Cosimo User Guide

Cosimo helps you study prompt and response pairs with spaced repetition. It is designed for keyboard and screen reader use.

Quick Start

Cosimo is a portable application. To install it, expand the release zip file into the folder where you want Cosimo to run, then start cosimo.exe from that folder. To upgrade, close Cosimo, expand the new release zip over the existing Cosimo folder, and allow application files to be overwritten. Release packages should not overwrite user data such as cosimo.ini, personal decks, review databases, reports, sidecar audio directories, or backups.

When Cosimo starts, it opens the last deck you used. On a new installation, that is normally decks/deck.cosimo-deck beside the executable. Use File -> Open Deck to open another .cosimo-deck file. If the deck file does not exist, Cosimo asks whether to create it.

To add material, use Card -> Add Card for one card, or Card -> Add Card Batch for repeated entry. A card has a prompt, which is the question or cue you see first, and a response, which is the answer you check yourself against. Keep prompts specific and avoid duplicate prompts.

The ready screen shows a summary, a list of card prompts, and a status bar. The status bar reports total cards, cards available now, and whether a scheduled study session is due or roughly how long remains until the next one. The ready screen refreshes about once per minute. Press F5 to refresh it manually.

Use Start to run the scheduled study session. During a normal study session, Cosimo first shows the prompt. You can rate your prediction:

You can also choose Reveal Response, or press Space, to show the response without recording a prediction. After the response is shown, choose recall feedback:

The recall feedback is what Cosimo uses most directly to schedule the card. Press 0 during a normal study session to end it early. Completed attempts are still saved.

Use Force Start for extra practice when you want to ignore the schedule. Use Study Flagged for cards you have marked for manual attention. Use Start Exam for typed-answer testing from a random sample, and Start Quiz for multiple-choice practice. Exam and quiz sessions are report-only: they do not change the spaced-repetition schedule.

After studying, Cosimo shows a summary. Choose Save Report to write a text report into the reports directory, or Continue to return to the ready screen.

Menu Organisation

How Cosimo Studies Cards

Spaced repetition is a study method based on reviewing material at increasing intervals. When something is new, difficult, or recently missed, it should come back soon. When something is easy and consistently remembered, it can wait longer. The goal is to spend more time on material that needs attention and less time repeating what is already secure.

Cosimo records your answers and ratings and uses that history to decide which cards are due.

Cosimo still uses review database, review history, scheduled review, and review audio for established parts of the application. In those phrases, review refers to the broader spaced-repetition process or to stored study history, not necessarily to one whole session.

Cosimo stores each completed attempt in the review database. For each card, it looks at the review history and estimates when the card should come back.

Cosimo uses an FSRS-6 scheduling algorithm. It tracks each card’s difficulty and stability. Repeated correct answers make the card more stable and lengthen the interval, while missed or wrong answers move the card back towards learning. Short learning steps are automatic: new cards normally use 1-minute and 10-minute learning steps, and review-state cards marked Again use a 10-minute relearning step. Cards marked Again are not automatically repeated at the end of the current session; Cosimo waits until the scheduled learning or relearning step is due.

The four recall buttons have the same labels in every state, but their scheduling effect depends on whether the study item is new, learning, review, or relearning:

Decks first studied with older beta versions may contain existing schedule rows from Cosimo’s earlier FSRS-like scheduler. Cosimo does not silently rewrite all of those due dates when the deck is opened. Each card moves to the current FSRS-6 calculation when it is next answered.

In scheduler terms, a lapse is narrower than any Again answer. It means a card that had already reached review state was marked Again and moved into relearning. Again while a card is new, learning, or already relearning is still recorded as an answer, but it does not add another scheduled lapse.

The exact interval is automatic. You do not need to manage dates yourself. For ordinary review-state cards, Cosimo may vary longer intervals slightly and deterministically. This avoids cards repeatedly clumping on the same future date. Short learning and relearning steps are not varied.

The minimum cards per session controls how many cards Cosimo tries to include when you press Start. For example:

The minimum is not a maximum. Due cards are not hidden just because there are more of them than the minimum.

When a session starts, due learning and relearning cards are shown before ordinary review cards, then new cards follow. Within the learning/relearning and review groups, the most overdue cards are shown first. Cards with the same due time, or due times within about one hour of each other, are ordered randomly within that group. New cards are also ordered randomly, so cards added in thematic groups are not shown in deck order.

Good cards are specific and unambiguous. A prompt should ask for one thing, and the expected response should be short enough that you can judge it consistently. Avoid duplicate prompts: one prompt should identify one card only. Duplicate responses are fine, but generated reverse cards need extra care because a duplicated response would become a duplicated reverse prompt.

Cards and Decks

A deck is a file containing cards. Cosimo deck files use the .cosimo-deck extension.

You can edit cards inside Cosimo, or edit the deck file directly in a text editor. The deck file stores prompts, responses, deck metadata, and card markers such as non-reversible status. Review history and ratings are stored separately.

Deck files also contain a small numbered marker at the end of each card, such as =1=. Leave that number in place when editing by hand. It lets Cosimo keep the card’s review history even if you change the prompt or response text. If you add new cards by hand, end each new card with === instead of inventing a numbered marker. Cosimo will choose a safe unused ID the next time it opens the deck. Use =i= instead of === for a manually added card that should be non-reversible. Cosimo may also write next_card_id in the deck metadata; leave it in place, since it prevents new cards from reusing older IDs after cleanup. When Cosimo opens an older deck, it backs up the original according to your deck backup setting and upgrades the deck file to the current format.

