DecksMD 2.10.0

Decks 2.10.0 — the AI workbench and card ids in your notes

• A new AI workbench tab lists every generation session with its source, pages, staged, flagged and saved counts, and this week's keep rate. Open it from the Decks panel's AI button, the "Open the AI workbench" command or the status bar, which counts staged and flagged cards while AI is on.
• "Needs attention" in the workbench gathers flagged cards from every session to keep, discard or "Fix in session". "Gap in the last session" counts the concepts in the latest session's PDF that have no card yet, with "Generate for it".
• Sessions save as you go, survive a restart, sync between devices that share the vault, and can be resumed or archived from the workbench. Resume brings back the rounds, kept and discarded cards, flags, the PDF and its chapter selection, and the destination note and profile.
• The AI generator always opens as a tab named after its source and reopens with the workspace. The "AI generator display mode" setting is removed.
• A session reads as a thread of prompts and the rounds they produced. An instruction rewrites the last round and keeps the original behind "Show originals", "Keep all" keeps a whole round, and a selected card expands in place with Keep or Discard.
• The composer has one + menu (note, image, PDF) and one options menu (model, card type, "Include generated cards as context"), and Generate turns into Stop while a round runs. The header holds buttons for the "Staged cards" panel, the destination and Save, and the Clear button is removed.
• After each round, a second pass checks the cards against card-writing rules. A flagged card names the rule, such as "two facts" or "needs the source", and offers a fix such as Split, Make cloze, Rewrite or Add context, with Undo.
• "N need attention" shows only the flagged cards. A round the check could not run on is marked "not checked" and can still be saved.
• Saving leaves out cards the destination deck already holds and cards whose front another deck holds, and says how many. No note is created when nothing is left to write.
• The generator can write multiple-choice questions (card type "Multiple choice"). Questions exam setup could not read are flagged, and fixes include Even out, Replace weak options and Make type-in. They are saved as a heading with a task list into a note on the Exams profile, and the "Exam draft" panel counts the destination's questions by type.
• "Ask" answers questions about the attached PDF or notes, citing pages, and "Make this a card" keeps an answer. "What did I miss?" checks the source against this session, earlier sessions and the destination note, with "Generate for N gaps".
• In a narrow pane, including on Obsidian mobile, the workbench lists one row per session and a session's PDF panel opens over the thread instead of beside it.
• Clicking a session's PDF chip opens a resizable PDF panel with Chapters, Pages, Concepts and, for multiple choice, Blueprint views. Pages shows the PDF with selectable text, and a selected passage can be asked about or turned into cards that keep its page.
• Cards generated from a PDF show a "p. N" chip that opens their page in the panel. The page is saved with the card, as a hidden %%p. N%% comment or in a table row's Notes column.
• Chapters shows a page strip under each chapter, shaded by how many cards cite each page, with a coverage figure and "Generate for the N gaps" for pages with no cards.
• Concepts extracts the examinable concepts from the selected chapters, cached per PDF, and marks each one Uncovered, Thin (one card for several facts), Failing or Covered, with Generate cards, Split or Repair these cards to match. A page grid separates pages with nothing to learn from pages whose concepts have no cards.
• Blueprint plans an exam over the selected chapters, with question counts per section, "Auto-weight by pages" and the destination note's question mix. Attempt defaults can be saved to the Exams profile, and generation runs one round per section.
• A PDF attached from the computer is copied to the plugin folder under ai-sources so a resumed session can reopen it. At startup, a copy no session uses is removed once it is more than 30 days old.
• In review, a card the workbench made from a PDF that has reached the leech threshold shows "Missed N times", "Re-read p. N" and "Repair this card" with its answer. Repair opens Refactor with AI, with that page's text as context.
• Exam results show "Weakest sections" for questions the workbench made from a PDF, with Generate cards or Repair these cards for each section and "Create session from misses". Each missed question shows its page.
• With DecksMD Pro, typed exam answers can be graded "By meaning", which you choose in exam setup, profiles or the blueprint. Answers the check is unsure about, or all of them when it cannot run, go to "Grade these yourself" before the results.
