Clean and refactor study decks

Export a baseline

Save the original deck with a date before editing. Keep review history in the original application if it matters; a plain text export usually contains content rather than scheduling state.

Save an untouched export and record the date, note type, card count, and destination application before making changes. Where the application provides a backup that preserves scheduling history, keep that as well as a text export. A text file may preserve prompts and answers without preserving everything associated with them. Open the export and confirm that it contains the fields you intend to edit. Work on a copy with a distinct name. This baseline lets you compare the result and recover from a mistaken merge without relying on memory. Cleanup should begin with a recoverable state, especially when the collection represents several months of review work.

Normalize the input

Use Prompt :: Answer lines. Remove accidental extra spaces and inconsistent separators. Normalization should make comparison possible without changing mathematical signs or meaningful punctuation.

Normalize layout differences before judging whether content is duplicated. Repeated spaces and different separators can make identical prompts look different to a person or a program. The cleaner normalizes prompt case, Unicode compatibility forms, and repeated whitespace for comparison, while preserving the output cards. Still inspect symbols and domain-specific capitalization yourself: a normalization useful for ordinary prose may obscure a meaningful distinction in a technical notation. Start with a small sample containing those edge cases. Keep a readable master that uses one separator convention consistently, and avoid hiding extra fields inside answers simply because the current input format has only a few columns.

Separate duplicate types

Identical prompt, answer and tag records can be removed safely. Identical prompts with different answers may represent ambiguity or different contexts. The cleaner preserves those conflicts for manual inspection.

An exact repeated record is different from a shared prompt with different answers. The cleaner removes repetitions of the same normalized prompt, answer, and tag combination. It preserves conflicting answers and reports them for inspection. Do not merge a conflict automatically: the answers may refer to different contexts that the prompt failed to name. For example, an abbreviation can have two legitimate meanings in different subjects. Repair those prompts by adding the subject rather than choosing an answer arbitrarily. Records with different tags may also remain separate. Decide whether that difference is intentional before treating a lower output count as evidence of successful cleaning.

Inspect long answers

Answers over twenty-five words are flagged, not automatically shortened. A long answer may be necessary, but often contains several independently testable facts. Split only when the learning target remains meaningful.

The long-answer warning is a review cue, not a rule that every answer above twenty-five words is wrong. Read the prompt and identify the response it requires. A short causal explanation may need several clauses; an overloaded list may need to be split into several cards. If you divide a card, preserve the relationship between its parts with an overview question or an application exercise. Do not shorten an answer by removing the conditions that make it correct. Conversely, move background exposition into a companion note when it is useful context but not part of the response you need to retrieve.

Repair the prompt

If two valid answers share one question, add context or create separate cards. Deleting one answer without checking can silently erase an important distinction.

A repaired prompt should make sense when shown alone and out of order. Replace unclear references such as “this process” with the process name or a brief setting. State which distinction, direction, or quantity the learner must retrieve. Then write a short grading criterion beside the answer. Try the revised card without looking at neighboring records. If two knowledgeable people could answer it differently and both be right, either accept those alternatives explicitly or narrow the question. Better wording often reduces apparent difficulty more effectively than repeatedly reviewing the same ambiguous text, because the learner can finally identify the intended task.

Protect source references

Retain tags or notes that identify the relevant course and source. A duplicate definition may still need multiple references. Decide how to combine that context before removing records.

Keep enough provenance to resolve future disagreements. A source title, section, lecture date, or stable page reference can live in a dedicated field or companion document. When two answers conflict, consult that source before rewriting either. If the course has changed its convention, label the old material rather than silently blending it with the new. Avoid putting essential answer conditions only in tags, because a review interface may hide tags. During cleanup, inspect a few cards whose sources you rarely open. Broken or vague references are easier to fix now than during a rushed revision session shortly before an assessment.

Import cautiously

The audit report is not part of the deck. Export only the cleaned TSV section when importing. Test a small batch and check the destination application's duplicate-matching behavior.

The audit output includes a report after the cleaned records. Remove that report before importing the deck into a tool that expects only cards. Then pilot a few changed records and inspect how the destination matches existing notes. Some import workflows create new notes where you expected updates, potentially separating edits from prior review history. Confirm front, back, tags, and duplicate behavior in a small test collection. Keep both the pre-cleanup export and the cleaned candidate until you have reviewed the destination. An apparently successful import notification does not establish that the intended records were updated or that scheduling data survived.

Schedule maintenance

Review recurring misses and confusing prompts weekly. Small corrections are easier to verify than a whole-deck rewrite before an exam. Track whether the changes reduce ambiguity, not just card count.

Choose a modest maintenance trigger, such as a weekly sample or a short cleanup after finishing a chapter. Look at repeated misses, conflicting prompts, and newly imported material first. Record the number of records removed separately from the number rewritten; they represent different kinds of work. After cleanup, test a few repaired cards with answers hidden and attempt an application question. A tidy file is helpful, but its purpose is better practice. If maintenance regularly consumes the whole study session, reduce the batch size and prioritize cards that are actively causing errors rather than polishing every record to the same level at once.

For a shared deck, nominate one person to reconcile edits before distributing a new version. Keep a short change log identifying corrected prompts and unresolved conflicts. Ask collaborators to report the card identifier alongside an issue. This avoids repeated discussions about similar-looking records and makes it easier to confirm that everyone has received the correction rather than an older export.

Copyable planning template

Topic: Clean and refactor study decks
Source and date:
What I need to recall or do:
First attempt without notes:
Error or uncertainty:
Correction and supporting source:
Next review or action:

Input example for Deck cleaner and prompt auditor:
ATP :: Adenosine triphosphate
ATP :: Adenosine triphosphate
ATP :: Energy transfer molecule

Frequently asked questions