Use Deck -> Deck Properties to view the deck format version and optional metadata fields such as title, author, description, URL, voice choices, and whether automatic audio generation is disabled. Use Deck -> Edit Deck Metadata to edit those metadata fields.

The default deck is decks/deck.cosimo-deck beside the Cosimo executable. On Windows portable installs, keep your decks in the decks folder beside the application. If the default deck is missing, Cosimo creates an empty one. You can open another deck from File -> Open Deck. If you choose a deck file that does not exist, Cosimo asks whether to create an empty deck there.

When Cosimo closes, it remembers the last opened deck and opens it again next time. It also keeps up to eight recently opened decks in File -> Recent Decks, omitting the current deck. If there is only one other recent deck, Cosimo shows it directly in the File menu instead of placing it in a submenu. It stores this and other application settings in cosimo.ini beside the executable. Cosimo also remembers the ready-list filter and sort mode. If Cosimo finds older cosimo-settings.txt or cosimo-last-open-deck.txt files, it migrates them into cosimo.ini and removes the old files.

Adding and Editing Cards

Use Card -> Add Card to add one card. The editor has prompt and response fields. It also has a Mark as non-reversible checkbox. Leave it unchecked for an ordinary card; check it when the card should never be used to generate a reverse card.

For repeated entry, use Card -> Add Card Batch. Enter a prompt and response, then press Ctrl+Enter from either field to add the card and return to the prompt field. Pressing Tab from the response field also adds the card and returns to the prompt field when both fields contain text. Press Escape to close the batch dialogue. Batch add does not show the non-reversible checkbox, to keep the repeated-entry flow short.

Use Card -> Edit Selected Card to edit the selected card from the ready list. After a card has been revealed during a study session, this command changes to Card -> Edit Current Card. You can also choose Edit Card or press e at that point. Cosimo opens the same editor used elsewhere, saves the card text, and returns you to the post-response rating step. The scheduling decision is still made only after you choose the recall feedback rating.

Use Delete or Card -> Remove Selected Card to remove the selected card from the deck. Removing a parent card also removes its generated reverse card from the prompt list, because the reverse card is derived from the parent card. The reverse review data remains in the database, but it is only useful again if you restore a deck backup containing the original parent card.

Use Alt+Enter, Card -> Card Details, or the prompt-list context menu to show details and statistics for the selected card.

Duplicate Prompts and Responses

Avoid duplicate prompts. A prompt should identify one card only. Cosimo refuses ordinary add and edit actions that would create a duplicate prompt. If a deck is edited by hand and duplicate prompts are introduced, Cosimo opens the deck with a warning. Use the Duplicated prompts filter to find them, or use Deck -> Resolve Duplicate Prompts to choose the response to keep for each duplicated prompt; the other cards with that prompt are removed from the deck, and their review data becomes dormant unless the review database is scrubbed. During duplicate resolution, Skip Prompt leaves the current duplicated prompt unchanged and continues with later duplicated prompts. Cancel Resolution stops the process entirely. Prompts already resolved remain changed.

Duplicate responses are allowed. They are often useful when different prompts share the same answer. The restriction is only on generated reverse cards: from each group of cards with the same response, at most one card may have an active generated reverse card. Otherwise, the generated reverse cards would have the same prompt and it would no longer be clear which original card was being asked. Cosimo blocks making or editing reverse cards when that would create this ambiguity. Mark a card non-reversible if its shared response should never be used as a reverse prompt.

Reversible and Non-Reversible Cards

Use Card -> Add Reverse Card, or press Ctrl+R while the prompt list has focus, to add a generated reverse card. A reverse card asks for the original response and expects the original prompt. Reverse cards have their own scheduling state and review history. Cosimo blocks making a card reversible when another card with the same response already has an active generated reverse card, because the reverse prompt would be ambiguous. Reverse cards are marked with (reverse) in the prompt list and can be found with the Reverse cards filter. Removing a reverse card removes only the generated reverse direction; it does not remove the parent card. The reverse card’s review data remains in the database and is used again if you make the parent card reversible later. When a generated reverse card is selected, the card menu and context menu offer removal of that reverse direction, not marking the parent card non-reversible.

Use Card -> Mark Selected Card Non-Reversible when a card should not have a generated reverse direction. This writes a marker into the deck file, removes any active generated reverse card for that parent card, and prevents Make Possible Cards Reversible from adding it again later. Use Card -> Clear Selected Card Non-Reversible Mark to remove that marker. The card then becomes eligible for manual or bulk reversal again, provided its response is not already used by another active generated reverse card. Non-reversible cards are marked with (non-reversible) in the prompt list, can be found with the Non-reversible cards filter, and show the non-reversible marker on the card details state line. When a non-reversible card is selected, the menus offer clearing the non-reversible mark rather than making a reverse card directly.

Use Deck -> Make Possible Cards Reversible to mark every currently eligible card reversible in one operation. Cosimo asks for confirmation and tells you how many cards can be changed. Cards that are already reversible, cards marked non-reversible, and cards whose response group already has an active generated reverse card are excluded. If review database backups are enabled, this operation refreshes the retained review-database backup before writing and the confirmation warns that the retained backup will be overwritten.

Use Deck -> Undo Last Bulk Reversal to remove only the reverse cards created by the latest bulk reversal. Manually created reverse cards and earlier bulk batches are kept. Parent cards and reverse review history remain in the database.

