# Dog-Eared

> A book club discussion planner drawn as a chalkboard menu. The host pastes the members' lines - a passage they marked, a hot take, a question, a reaction - one per line as "Name: text (ch. 5)", names the book, says how far everyone has read and how long the meeting runs (45, 60 or 90 minutes). Lines that point past that chapter or page are held back in a spoiler jar and never sent. The rest become the night's menu: an easy starter, main courses of Socratic questions built on members' own lines, a steel-man side dish that argues the view fewest lines take, and a dessert to close on.

URL: https://dog-eared.skillsafe.ai/
API tutorial: https://dog-eared.skillsafe.ai/api.html

## Free, in the browser
- The line reader: who wrote each line (a "Name:" prefix; group-chat timestamps are dropped), what kind it is (quote, take or question - from punctuation alone), and the chapter or page it points to (ch. 5, chapter 5, p. 88, page 88, pp. 12-14).
- The spoiler fence: "chapter 16", "page 210" or "the whole book". A line pointing past the fence in the same unit is held back and never sent (a line with no chapter or page is sent as written); a line in the other unit is sent and flagged as unchecked.
- The board: tonight's fresh lines before a run, then the courses with their minutes. Minutes are set by the page, not the model: starter 10%, side 15%, dessert 15%, the rest shared by the main courses.
- The chalk tally of who sent lines, the seats to save for people on the guest list who sent nothing, and a warning when one voice sent more than half the lines.
- Exports: a group-chat card (the questions only, or on a needs-more reply the asks for members), a host card (.txt, questions with minutes), Markdown, a CSV of questions, JSON, print; kept menus (in the browser, or in your account when signed in).

## Metered (gpt-terra), one task
- menu: the lines -> a verdict by ordered rules and the courses. Verdicts, first match wins: needs-more (fewer than three lines worth discussing - an opinion, a reading, a reason, a question or a quote; 2-3 questions to send members instead of a menu), light-bite (three to five: two main courses), full-table (six or more: three main courses). Each main course: the line ids it is built on, a dish name, a one-sentence restatement of those lines, a Socratic move (clarify, assumptions, evidence, implications, perspectives - no two courses share one), an opening question and exactly two follow-ups.
The model may use only the lines: no character, event, place, chapter, quote, theme or ending they do not mention, even for a famous book; members by name only, never he or she; no digits that are not in the input; every question open and ending with a question mark. The page checks every reply: course count matches the verdict, every course is built on lines that were sent, no line feeds two courses, moves are distinct, and no capitalised name, number or pronoun appears that the lines do not hold.

## Sources
Built on the SkillSafe skill @neurofoo/socratic - https://skillsafe.ai/skill/@neurofoo/socratic - (Socratic questioning in layers: clarify the claim, probe assumptions, examine evidence, explore implications, alternative perspectives, steel man).