of preformatted text where line breaks matter (poetry stanzas,
    code samples, ASCII layouts). Children are the text leaves, one per line.

role_types:
  heading: A heading-level leaf (chapter, section, or subheading text).
  text: Ordinary prose or narrative text.
  math: A mathematical expression (use LaTeX notation).
  image: >
    An image leaf referencing an extracted image. Carries `image_id`
    (one of the IDs listed in the input). Has no `text`. Emit as a bare
    leaf in reading order when the image stands alone; wrap in an
    `image_group` container only when the image has associated content
    (caption, label, or overlay text).
  caption: A caption tied to an image or figure.
  label: A short label, tag, or annotation.
  activity_number: Numbering for an activity (e.g. "1.", "A)").
  activity_instruction: A general instruction for the learner at the activity level.
  activity_question: The question or prompt of the activity.
  activity_fill_in_the_blank: Text with blanks to fill in (use ___ for blanks).
  page_number: A page number.
  header: A running header.
  footer: A running footer.
  quote: A quoted passage.
  book_metadata: Publisher, copyright, ISBN, edition, or author metadata.

section_types:
  front_cover: >
    The very first page of the book showing the title, author, and cover artwork.
    Only the first page qualifies — interior title pages do not.
  inside_cover: >
    A page immediately after the front cover or before the back cover,
    typically containing publisher info, ISBN, copyright, or edition details.
  back_cover: >
    The very last page of the book, often with a summary, barcode, or publisher logo.
    Only the final page qualifies.
  separator: >
    A divider page between major sections of the book. Contains minimal content —
    usually just a title, chapter number, or decorative element. No substantive text or activities.
  credits: >
    A page listing contributors, acknowledgements, photo credits, or funding sources.
    Distinct from inside_cover (which has publisher/ISBN info).
  foreword: >
    An introductory page such as a preface, author's note, or "about this book" section.
    Appears before the main content begins.
  table_of_contents: >
    A page listing chapter or section titles with their corresponding page numbers.
  boxed_text: >
    Text presented inside a visible box, border, or colored callout area that
    visually separates it from the surrounding content. Used for tips, definitions,
    key facts, or highlighted information.
  text_only: >
    A reading section that contains only text — no images, no activities.
    Paragraphs, headings, and prose passages.
  text_and_single_image: >
    A reading section that contains text accompanied by exactly one image.
    The image supports or illustrates the text.
  text_and_images: >
    A reading section that contains text accompanied by two or more images.
  images_only: >
    A section that contains only images with no meaningful text content.
    Captions or labels alone do not count as text.
  activity_matching: >
    Match items in a strict one-to-one relationship.
    Each item on one side corresponds to exactly one item on the other side,
    and both sides have the same number of elements.
  activity_fill_in_a_table: >
    Complete a table by filling in missing cells.
    The structure is a grid with rows and columns, and learners must enter
    information into specific cells.
  activity_multiple_choice: >
    Select the correct answer from a list of predefined options.
    Each question presents one or more possible answers, and the learner
    chooses the best option.
  activity_multi_select: >
    Select all options that apply from a list of predefined options.
    More than one option per question may be correct, and the learner must
    pick every option that satisfies the prompt (a "check all that apply"
    or "mark every X" task). Use checkboxes, not radios.
  activity_underline_text: >
    Underline one or more words, phrases, or sentences directly in the text.
    The learner selects inline text segments instead of separate option cards,
    and the correct answer may be one segment or multiple segments.
  activity_true_false: >
    Decide whether each statement is true or false.
    The learner evaluates given statements and selects between two options:
    true or false.
  activity_open_ended_answer: >
    Provide a free-form response without predefined options.
    The learner must generate their own answer, typically as a sentence,
    explanation, or short text.
  activity_fill_in_the_blank: >
    Complete sentences or phrases by filling in missing words or values.
    The learner supplies the missing information directly in the blanks.
  activity_sorting: >
    Group multiple items into categories.
    There are more items than categories, and each category contains
    multiple items (many-to-one classification).
  activity_other: >
    An interactive activity or exercise that does not fit any of the specific
    activity types above. Use this ONLY when the underlying learning mechanic
    is truly unique (e.g., free drawing, crafts, physical movement).
    Do NOT use this just because the physical action is circling, coloring,
    or underlining — classify based on the cognitive task, not the action.
  other: >
    Any section that does not clearly fit any of the above types.
    Use this only as a last resort.

