Build a citation you can explain

Opening problem

A memorandum cites “Box 4,” a video title without a timestamp, and a dataset's homepage. The reader cannot identify the collection, find the quoted sentence, or reproduce the reported count. Each item exists, but the citations identify containers rather than the material used. Unusual sources require more attention to identity and location, not less.

Learning objectives

Build archival citations from creator through repository; distinguish creation, publication, accession, and access dates; identify unpublished manuscripts and correspondence; cite audiovisual and social material with useful locators; and preserve versions of datasets, software, and generated records.

Think in layers of identity

For a conventional book, author, title, edition, and page often make the source readily retrievable. An archival item may require several additional layers: document, file, series, collection, box or folder, record group, and repository. A digital object may require a version, persistent identifier, file name, timestamp, or record key. Select the layers that identify the actual item without adding unrelated catalog information.

The National Archives' guidance for unpublished congressional records identifies the record, file unit, series, Congress, record group, and repository as relevant retrieval elements. Its general citation leaflet also warns against relying on record-group and box numbers alone. Those recommendations explain why a source's collection context can matter as much as the item's title. National Archives, Citing the Records of Congress; General Information Leaflet 17.

Keep the creator and the custodian separate. A private letter can be held in a public archive. A court file can include a party's memorandum, an expert report, and a court order. The repository's identity does not turn every item into the repository's own statement. Inspect the item and its descriptive record before assigning authorship.

Archival source cards

Transcribe the catalog's identifier exactly, including punctuation that distinguishes a series, box, folder, reel, or item. Record a stable catalog URL if one exists. If an image viewer exposes both a collection link and an item link, preserve the item link and the collection context. A viewer image number is useful only when the reader can identify the corresponding digital object.

Describe an untitled item clearly, using an appropriately bracketed descriptive title if the receiving style calls for one. Do not turn your description into an apparent historical title. Identify a recipient when it helps distinguish correspondence. If a document's author cannot be established, omit the author or describe the uncertainty; the collection donor is not automatically the document's creator.

Use the item's creation date when that is the relevant date. Accession describes entry into a collection; digitization describes creation of a digital representation; catalog revision describes the finding aid. These dates can all differ. Retain them in separate fields rather than selecting the most recent date as a default publication date.

For an undated item, use an undated notation appropriate to the receiving style. Use an approximate date only when a source supports the approximation. A phrase such as “circa 1910” should come from the item, catalog, or documented inference, not from how old the handwriting looks. Record the basis in the source card if it matters to the argument.

Annotated model: a versioned research publication

DataCite Metadata Working Group, DataCite Metadata Schema Documentation for the Publication and Citation of Research Data and Other Research Outputs, version 4.5 (DataCite e.V. 2024), https://doi.org/10.14454/g8e5-6293.

The official version page supplies this metadata and gives a separate identifier for the schema itself. The model demonstrates why related digital objects should not be treated as one undifferentiated website. It is a bibliographic model; a substantive claim about a schema property would also require inspection of the relevant documentation passage. DataCite's official version 4.5 page.

Unpublished manuscripts and working documents

Identify an unpublished work by its author, title or descriptive title, date or version, relevant page or section, and repository or other access information. Label its status accurately. “Unpublished manuscript,” “working paper,” “draft,” and “forthcoming article” describe different situations. Do not call a draft forthcoming unless publication status is actually established.

A paper can be publicly accessible yet unpublished in the sense relevant to its citation. Conversely, a work deposited in a repository may be a copy of a published article. Read the item and repository metadata. A download link is an access route, not a publication-status determination.

If a manuscript later appears in a book or journal, cite the version used. The edited version may alter words, pagination, or conclusions. When the difference matters, cite the historical manuscript and the later publication separately. A facsimile preserves a representation of the original; a transcription introduces an additional editorial layer that may contain errors or normalization choices.

Correspondence, interviews, and personal communications

For correspondence, identify author and recipient when known and relevant, document type, date, locator, and custody. An email thread can contain several authors and times. Select the particular message that supports the proposition. If quoting an attachment, identify the attachment rather than attributing its contents to the covering message without checking authorship.

An interview citation should identify who was interviewed, by whom when relevant, the date, form of record, and location or repository. A transcript page and an audio timestamp can point to different representations of the same conversation. If the quoted words were checked against the recording, record that fact; if only the transcript was inspected, do not imply an audio check.