Suspending, Deferring, and Flagging

Use Card -> Suspend Selected Card, the context menu on the prompt list, or Ctrl+S while the prompt list has focus, to keep the selected card in the deck while preventing it from appearing in normal or forced study sessions. The prompt list marks suspended cards with (suspended). Suspended cards can still be found with the Suspended cards filter and unsuspended from the same menu, context menu, or shortcut. Suspended cards are also ignored when the status bar reports whether a review is due now or when the next review is due.

Use Card -> Defer Selected Card until Tomorrow, or the context menu on the prompt list, to postpone the selected card temporarily without changing its stored schedule. A deferred card remains in the deck and returns automatically when the deferral expires. While deferred, it is marked with (deferred), appears in the Deferred cards filter, and is excluded from scheduled sessions, forced sessions, ready-screen counts, and status-bar due calculations. Use Card -> Clear Selected Card Deferral, or the context menu, to make it available again immediately.

During a study session, before revealing the answer, use Ctrl+D or the Defer Card until Tomorrow button to postpone the current card and move on. Deferral is not available after the answer has been revealed. If every card in the session is deferred before any answer is recorded, Cosimo shows a notice that no cards were studied.

Use Card -> Flag Selected Card, the context menu on the prompt list, or Ctrl+F while the prompt list has focus to flag or unflag the selected card. During a study session, Ctrl+F flags or unflags the current card. The review screen also has a Flag/Unflag button in both phases, after the relevant rating buttons. Flagging is for manual follow-up: use it when a card may need editing, feels confusing or ambiguous, has a dubious answer, or otherwise needs attention after the session. Flagging does not affect scheduling or whether the card appears in normal study. Flagged cards are marked with (flagged) in the prompt list and can be found with the Flagged cards filter.

The Ready Screen

If the minimum cards per session causes Cosimo to add extra cards that are not due yet, the scheduled count also shows the due-card count in parentheses. For example, Scheduled now: 10 (6 due) means a study session would include 6 due cards and 4 extra cards to reach the minimum.

The main window status bar gives a short overview: total cards, cards due now, and either that a review is due now or roughly how long until the next scheduled review. Durations are rounded up to speech-friendly units such as 2 days, 4 weeks, or 3 months. During a session, the status bar changes to session context, such as scheduled study session, unscheduled study session, exam, or quiz, with current progress.

While the ready screen is open, Cosimo refreshes the ready summary, list filters, and status bar about once per minute. A card that becomes due while the application is open will therefore become available for Start without closing and reopening the deck.

Use View -> Deck Order, View -> Next Scheduled Review, View -> Last Reviewed, View -> Prompt Text, View -> Response Text, or View -> Estimated Recall to choose how the prompt list is sorted. Use View -> Ascending or View -> Descending to choose the direction. The checked View menu items show the current sort criterion and direction. When you choose a different sort criterion, Cosimo resets the direction to that criterion’s default. Last reviewed defaults to descending, putting the most recently reviewed cards first and new cards at the bottom; ascending last reviewed puts the oldest reviewed cards first and still leaves new cards at the bottom. Estimated recall uses the same scheduling estimate as the learning dashboard; ascending estimated recall puts cards estimated as less likely to be remembered first and leaves cards without an estimate, such as new cards, at the bottom.

The text filter narrows the list by matching text in the prompt or response. The filter combo box controls which type of cards are shown, including flagged cards, duplicated prompts, duplicated responses, non-reversible cards, deferred cards, problem cards, asymmetry issues, high-confidence failures, and low-confidence successes. Its counts follow the current text filter. The duplicated-response filter is especially useful before making cards reversible.

The Card menu can also act on the cards currently shown in the ready list. Use Flag All Shown Cards or Unflag All Shown Cards to mark a whole filtered set for manual follow-up, for example all problem cards before choosing Study Flagged. Suspend All Shown Cards and Unsuspend All Shown Cards work the same way. Cosimo asks for confirmation and reports how many shown cards will actually change; cards already in the requested state are left unchanged.

The Problem cards filter shows active study items that appear unusually troublesome in the current deck. It waits until an item has at least 4 attempts, then uses review behaviour rather than raw age. A card can appear there if at least 2 of its last 5 attempts were Again, if its lifetime Again rate is in the deck’s highest 20% and at least 25%, if its scheduled lapses per attempt are in the deck’s highest 20% and at least 20%, or if its scheduling difficulty is in the deck’s highest 20% and at least 8 on the FSRS 1-10 difficulty scale. Suspended and currently deferred study items are not shown there.

The Asymmetry issues filter is for reversible cards. When both directions have at least 4 attempts and one direction is substantially weaker, the filter shows the weaker study item. Cosimo compares a lightly smoothed strict recall rate, where only Good and Easy count as recalled, and treats a gap of 30 percentage points or more as substantial. This is useful when, for example, seeing an Indonesian word and remembering English is easy, but producing the Indonesian word from English is still unreliable.

The High-confidence failures and Low-confidence successes filters use recent confidence-rated attempts. High-confidence failures shows active study items where you recently chose Probably know or Certainly know but then answered Again; the latest confidence-rated attempt is enough, or at least 2 of the last 5 confidence-rated attempts. Low-confidence successes shows active study items where at least 2 of the last 5 confidence-rated attempts were Don't know or Probably don't know followed by Good or Easy. Suspended and currently deferred study items are not shown in either filter.