metadata:
  prompt: metadata_extraction

page_sectioning:
  prompt: page_sectioning
  max_refinements: 0

image_meaningfulness:
  prompt: image_meaningfulness

image_cropping:
  prompt: image_cropping

default_render_strategy: two_column

render_strategies:
  llm:
    render_type: llm
    config:
      prompt: web_generation_html
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  llm-overlay:
    render_type: llm
    config:
      prompt: web_generation_html_overlay
      max_retries: 25
      timeout: 180
      visual_refinement:
        enabled: true
  two_column:
    render_type: template
    config:
      template: two_column_render
  two_column_story:
    render_type: template
    config:
      template: two_column_story
  one_column:
    render_type: template
    config:
      template: one_column_render
  fixed_layout:
    render_type: fixed_layout
  activity_multiple_choice:
    render_type: activity
    config:
      prompt: activity_multiple_choice
      answer_prompt: activity_multiple_choice_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_multi_select:
    render_type: activity
    config:
      prompt: activity_multi_select
      answer_prompt: activity_multi_select_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_underline_text:
    render_type: activity
    config:
      prompt: activity_underline_text
      answer_prompt: activity_underline_text_answers
      model: openai:gpt-5.4
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_true_false:
    render_type: activity
    config:
      prompt: activity_true_false
      answer_prompt: activity_true_false_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_fill_in_the_blank:
    render_type: activity
    config:
      prompt: activity_fill_in_the_blank
      answer_prompt: activity_fill_in_the_blank_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_fill_in_a_table:
    render_type: activity
    config:
      prompt: activity_fill_in_a_table
      answer_prompt: activity_fill_in_a_table_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_matching:
    render_type: activity
    config:
      prompt: activity_matching
      answer_prompt: activity_matching_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_sorting:
    render_type: activity
    config:
      prompt: activity_sorting
      answer_prompt: activity_sorting_answers
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true
  activity_open_ended_answer:
    render_type: activity
    config:
      prompt: activity_open_ended_answer
      max_retries: 5
      timeout: 180
      visual_refinement:
        enabled: true

section_render_strategies:
  activity_multiple_choice: activity_multiple_choice
  activity_multi_select: activity_multi_select
  activity_underline_text: activity_underline_text
  activity_true_false: activity_true_false
  activity_fill_in_the_blank: activity_fill_in_the_blank
  activity_fill_in_a_table: activity_fill_in_a_table
  activity_matching: activity_matching
  activity_sorting: activity_sorting
  activity_open_ended_answer: activity_open_ended_answer

quiz_generation:
  # Style quizzes to match the book (typography + colors derived from the book).
  match_book_style: true
  quiz_section_types:
    - boxed_text
    - text_only
    - text_and_single_image
    - text_and_images
    - images_only

pruned_role_types:
  - header
  - footer
  - page_number

pruned_section_types:
  - back_cover
  - credits
  - inside_cover

concurrency: 32

image_filters:
  min_side: 100
  max_side: 5000
  min_stddev: 2
  meaningfulness: true
  cropping: true

# rate_limit:
#   requests_per_minute: 60

# Text-to-speech (read-aloud) settings. `default_provider` and per-language
# `providers` routing decide which engine synthesizes each language.
#
# Gemini defaults to the faster gemini-2.5-flash-preview-tts model; set `model`
# to switch to Pro (higher quality, slower, lower throughput).
#
# `rate_limit` controls the adaptive request limiter for Gemini TTS:
#   - requests_per_minute: auto  → start at the model's documented ceiling
#     (flash 150 / pro 125) and automatically back off on 429s, then recover.
#     This is the default when omitted.
#   - requests_per_minute: 30    → pin the starting ceiling to a fixed number
#     (still backs off on 429s).
#   - min_/max_requests_per_minute → override the adaptive floor/ceiling.
#
# speech:
#   default_provider: gemini
#   providers:
#     gemini:
#       model: gemini-2.5-flash-preview-tts   # or gemini-2.5-pro-preview-tts
#       languages: [en, es, fr, pt-BR, sq]
#       rate_limit:
#         requests_per_minute: auto
#         min_requests_per_minute: 3
#         max_requests_per_minute: 150

# Reader-customization defaults and locks applied to the exported web ADT.
# `default_settings` seeds the reader's starting values (only overrides
# localStorage when the corresponding key is locked).
# `locked_settings` hides those controls from the reader's settings panel.
#
# Example:
# default_settings:
#   dock_layout:
#     width: full        # compact | full
#     position: bottom   # top | bottom
#     align: spread      # center | spread
#   theme: dark          # light | dark | system
#   icon_size: md        # sm | md | lg
#   reduce_motion: false

locked_settings:
  - dockLayout
  - theme
  - iconSize
  - reduceMotion