• With DecksMD Pro, the Concepts view also links cards to the concept they test when their wording doesn't name it. A kept card that tests the same fact as one already in the destination note shows "Similar to: …", and nothing is dropped.
• Card ids are now written in the hidden markers next to your cards, so a reworded card keeps its schedule on every device, including the DecksMD app. Existing markers are upgraded in the background, a few notes a minute and never during a sync, so notes change on disk even when you haven't edited them.
• Update Decks on every computer that uses the vault, and update DecksMD to 1.1.0 or later. Until a device updates, a card reworded there after its marker was upgraded can show as new on that device. Its review history is kept.
• Before 3 January 2027, open Decks on each computer that uses the vault and leave it running until Settings → File locations → "Card identity" reads up to date. Until then it shows how many cards still depend on that computer's data. A later release stops reading the old markers.
• When two different cards in a deck get the same card id, a notice names both and says only the first is studied until one is reworded. It shows even with notices turned off, and the duplicate-card notice no longer counts these cards.
• Cards answered in an exam now get their marker, as reviewed cards do. Rewording a cloze line in Decks' card editor keeps the history of each deletion still on it, including when you add deletions.
• Re-importing an Anki package into the same folder keeps progress made in Decks, and cards edited in Anki keep their history. Anki's scheduling applies only to cards not studied in Decks or answered more recently in Anki, manual reschedules in Anki are not imported as reviews, and the same front in two Anki decks no longer gets a " (2)" suffix.
• Restoring a backup first saves a backup of the data it replaces, and changes this device made after the backup, such as deletions, no longer replay over it on the next start. Manual backups and the one taken before rebuilding scheduling include the time in their filename, so they no longer replace the day's automatic backup.
• A card rated Again now comes back before the session ends: once nothing else is left, it is shown again. Burying or suspending a card mid-session takes it out of the remaining count.
• "Rebuild card scheduling" now changes only cards that show as new but have review history, instead of every reviewed card. Profiles with a retention target of exactly 0.5 or 0.995 can be reviewed again, and the field accepts a comma as the decimal mark.
• With "Shuffle questions" off, exams keep the note's order. "Tolerant of small typos" grading also ignores punctuation and a leading "the".
• Exam results and the self-grade step render Markdown, so math shows instead of raw LaTeX. The page-number strip is one scrolling line above Previous and Next, the rings on selected options are no longer clipped, and Submit is hidden while "Were you correct?" is open.
• The forecast, overdue cards included, leaves out suspended and buried cards, as the deck list does. The backlog line simulates each deck with its own daily review limit and FSRS settings, and lines up with the bars by date.
• "Due today", "Due tomorrow" and the forecast's day labels follow the "Study day starts at" setting and your local time, so they no longer show the wrong day around midnight or in time zones far from UTC. Cram and the forecast also count calendar days correctly across daylight-saving changes.
• In statistics, "Today" no longer shows another day's reviews, the week and month figures cover exactly 7 and 30 days, and "All history" covers your whole history instead of the last 30 days.
• Reset card and Reset selected in the flashcard manager now reset the cards, and Reset card in review moves on to the next card.
• Editing a heading card that carries notes, including an exam question with an explanation, no longer fails, and the edit keeps the notes. An edit that would empty either cell of a table row is refused instead of losing the card and its reverse at the next sync.
• Editing a reverse card, opening it in its note and refactoring it with AI now act on the note's own card, the right way round. Refactoring several cards with AI rewrites each note once instead of treating its reverse as a separate card.
• A note that briefly reads as empty, for example during a cloud sync, is synced again on the next pass instead of waiting for its next edit.
• While the deck list is filtered, a folder or section totals only the decks it still shows.
• An image occlusion mask with a missing or invalid position gives no cards until it is fixed, instead of showing as a tiny mask at the edge, and the cards then return with their history.
• Counts use correct singular and plural forms in all 13 languages, such as "(1 card)" in the deck list, and Decks Pro is now called DecksMD Pro throughout.