The card details dialogue shows the prompt, response, card ID, current card state, attempt count, last attempt, scheduling state, availability, estimated recall now, recent recall, lapse summary, and attempt history for the selected card. Card state combines visible markers such as suspended, flagged, deferred, reverse, has reverse, and non-reversible on one line. Availability explains whether a card is new, scheduled for later, ready for an ordinary scheduled review, or waiting for a learning or relearning step. When there is a scheduled time, the availability line includes both the relative time and the exact UTC timestamp.

Generated reverse cards show the parent card ID, because editing the parent card in the deck also changes its generated reverse direction. Date fields in the summary, such as last attempt and schedule last updated, are shown first as a complete relative duration, such as 1 day, 1 hour, 1 minute, 1 second, followed by the exact UTC timestamp in parentheses. Past dates use the same complete relative duration followed by ago. The attempt history list at the bottom uses exact UTC timestamps only.

Study Sessions

Start uses the schedule. It shows due cards first. If fewer cards are due than your configured minimum cards per session, Cosimo randomly adds extra cards from the rest of the deck.

Force Start ignores the schedule. It puts every card that is not suspended or deferred into a random order. Use this when you want to study even though no cards are due, or when you want a longer practice session.

Study Flagged starts an unscheduled study session through flagged cards only. This session includes flagged cards whether or not they are currently scheduled, but excludes cards that are suspended or deferred. The order is randomised, and the session does not clear the flags; remove flags manually when the cards no longer need follow-up. The button is disabled when there are no flagged cards available to study.

When a prompt is shown, you can rate how likely you think you are to know the response:

You can also choose Reveal Response, or press Space, to show the response without recording a prediction. This is stored as Not answered in the session summary.

After you rate your prediction or reveal the response directly, Cosimo shows the response. Then rate what actually happened:

This feedback is what Cosimo uses most directly to schedule the card. The same recall rating can have a different scheduling effect depending on whether the card is still in a short learning step or has already reached ordinary review state; see How Cosimo Studies Cards for the details.

If you notice a mistake after the response is shown, choose Edit Card, use Card -> Edit Current Card, or press e. Cosimo opens the same editor used elsewhere, saves the card text, and returns you to the post-response rating step. The scheduling decision is still made only after you choose the recall feedback rating.

If an older session or future study mode contains repeated questions for the same study item, Cosimo shows the fuller distinction between unique cards, questions answered, time per question, and time per unique card. The normal current study modes do not repeat Again cards in the same session. A card and its generated reverse direction count as separate studied cards.

Cards deferred before the answer is revealed are not counted as answered questions in the session summary.

Choose Save Report to write a text copy of a normal session summary to the reports directory beside the executable. Cosimo creates the directory if needed. Normal report filenames use the deck file name without its extension and the session finish time in UTC, for example Indonesian-2026-05-03-06-04.txt. If a report with that name already exists, Cosimo adds a numeric suffix instead of overwriting it. The report text includes the full deck path.

Exams and Quizzes

Start Exam samples from the active deck instead of using the schedule. Suspended and deferred cards are excluded. Cosimo asks you to choose a number of questions or a percentage of active cards, then randomly selects and orders the cards to reduce cues from deck order.

If the deck has generated reverse cards, exam setup includes a study-direction choice: standard only, reverse only, or both directions. If the deck has no generated reverse cards, this choice is hidden and the exam uses the standard direction.

During an exam, Cosimo shows each prompt and asks you to type the response. It does not show feedback while the exam is running. Exact answers are counted as correct automatically. At the end, for each answer that was not an exact match, Cosimo shows the prompt, your answer, and the expected response, then asks whether the answer should count as correct.

If the Exact checkbox is selected when starting the exam, non-exact answers are not graded manually. They count as incorrect, and Cosimo shows each one for review with the prompt, your answer, and the expected response. Choose Next or press Alt-N to continue reviewing inexact answers, or choose Summary to go straight to the exam summary.

If the Timed exam checkbox is selected, enter a whole number of minutes for the exam. The remaining time is shown in the status bar. Cosimo plays a very short tone at each 10% boundary of the time limit. When time runs out, any text currently in the answer field is submitted, four short tones play, and Cosimo shows a time-out notice with the number of submitted answers. The notice starts with its message focused; tab to Next to continue to grading, answer review, or the summary.

When a non-timed-out exam finishes with at least one submitted answer, Cosimo plays two long tones and shows an Exam Finished notice before grading, answer review, or the summary. This gives users who are relying on exam audio a clear boundary between answering questions and checking answers.

During an exam session, choose End Exam to stop after the answers submitted so far. If no answers were submitted, Cosimo shows a notice.

Exam sessions show a separate exam summary. It reports selected questions, submitted answers, correct answers and percentage correct, exact matches, accepted non-exact answers, incorrect answers, total time, time per question, whether the exam ended early, and whether a timed exam expired. Exam summaries do not change the review database. Choose Save Report to write a text copy of an exam summary to the reports directory. Exam report filenames add -exam, for example Indonesian-exam-2026-05-03-06-04.txt.

Start Quiz is also unscheduled and report-only. It is intended for multiple-choice recognition practice, such as preparing for external tests that use quiz-style questions. Cosimo samples active cards randomly and excludes suspended and deferred cards. Unless the requested quiz size makes this impossible, it avoids putting both directions of the same reversible card in one quiz, so one question does not cue the other direction.

