BibTeX Architecture: Decoupling Bibliographic Data from Presentation in LaTeX

Requirements: Stable References Across Changing Formats

Technical documents require citations that remain correct while the surrounding text evolves. The core requirement is separation: authors manage a single source of bibliographic truth, while publishers, journals, or thesis templates impose distinct formatting rules. BibTeX satisfies this by treating the bibliography as a data pipeline rather than inline text.

Smallest Suitable Design: Three Files, Two Engines

The minimal working set consists of:

• A .bib database containing structured entries (key-value pairs)
• A .bst style file implementing formatting logic
• A .tex source referencing citations by key

Two engines execute the pipeline: latex (or pdflatex, xelatex, lualatex) and bibtex. The .bib file never contains presentation markup—no italics, no punctuation, no author-order conventions. Those decisions live entirely in the .bst file.

Data Boundaries: Where Structure Ends and Policy Begins

A .bib entry declares typed fields:

@article{knuth1984texbook,
  author  = {Donald E. Knuth},
  title   = {The TeXbook},
  journal = {Computers & Typesetting},
  year    = {1984},
  volume  = {A},
  publisher = {Addison-Wesley}
}

The .bst file decides whether author appears as "Knuth, D. E." or "Donald E. Knuth", whether the title is italicized or quoted, and how multiple authors are joined. This boundary is strict: the .bib file is portable across IEEE, APA, ACM, and custom house styles without modification.

Trust flows one way: the .tex file trusts the .bbl output generated by bibtex. The .bbl file is a generated artifact—essentially a thebibliography environment with formatted entries. It should not be edited by hand; changes are lost on the next bibtex run.

Operational Checks: The Four-Pass Verification

Correct compilation requires a specific sequence. Run each command from the project root directory where main.tex resides. No elevated permissions are needed.

1. pdflatex main.tex — generates main.aux with citation keys
2. bibtex main — reads main.aux, resolves keys against .bib files, writes main.bbl
3. pdflatex main.tex — incorporates main.bbl, updates cross-references
4. pdflatex main.tex — stabilizes page numbers and reference links

Verification: open the resulting PDF and confirm no [?] markers appear where citations should be. Check the .bbl file directly—it should contain formatted \bibitem entries matching the chosen style.

To test style independence, change only the \bibliographystyle{plain} line to \bibliographystyle{alpha} or \bibliographystyle{ieeetr}, then re-run the four-pass sequence. The bibliography reformats automatically; the .bib file is untouched.

Failure Modes: The Citation Loop

Most failures manifest as unresolved citations ([?]) or duplicate label warnings. Common causes:

• Missing .aux file — first pdflatex pass skipped or failed
• Key mismatch — \cite{knuth1984} in .tex but @article{knuth1984texbook} in .bib
• Malformed .bib entry — missing comma, unbalanced braces, or undefined entry type
• Style file not found — \bibliographystyle{custom} without custom.bst in search path

Diagnostic: examine main.blg (BibTeX log) for parse errors. The log reports line numbers in the .bib file and undefined citation keys.

Non-ASCII characters (accents, non-Latin scripts) cause silent corruption or errors in traditional BibTeX. The workaround is switching engines: use xelatex or lualatex with bibtex, or migrate to biber + biblatex which natively supports UTF-8.

Conditions That Would Change the Design

BibTeX's architecture shows its age in three scenarios:

1. Complex Sorting or Multi-Script Bibliographies

The .bst language is a stack-based, postfix notation (similar to Forth) that few authors can modify. Implementing locale-aware sorting (e.g., Chinese stroke order, Arabic alphabetical) requires writing or finding a specialized .bst file. biblatex with biber handles this via Unicode Collation Algorithm configuration in the .tex preamble.

2. Rich Metadata Beyond Standard Fields

BibTeX's fixed field set (author, title, year, journal, etc.) cannot cleanly represent DOIs, ORCIDs, dataset identifiers, or custom taxonomy tags without abusing note or adding non-standard fields that styles ignore. biblatex defines extensible data models.

3. Single-Pass Build Requirements

Continuous integration pipelines that forbid multiple passes (or tools that only invoke latex once) cannot use BibTeX. biblatex with biber still requires multiple passes but integrates better with modern build tools like latexmk that automate the sequence.

Practical Verification Checklist

• Create test.bib with two entries of different types (@article, @book)
• Create test.tex with \cite for both, \bibliographystyle{plain}, \bibliography{test}
• Run the four-pass sequence
• Confirm PDF shows formatted references, no [?]
• Change plain to alpha, re-run passes 2–4
• Confirm citation labels change from numeric to alphabetic (e.g., [Knu84])

If all steps succeed, the decoupled architecture is functioning as designed. The .bib file remains the single source of truth; the style file remains the single point of presentation control.