API reference

Every known release point, newest first.

titles_affected is what OLRC says the release point changed; ingested_titles is what this database actually holds for it. The two differ for as long as the backfill is incomplete, and conflating them would make an empty answer look like a missing law — which is why they are two parameters and not one.

What this deployment holds, and when it last asked OLRC what exists.

A versioned mirror of a living source has two ways of being wrong and only one of them is visible from the text: it can be missing a release point that was published last week, or it can have stopped asking altogether. The second is the dangerous one, because a corpus that has quietly stopped updating is indistinguishable from a corpus with nothing to update — every page still renders, every date still looks authoritative. So the answer here is two facts side by side: the newest release point loaded, and the timestamp of the last poll of uscode.house.gov.

Polled daily; stale goes true after a week without a successful check (storage.SOURCE_CHECK_STALE_AFTER).

Every ingested title, in the Code's own order — 1, 2, … 5, 5a, 6, … 11, 11a, 12, … (ADR-0025). The ordering is the Repository's contract, not this handler's, so the same list arrives sorted from any implementation.

11 usc 523(a)(1) → /us/usc/t11/s523/a/1, and whether it is there.

Parsing is citeparse.parse_citation, which is pure — it knows what string a citation names and nothing about what exists. Existence is answered here, with the batched labels() lookup already built for citation hover text, so this costs one query and needs no new Repository method (architecture rule 1).

A string that cannot be parsed is a 422: the request was malformed. A string that parses but names nothing loaded is a **200 with exists: false** — it is a well-formed question with the answer "not here", and telling the reader which part was wrong beats an error page.

Reading order at one release point, skipping nothing: repealed, omitted and transferred sections keep their place and arrive badged (gotcha 9).

What a list of citations say, in one request.

A section's text can carry forty cross references, and a reader that labels them — hover text, so a citation is legible without following it — has to ask once, not forty times.

The section's change timeline — one entry per distinct text, not one per release point, since most release points leave most sections untouched.

Diffs the two release points' verbatim XML (docs/adr/0016): a generic text diff, computed here because it needs no USLM vocabulary at all — presentation (wrapping ops in <ins>/<del>) is frontend/src/lib's job, same as the section renderer itself.

No release parameter, by design: a guid identifies (provision, release point) on its own, which makes it the stable citation for this exact text at this exact moment (ADR-0003).

Resolve a US Code identifier at a release point.

Longest-prefix match means /us/usc/t16/s45f/c/5 returns section 45f with paragraph (c)(5) extracted and anchored, rather than the paragraph alone — sub-section provisions are cut from the section at request time (ADR-0001), so the reader never loses context.

If the identifier names no section, it is tried as a structural node, and a table of contents comes back instead.

"My Provisions": the current user's first watchlist, auto-created.

Defaults for a user who has never saved a preference — get_settings does not write, so reading this costs no row (storage/accounts.py).