If the deck has generated reverse cards, quiz setup includes a study-direction choice: standard only, reverse only, or both directions. If the deck has no generated reverse cards, this choice is hidden and the quiz uses the standard direction.

During a quiz, Cosimo shows the prompt and a list of response choices. Choose one answer and submit it. Feedback is withheld until the end. After the quiz, Cosimo shows each submitted answer with the prompt and your selected response. If the selected response was wrong, it also shows the correct response. Choose Next or press Alt-N to move through the answer review, or choose Summary to go straight to the quiz summary. Quiz mode does not record attempts, change card schedules, or add failed answers back into the same session. It needs at least four distinct answer choices in the current study direction, so forward questions draw choices from responses and reversed questions draw choices from prompts. Prompts from cards marked non-reversible are not used as choices for reversed questions.

Quiz sessions show a quiz summary with selected questions, submitted answers, correct and incorrect answers, percentage correct, total time, time per question, and whether the quiz ended early. Quiz summaries do not change the review database. Choose Save Report to write a text copy of the quiz summary to the reports directory. Quiz report filenames add -quiz, for example Indonesian-quiz-2026-05-03-06-04.txt.

Learning Dashboard

Use Study -> Learning Dashboard, or press Ctrl+L, to show a deck-wide learning summary. The dashboard is split into separate read-only sections so you can tab between the summary, distribution, workload, diagnostics, and activity areas instead of reading one long field. It reports several kinds of information:

The recall section is an estimate from Cosimo’s scheduling model, not a direct test result. For each active card with scheduling data that is not suspended or deferred, Cosimo uses the last review time, the stored stability, and the actual scheduled interval to estimate how likely you are to recall the card now. New cards without scheduling data, suspended cards, and currently deferred cards are not included in this estimate. Learning and relearning cards use their short scheduled interval as the effective 90% point, so a card due after a short failed-card step does not look safer than it is.

Available now counts active cards that are not suspended or deferred and can be studied immediately. This includes new cards, which Cosimo can place in a study session immediately, and cards with existing schedule data whose due time has arrived. Of these, already scheduled and due now is a subset of Available now: it counts only available cards with existing scheduling data. It uses the same due-time boundary as Available now, so a card due at the exact time the dashboard was generated is included. The dashboard puts Available now first because it is the operational number: how many cards can actually be studied now.

Average estimated recall is the average of those per-card estimates. The dashboard also shows the median estimate and the weakest estimated cards so an average does not hide a small set of weak material. The weakest-card average is the average estimated recall of the lowest 10% of estimated active cards, rounded up so at least one card is included. Likely known counts estimated cards at or above 90% recall. Mostly stable is 70-89%, Fragile is 50-69%, and Very low is below 50%. At-risk cards counts the fragile and very-low bands together. These numbers are best used for triage and trends, not as a promise that exactly that percentage of cards will be answered correctly in the next session.

The workload forecast shows how many active cards are scheduled now and within the next 24 hours, 3 days, 7 days, and 30 days. These counts are cumulative: a card available now is also included in the later horizons. Suspended cards and deferred cards are excluded. When Cosimo has recorded session timing, the forecast also estimates study time from the current average time per unique card studied. This is only an estimate: it reflects repeated attempts in previous recorded sessions, but it cannot know how many learning or relearning steps will become due during a future session.

The Summary section describes the shape of the next 7 days of workload only when at least 5 cards are forecast. If at least half of the 7-day workload is due within 24 hours, Cosimo says most of it is due within 24 hours. If at least three quarters is due within 3 days, Cosimo says most of it is due within 3 days. Otherwise, Cosimo says the workload is spread across the week. When timing data exists for the last 7 days, the Summary also compares the estimated 7-day workload with your recorded study time in the last 7 days. If the difference is under 5 minutes and under 20% of the recent recorded time, Cosimo describes the estimate as close to recent study time; otherwise it describes it as higher or lower and shows both durations.

The calibration section treats your prediction before seeing the response as a target interval for whether you will recall the answer, not as a precise percentage. Don't know means a 0-5% recall target, Probably don't know 5-32%, Probably know 68-95%, and Certainly know 95-100%. The middle range is deliberately uncovered; reveal without a confidence rating if you do not want to make a prediction. The recall rule is configured in Tools -> Options. The lenient rule counts Hard, Good, and Easy as recalled, while Again counts as not recalled. The strict rule counts only Good and Easy as recalled. The dashboard shows the active rule before the calibration figures. The target recall range is the average of the target intervals for answered confidence ratings. The observed recall rate is what actually happened under the current recall rule. Calibration is within target if observed recall falls inside the target range, under-confident if observed recall is above the range, and over-confident if observed recall is below it. The per-confidence lines show the same comparison for each confidence rating. Cards revealed without a prediction are counted separately.

In the diagnostics section, High-confidence failures means you chose Probably know or Certainly know and then answered Again. Low-confidence successes means you chose Don't know or Probably don't know and then recalled the answer under the active recall rule.