Private or restricted communications present an access question that citation alone cannot solve. Use identifying detail appropriate to the document's lawful audience and applicable confidentiality requirements. If a source cannot be shared, describe the actual access limitation rather than suggesting a publicly retrievable item exists. Do not invent a repository or state that a document is on file with an author unless that custody is real.

Audio, video, and transcripts

Collect creator or speaker, title, publisher or platform, event or publication date, version, stable URL, and timestamp. Identify whether the citation concerns the event as a whole, a particular statement, or a visual segment. Use a timestamp range for a passage spanning several seconds or minutes, and state hours where omitting them could confuse a long recording.

An edited clip and a full recording are different versions. A statement can begin at 02:10 in the clip and 41:35 in the full event. Cite the version actually inspected and do not transplant timestamps between them. The date of the event and the date of upload may differ; preserve both when necessary to identify the material.

Captions and transcripts can be useful navigation aids, but compare a quotation with the actual source when the wording matters. Label a supplied transcript as such. If a speaker is uncertain or a word is inaudible, disclose that uncertainty rather than silently replacing it with a plausible guess. Accurate attribution includes knowing whether the words came from the speaker, an interviewer, or an on-screen caption.

Social posts, messages, and changing pages

For a public post, collect the account's display name, identifying handle, platform, date and time where supplied, identifying text or title, permanent item URL, and relevant preservation information. An account's display name can change; the stable identifier helps distinguish the particular item. A screenshot alone may omit the URL, date, surrounding thread, or indication that the post was edited.

Do not treat a repost as proof that the reposter authored the original statement. Identify quoted or embedded content and inspect the original if the proposition concerns its author. Similarly, a reply can refer to earlier context that changes its meaning. Save enough of the thread to understand the passage without republishing irrelevant personal information.

If a page changes, distinguish the live version from an archived capture. Record the capture date and the date of the underlying content where known. A later archive capture does not prove that the same text appeared on an earlier event date. The archive's timestamp establishes the capture it reports, subject to the service's limitations, not every fact about authorship or original publication.

Data, software, and generated records

A dataset citation should identify creator, title, version, repository or publisher, release date, and persistent identifier when supplied. A data-derived claim also needs enough information to identify the subset and method: table, variable, row range, query, filter, or calculation. A citation to a dataset homepage cannot explain which release or transformation produced the reported number.

For software, identify the program, credited creator or organization, version or release, and repository or publication identifier. If a result depends on source code at a particular revision, preserve the revision identifier in the research record and cite it when useful. A moving branch name does not identify a stable historical state.

A generated report may combine inputs, settings, code, and an execution date. Preserve those components when they affect reproducibility. Cite the underlying sources for external legal or factual claims and identify the generated report as the output examined. Its fluent presentation or automated origin does not establish that a cited authority exists or supports the reported conclusion.

Separate a platform's description of its method from your actual use of it. If the provider says its software supports a feature, that does not prove the feature was enabled in your run. Record the settings and output actually used. A release-specific source card lets an editor distinguish an observed result from a general product claim.

Worked editing session: rebuild the missing path

Fictional training example — not legal authority. The supplied archive record identifies a letter from Mara Vale to Omar Reed, April 3, 1932, page 2; Folder 8, Box 4, Harbor Port Committee Papers, Series B; Harbor Civic Archive, item HCA-17-44. Its catalog link is https://example.org/archive/HCA-17-44. The letter says the writer had not received a reply. No reply is included in the packet.

“Box 4 proves the committee rejected the request” fails twice. The citation omits the collection and item, and the proposition converts an absence of reply reported by one writer into a committee decision. A supported statement would say that Vale reported not receiving a reply as of April 3. Cite the letter and page 2, followed by enough collection and repository information to retrieve it.

A workable descriptive form is: Letter from Mara Vale to Omar Reed 2 (Apr. 3, 1932), Folder 8, Box 4, Harbor Port Committee Papers, Series B, Harbor Civic Archive, item HCA-17-44, https://example.org/archive/HCA-17-44. The precise punctuation may change under a receiving style. The identifying layers and limited attribution should survive.

If the catalog was revised in 2025, that does not make the letter a 2025 publication. If an OCR transcript says “received” where the image says “not received,” the quotation must follow the inspected original or accurately disclose the transcription being analyzed. The source's legibility and provenance matter before polishing the citation.

