Algorithm References

Every built-in algorithm cites its source in its class docstring (the Reference: line for paper-based DP mechanisms; an explicit "non-private baseline" note for textbook estimators). Those docstrings are the single source of truth and are rendered, per task, on the API-reference pages:

Two kinds of entry appear in those docstrings:

Paper-based DP mechanisms cite the title + venue/arXiv/DOI of the work the implementation follows.
Standard non-private baselines (OLS, sample mean/median, plain gradient descent / Newton's method / ERM) are textbook estimators with no single canonical paper; they are labelled as such explicitly rather than given a stretched citation. They carry no privacy guarantees and exist as reference points for the DP mechanisms.

Citations are stated only where verified. One algorithm (TruncatedExponentialMechanismBGSUZ) is flagged below: its source could not be verified, so no formal citation is asserted.

Flagged: uncertain citation

`TruncatedExponentialMechanismBGSUZ`

The original source code attributes this scalar median mechanism only by the author-initial acronyms "BGSUZ" and "AJRV", with no paper title, year, or link. The mechanism itself is a Propose-Test-Release check on Tukey-depth volumes followed by a truncated exponential mechanism over the surviving candidate interval.

Best guess (not committed as a citation): the "BGSUZ" acronym plausibly refers to a Tukey-depth / PTR median-estimation line of work, and the construction is consistent with the propose-test-release framework of Dwork & Lei, "Differential Privacy and Robust Statistics", STOC 2009. However, no specific paper matching the "BGSUZ"/"AJRV" initials was verified against the implemented mechanism, so asserting any of these as the source would be a guess. A wrong citation is worse than none, so the class docstring states the source is unverified and this entry is left open. Resolving it is tracked as a loose-TODO ticket in the project coordination inbox.