The recall trend section counts stored answers for cards in the current deck. Overall recall uses the ordinary dashboard rule that Hard, Good, and Easy count as recalled, while Again counts as not recalled. The feedback lines show percentages first, followed by counts in parentheses. The recent window compares the last 7 days with the previous 7 days; Cosimo only makes a qualitative trend statement when both windows contain at least 5 answers. A change of 5 percentage points or more is described as improved or declined; smaller differences are described as broadly stable. Reverse-card Again rate trends use the same 7-day windows, 5-answer minimum, and 5-point threshold, but a falling Again rate is better and a rising Again rate is worse. The lapse lines come from the current scheduling state of active cards and help identify whether the deck contains material that repeatedly falls back from review state into relearning. These scheduled lapses only count Again on cards that had already reached review state; ordinary Again rates are reported separately. Retention statements about recent learning count recently learned cards that have not had a later Again answer. Cosimo needs at least 5 recently learned cards before it describes recent retention; 90% or more retained is retained well, 70-89% is mostly retained, and below 70% is shown as follow-up lapses.

When timed session data exists, the Summary section includes average study time per active study day so this headline metric is visible without tabbing to the full activity section.

Cosimo records the first time the review database sees each physical deck card. This is not necessarily the card’s true creation date: existing decks will have their current cards recorded when the feature first sees them. For this reason the dashboard describes it as cards first seen by this review database. Once the database has older observations, the Summary can show a Deck flow line for the last 7 days: how many deck cards were first seen, and how many standard-direction cards reached review state. This is descriptive information only. It is meant to help you judge deck growth and learning throughput without Cosimo telling you whether to add more or fewer cards.

The study activity section is based on recorded attempts and study-session summaries. Stored answers counts individual stored answers. Days with answers counts UTC days on which at least one answer was recorded. Session, answer, and time totals labelled as recorded sessions come from completed session summaries. Timed study days counts UTC days with at least one completed session summary and is the basis for average study time per active study day. Recent recorded study time and recent recorded questions compare the last 7 days with the previous 7 days. This describes actual study activity recorded in completed sessions; it is not a historical forecast of how many cards were scheduled. The Summary describes recent study time as higher, lower, or broadly similar to the previous 7 days when timed session data exists; differences under 5 minutes and under 20% of the previous window are treated as broadly similar. If older data predates session summaries, the dashboard shows Older answers without session timing so the stored-answer total is not confused with the smaller timed-session total. That line is hidden when the value is zero. The same section also reports how many physical deck cards have been first seen by this review database, first-seen counts for the last 7 days, previous 7 days, and older observations when enough history exists, and active standard deck cards that have not yet been studied.

Use the dashboard’s Save Report button to save the complete dashboard report in the reports directory. The saved text uses Markdown-style section headings and includes every dashboard section, even though the dialogue shows them in separate fields. The report file name uses the deck name and the UTC time when the dashboard was opened, and the report includes the full deck and review database paths. Use Copy to Clipboard to copy the same complete report text to the clipboard.

If the deck has generated reverse cards, the dashboard adds a standard/reversed comparison section, because one direction can be much easier than the other. If the two estimated recall averages differ by less than 5 percentage points, Cosimo describes the directions as similar.

Audio

Audio support has two separate parts. Playback uses existing WAV files while you are studying, previewing cards, or reviewing exam answers. Automatic generation uses voice.exe, either from the command line or through Tools -> Generate Audio, to create those WAV files from installed speech voices.

Review audio playback is optional and currently supports WAV files only. Configure it in Tools -> Options -> Audio during study:

The separate Guiding tones for automatic audio option adds short orientation cues when automatic audio during study is on: one short tone before prompt audio, two short tones before response audio, and a three-note arpeggio when the review summary is shown. The option is off by default.

Ctrl+P activates the same play/stop action when review audio is available. On the ready card list, Ctrl+P or the Play/Stop Audio button previews the selected card by playing its prompt audio and then its response audio. Moving to the next card, revealing the response, finishing the session, or closing the window stops any current card audio immediately. In exam mode, prompt audio is available while answering the question.

Tools -> Options -> Automatically play audio during exam grading and review controls what happens after an exam answer has been submitted and Cosimo is showing a grading or review dialogue. No automatic audio leaves that phase silent unless you ask for audio manually. Question, Answer, and Both automatically play the prompt, the expected response, or the prompt followed by the expected response. In grading and review dialogues, Ctrl+P always plays the prompt followed by the expected response when those audio files exist, even when automatic exam grading/review audio is off.

Tools -> Options -> Automatically play audio during quizzes controls quiz audio. No automatic audio keeps quiz sessions silent unless audio is played manually in the answer review. Prompt plays prompt audio when each quiz question appears. Choices plays the audio for a response choice when that choice is selected and stops it when the choice list loses focus or the selection changes. Both combines those behaviours. During quiz answer review, the same setting controls automatic prompt and correct-response playback; Ctrl+P plays the prompt followed by the correct response when those audio files exist.

Audio files live in the same-stem directory beside the deck. For card 1 in Indonesian.cosimo-deck, use Indonesian/1.prompt.wav for the prompt side and Indonesian/1.response.wav for the response side. Generated reverse cards reuse the parent card’s files with prompt and response swapped. The audio generation helper can write and overwrite files in this directory, so keep separate copies of any recordings you created yourself.

Cosimo’s automatic deck and review-database backups do not include sidecar audio files. If you record or edit your own WAV files, back up the sidecar directory yourself before using Tools -> Generate Audio or voice --overwrite.

