How to diagnose Anki duplicate imports without deleting useful work

Choose one repeated-looking prompt and inspect its source record. Determine whether the similar prompt comes from that same note or a separate note. Anki’s getting-started manual explains the note/card distinction. A reverse direction or another generated prompt is not necessarily an accidental extra record. Conversely, two separate notes can carry the same visible front while differing in source, explanation, or history. The diagnosis depends on the object, not the screenshot.

In the fictional vocabulary sheet, a learner sees a term twice during study. One appearance asks for meaning; the other asks for production. That may be deliberate. Elsewhere, two notes both ask the same direction because the source term was imported under different text. These are different maintenance tasks. Write a short record for each case: source note identified, generated direction identified, and reason it appears duplicated. This prevents a bulk deletion based on the wrong layer of the collection. Keep the explanation even if you later decide that a legitimate reverse direction is unnecessary.

The official text-import manual describes how matching uses the first field and the chosen match scope. Inspect the actual settings in your current import rather than assuming a filename or destination deck supplies a stable identity. Write what should count as the same source record. If your expectation depends on a value that changed between exports, the operation may not match the old note in the way you intended.

Use a tiny test worksheet with one unchanged record, one answer-only correction, and one intended new record. Record the expected outcome before importing. For the fictional course sheet, the learner expects the corrected definition to update an existing term and the new term to become a new note. If the test behaves differently, stop before the full file. Compare the relevant first-field values, note type, and match-scope choices with the manual. A visible term that looks identical to a person can still differ through punctuation or surrounding characters, so inspect the stored text carefully.

A human editor may treat a spelling correction or added qualifier as the same conceptual entry. A text-matching workflow may not share that interpretation. If the value used to identify a record changes, write down whether you intend a new question, an update to an existing question, or a replacement after source correction. These are editorial decisions that should precede the import. Do not ask the software to infer your intention from similarity alone.

In the fictional vocabulary sheet, the first version uses a broad English cue. The revised sheet adds a sentence to disambiguate the intended meaning. That is a good question-design edit, but it may also change the field used by the import to recognize the record. The learner tests the update on a copy and inspects the resulting source note rather than assuming the improved wording will automatically attach to the previous record. Preserve both versions while resolving the mapping. Deleting the older one immediately can erase the evidence needed to understand the discrepancy.

Create a side-by-side comparison containing the front, answer, source, explanatory fields, and relevant generated directions. Do not assume the newest visible wording is the most accurate. One record may have a corrected source reference while the other has a clearer answer. Compare both against the original material and decide what the intended surviving content should be. If native history is valuable, consult the documented operation before making an irreversible cleanup choice.

For the course-sheet example, the learner finds that one repeated record includes an exception omitted from the other. They verify whether that exception belongs to the assessed rule. If it does, they preserve it in the planned correction. If it is unrelated extra detail, they move it to an appropriate separate task rather than blindly concatenating both backs. Duplicate cleanup should not create an overloaded question. The goal is a defensible content record and a clear future import rule, so the next revision does not recreate the same confusion.

Use a content-lineage note for the selected survivor: source version, correction made, and reason the other record was considered redundant. If another import later reintroduces the old wording, that note explains the decision without relying on memory. In the fictional course sheet, the exception is retained because the current source requires it, while an obsolete gloss is retired. The lineage makes cleanup a verifiable editorial choice rather than a contest between whichever record appears first in Browse.

Make a recovery copy that matches the content and behavior you care about. Select a small group whose relationships you can explain and record the proposed action. Inspect the operation’s scope: deleting a note can affect generated prompts associated with it. A result list containing similar fronts does not establish that every selected record is redundant. Verify source meaning and intended directions before a batch action. Keep uncertain cases out of the first cleanup group.

After the small cleanup, inspect the intended survivor and its generated questions. Test the next revised import on the same recoverable sample. If duplicates recur, the identity assumption is still unresolved, even if the current collection looks tidy. Document the export procedure and relevant settings with a plain-language explanation. For a shared spreadsheet, tell collaborators which field values must remain stable and which content can be corrected through the tested process. A maintenance rule that depends on one person remembering an undocumented detail is fragile as soon as another editor joins.

Do not export an unexplained duplicate group as a migration shortcut. First identify the intended source records and explicit question directions. A receiving app cannot reliably decide which of two near-identical answers reflects your course source. Use the cleaned, verified sample to evaluate content preparation, and keep a log of any field mapping or exclusions. The transfer should make the selected task clearer rather than hide the unresolved identity issue in another interface.

Elly supports CSV, TSV, and TXT content workflows, not native preservation of Anki packages, templates, media, scheduling, or review history. Keep the original collection if those matter. A separate permitted source-based drafting task can help rewrite a confusing cue, but its output still needs source checking. Judge the test by whether the resulting question asks the intended thing and whether future updates remain explainable. A smaller verified sample is a better starting point than a large repeated set whose content lineage has already become unclear.