Transcription and documentary uncertainty

Distinguish a transcription error from a source error. A faded handwritten word may be uncertain; a printed date may be internally inconsistent; a catalog may disagree with the item. Record the conflict. Do not silently repair the source in a quotation and then cite it as though no editorial judgment occurred.

Use brackets or an explanatory note to disclose necessary editorial material under the receiving quotation convention. Mark illegible material rather than inventing missing words. An omission must not join fragments into a proposition the original did not express. Preserve an image or exact locator in the research file so the disputed passage can be reexamined.

For physical objects, identify the creator only when known and relevant, the object's title or description, date, collection, accession or catalog number, and custodian. A photograph of an object and the object itself are separate sources. If an argument rests on the photographed view rather than direct inspection, say so and cite the image's provenance.

Common mistakes and corrected approaches

Compact reference

Identify creator, item, date or version, locator, container, and custodian. Preserve stable identifiers. Distinguish originals, copies, transcripts, and generated outputs. Disclose material uncertainty and keep the proposition within what the inspected item establishes.

Reconstruct the reader’s retrieval path

The National Archives' still-picture guidance asks image researchers to retain record-group, series, box or folder where relevant, and item information. For catalog images, it specifically encourages providing the Local Identifier along with the National Archives Identifier when making inquiries. These identifiers serve related retrieval functions; a convenient web address should not cause either field to disappear from the source card. National Archives, Still Picture Reproductions, under “How to Cite Records when Inquiring About Specific Images,” https://www.archives.gov/research/order/still-pictures (page last reviewed Mar. 7, 2019).

The guidance is an inquiry and retrieval source. It does not verify the creator, date, subject, or rights of every photograph in the archive. That information must come from the particular item's record and, when needed, the item itself. A catalog's institutional authority should not be expanded into a claim that every caption or inferred identification has been independently confirmed.

Fictional training example — not legal authority. A training archive supplies a photograph labeled “Harbor hearing room,” item HB-P-4-17, catalog record 9001. The caption identifies a room and an approximate year, but names no person. A memorandum says that a named committee chair approved a proposal at the photographed meeting. The image's existence and item identifiers do not support that account. A supported description identifies what the inspected image and catalog actually show, while the proposed account of approval requires another source.

Fictional training example — not legal authority. The improved source card separates object (photograph HB-P-4-17), catalog (record 9001), description (supplied caption), and claim (room depicted). If a later archive migration changes the URL, those identifying fields can still help a researcher locate the same object. If the catalog corrects the approximate year, preserve which description your historical argument used and reassess sentences affected by the correction.

Test an unusual source by walking the reader's route in order: repository → collection or release → item → representation → relied-on part. A recording's timestamp belongs to a particular clip or full recording. A dataset's row selection belongs to a particular release. A transcription's page belongs to that transcription, even when the original has different foliation. The citation should preserve every step needed to avoid a wrong turn.

Then remove detail that serves no retrieval or analytical purpose. A catalog's administrative update date may belong in your notes without displacing the item's creation date in the citation. A collection donor's name may explain custody without establishing authorship. The goal is a short usable path, not the longest possible string of metadata.

Before closing the work, attempt that path from the citation alone. If it reaches only a collection homepage, add the item identifier. If it reaches the wrong release, add the version. If it reaches a recording but leaves the quotation unlocated, add the timestamp range. If access is restricted, state that limitation accurately. Each repair answers a concrete retrieval failure.

Takeaway

When familiar citation components are unavailable, preserve the path a reader needs to reach the exact item and understand its provenance. An unusual source deserves a precise locator and an accurately limited claim.

Chapter source note

The archival retrieval guidance draws on the National Archives' congressional-record citation instructions and General Information Leaflet 17, revised 2010, particularly printed pages 2–4. The real digital-publication model uses DataCite's official version 4.5 metadata. These sources were inspected October 8, 2026. The licensed Bluebook 22 Rule 23 consultation informs the distinction among archival creator, item, date, and custodian. Fictional packets supply their complete identifying information; no private interview, historical letter, or dataset result is invented as real evidence.

Image-retrieval source: National Archives, Still Picture Reproductions, inquiry-identifier guidance, inspected October 8, 2026; the page states a March 7, 2019 review date.