Windows packages also include voice.exe, a helper for generating these WAV files from installed OneCore or SAPI voices. Run it from a command prompt. For example, if voice.exe is beside cosimo.exe, this command reads decks/Indonesian.cosimo-deck and writes audio files under decks/Indonesian/:

voice Indonesian --prompt-voice "Indonesian voice name" --response-voice "English voice name"

You can also use Tools -> Generate Audio from Cosimo after opening a deck. That menu item is available only when the deck has Prompt voice, Response voice, Prompt voice engine, and Response voice engine metadata, so Cosimo does not silently generate audio with the wrong system default voice. If the menu item is disabled, use Deck -> Edit Deck Metadata to choose those voices and engines, and check whether the deck is marked as a manual audio deck. It shows a progress dialogue while voice.exe runs for the current deck and writes the helper output to a text file in the reports directory. The menu item does not overwrite existing audio unless the helper manifest says the generated card text, selected voice, or selected voice engine has changed. If no manifest exists yet, matching existing WAV files are regenerated once so the new manifest reflects the current card text, voice selection, and engine selection. If a manifest exists but an individual audio file has no entry, that file is left untouched and added to the manifest so later card text, voice, or engine changes can be detected. When it finishes, Cosimo reports the number of cards for which audio was generated, how many audio files were new, how many were regenerated because card text, voice selection, or engine selection changed, how many untracked files were regenerated to seed a missing manifest, how many existing files were added to the manifest, and how many files were skipped. If the report mentions files force-regenerated by explicit overwrite, that refers to the command-line --overwrite option, not normal regeneration of stale managed audio.

This regeneration policy is intended for audio produced by voice.exe. If you manually recorded or edited a file whose name matches a selected card side, a later deck edit, voice change, or engine change can cause Tools -> Generate Audio to overwrite that recording. Use --dry-run first if in doubt, and keep your own backup of recordings before generating audio. For a deck whose audio you maintain manually, enable Manual audio deck (protect against automatic audio generation) in Deck -> Edit Deck Metadata. This writes audio_generation=disabled to the deck header, disables Tools -> Generate Audio, and makes voice.exe refuse to generate audio for that deck. It does not stop playback of existing audio.

Use voice --list-voices to see installed voices. For language learning, choose a voice that matches the language of each side: if the prompts are Indonesian and the responses are English, use an Indonesian prompt voice and an English response voice. If both sides are in the same language, --voice "Voice name" uses the same voice for both. Use --engine onecore or --engine sapi to list or generate with a specific engine; --prompt-engine and --response-engine can select different engines for the two sides.

Deck metadata can also store default voice filters and voice engines. Set Prompt voice, Response voice, Prompt voice engine, and Response voice engine in Deck -> Edit Deck Metadata, or add prompt_voice=, response_voice=, prompt_voice_engine=, and response_voice_engine= to the deck header. voice.exe uses those defaults when the matching command-line option is not supplied. The same editor can mark a deck as manual audio, which adds audio_generation=disabled. When voice.exe is available beside cosimo.exe, the metadata editor lists installed voices for the selected engine. The voice fields are still editable, so you can type a filter manually if listing fails or if you want to use a value not shown in the list. The engine selector can list OneCore voices, SAPI voices, or both; when you choose a listed voice, Cosimo saves the matching engine automatically. If you type or edit a voice filter manually, choose onecore or sapi before saving so voice.exe knows which engine to use. Command-line options still override deck metadata, but they do not override audio_generation=disabled.

The helper writes an audio-manifest.tsv file in the audio directory. It uses that manifest to skip audio whose card text, selected voice, and selected voice engine have not changed, and to regenerate files when any of them has changed. Existing audio files without manifest entries are regenerated if there is no manifest at all; if a manifest already exists, they are left alone on the current run but added to the manifest so future changes can be detected. Use --overwrite if you want to refresh all selected files immediately. --dry-run shows what would be generated without writing files.

Files, Backups, and Maintenance

Back up both the deck file and its matching review database. The deck file contains your cards, deck metadata, and non-reversible card markers. The review database contains scheduling state, ratings, and review history.

Cosimo also creates automatic deck-content backups. Before it overwrites an existing deck, it copies the current deck into the app-level backups directory beside the executable. Backup files have rising numbers, for example backups/000001-deck.cosimo-deck. The backups/index.txt file records each backup filename and the original deck path.

For another deck, the review database is beside the deck and uses the same base name. For example, spanish.cosimo-deck uses spanish.reviews.sqlite3; its sidecar audio directory, if present, uses the same base name as the deck.

Close Cosimo before copying or restoring backup files. Restoring only the .cosimo-deck file restores the cards but starts them without their previous review history. Restoring both files preserves the cards and their scheduling history.

Cosimo checks the review database integrity when opening a deck. If Cosimo says the review database failed an integrity check, close Cosimo and restore the matching .reviews.sqlite3 file from your backup.

Automatic deck backups contain only the deck text, not review history. To restore one, close Cosimo and copy the chosen backup over the active .cosimo-deck file. To restore scheduling history, restore the matching .reviews.sqlite3 file from your own backup or from Cosimo’s retained review-database backup. Automatic deck backups and review-database backups do not contain sidecar audio; restore audio from your own backup of the sidecar directory.

Automatic full deck backups are different: when enabled, they are retained compressed bundles that include the deck, sidecar directory, and review database. They are intended as a daily safety net for the currently used deck, not as a replacement for manual exports before risky changes.

File -> Export Full Deck writes a compressed .zip bundle containing the current .cosimo-deck file and its same-stem sidecar directory, including audio files and audio-manifest.tsv when present. It does not include the review database. Cosimo suggests a timestamped filename and lets you choose where to save the bundle. Use this export when you want to share deck content without your personal study history.

File -> Export Full Deck with Study History writes the same bundle plus a consistent snapshot of the review database. Use it for moving your own study setup to another machine or making a complete manual backup. The archive contains personal study data such as schedules, ratings, review history, suspensions, flags, and deferrals.

File -> Full Deck Backup writes the deck-and-sidecar bundle automatically into the app-level backups directory and records it in backups/deck-bundle-index.txt. It does not include the review database.

The optional daily automatic full deck backup also writes into backups, but uses backups/automatic-deck-bundle-index.txt and keeps a retained backup per deck path. It includes the review database.

File -> Restore Full Deck restores one of these compressed bundles into Cosimo’s decks directory. Cosimo first checks the bundle, then shows the deck file name, number of cards, number of audio files, file count, and whether the bundle contains study history. If existing files would be overwritten, Cosimo warns you and writes a full compressed backup of the existing target deck before restoring. If sidecar files would be overwritten but no matching target deck exists to back up, Cosimo refuses the restore instead of overwriting them.

Restoring a bundle can still cause data loss. It may replace newer card text, newer sidecar audio, a newer audio-manifest.tsv, or newer scheduling and rating history in the .reviews.sqlite3 file. A bundle without study history will not restore scheduling or rating history; a bundle with study history will replace that history with the snapshot contained in the zip. Close other programs that may be using the same deck or review database before restoring.

Use Tools -> Scrub Review Database to permanently delete dormant review database rows for cards that are no longer in the current deck and for inactive generated reverse cards. Active cards, active generated reverse cards, current schedules, current suspensions, and current flags are kept. Non-reversible markers are stored in the deck file, not the review database. If review database backups are off, Cosimo warns that the scrub is irreversible. If review database backups are enabled, Cosimo refreshes the retained review-database backup before scrubbing and warns that the retained backup will be overwritten. After deleting inactive rows, Cosimo vacuums the database so the SQLite file can reclaim unused space.

Use Tools -> Reclaim Database Space to run the same SQLite vacuum without deleting dormant review data. This rewrites the review database file and may take some time on a large database. If review database backups are enabled, this command does not overwrite the retained backup; if backups are off, Cosimo reminds you to make sure you have a backup if concerned.

Use Tools -> Options -> Delete Old Backups to keep only the newest automatic backup for each original deck path. Cosimo asks for confirmation before deleting or renumbering backup files.

Use Tools -> Wipe Backup Directory only when you deliberately want to empty the application backups directory. Cosimo warns before doing this because the command deletes every file and folder in that directory, including files that may not have been generated or tracked by Cosimo. The command keeps the backups directory itself, but its contents are not recoverable from Cosimo after deletion.

Accessibility

Cosimo is designed for keyboard and screen reader use. Most study actions can be completed without opening menus:

Review audio can be used manually or automatically. When automatic audio during study is enabled, the optional guiding tones provide non-speech cues: one short tone before prompt audio, two short tones before response audio, and a three-note arpeggio when the review summary is shown. Timed exams also use tones to mark time progress and timeouts.

The learning dashboard is split into separate read-only sections so screen-reader users can tab between summary, distribution, workload, diagnostics, and activity instead of reading one long field. The dashboard also has Save Report and Copy to Clipboard buttons for reviewing the full report outside the dialogue.

The main status bar reports deck status and review timing in short, speech-friendly terms. Scheduled times in card details include relative durations before exact UTC timestamps, because relative times are easier to understand with speech.

Known issue: when you tab or shift-tab into the prompt list, some screen reader and wxWidgets combinations announce the selected item twice. Once you are in the list, arrow-key navigation should work normally.

Settings and Language

Set the minimum cards per session in Tools -> Options. The options dialogue can also configure review-availability sounds, review audio, guiding tones, exam grading/review audio, the learning-dashboard calibration recall rule, deck and review-database backups, language, and minimise-to-tray behaviour.

Sound notifications can be set to Always, When Cosimo is visible, When Cosimo is in the system tray, or Never. When the chosen policy permits sound, Cosimo plays a short sound when the timed ready-screen refresh notices that reviews have become available, or that more reviews have become available since the last automatic refresh. User-triggered changes, such as adding or removing cards, and manual F5 refreshes do not play this sound. Cosimo also stays silent while a study session is active, even if more cards become due during that session.

With minimise-to-tray enabled, using the normal window minimise action hides Cosimo in the system tray instead of leaving it on the taskbar. Activating the tray icon restores the window, and the tray context menu contains Restore and Quit. Closing Cosimo still exits the application.

Cosimo currently supports English, Spanish, and Galician. Change the language in Tools -> Options. The language choice is saved and used next time. If no language has been saved yet, Cosimo uses Spanish when the system locale is Spanish, Galician when the system locale is Galician, and otherwise falls back to English. Unknown locales also fall back to English.

Translations are maintained in gettext-style .po files, with a small amount of source-code wiring for each new language. If you want to work on a translation, contact David Picón Álvarez first at @modulux@node.isonomia.net so the work can be coordinated.

The Help menu opens the local user guide, the project wiki, and the new ticket page for bug reports. Bug reports are handled through Fossil tickets on the project site.

Tips

Why the Name Cosimo?

Cosimo is named in honour of Cosimo Piovasco di Rondò, the protagonist of Italo Calvino’s novel The Baron in the Trees. Calvino’s Cosimo is a notorious polymath and polyglot, which makes his name a fitting one for a study tool.