Skip to content

Releases

v0.2.3

Choose a tag to compare

@dereckmezquita dereckmezquita released this 19 Sep 06:26
79bcb0a

The README now follows the same shape as every other package in the fleet.

This release does not change any code; it reshapes the README into the fleet's canonical structure so a reader used to one package's documentation already knows where to look in this one. Two sections that explained background concepts have been folded into the sections whose design decisions they motivate, and three sections that name the rendered documentation site, how to cite the package, and its licence have been added, since every package README carries them and this one previously did not.

  • Merged ## What this is into ## Technical overview: both described what the package is, so the second paragraph now continues directly under the technical summary.
  • Merged ## The revision trap (why vintages matter) into ## Design philosophy: it explains the problem the vintage-first design solves, so it now opens that section as motivating prose ahead of the existing bullet list.
  • Bolded the opening sentence of the plain-English lead paragraph, per the fleet's two-level documentation shape; no other wording changed.
  • Added ## Documentation (the pkgdown site and NEWS.md; the package ships no vignettes yet), ## Citation (built from DESCRIPTION), and ## Licence (MIT); none of these existed before.
  • Renamed ## The API key to ## Quick start, wording unchanged.

v0.2.2

Choose a tag to compare

@dereckmezquita dereckmezquita released this 19 Sep 02:28
c357cd6

Tidy the prose in the README and NEWS: drop the leftover "In plain terms" / "In plain English" labels now that the plain-first sentence stands on its own.

Earlier releases wrote the plain-English explanation as a labelled aside — "In plain terms: ..." in the README, "In plain English: ..." in three NEWS entries — a pattern the project has since moved away from in favour of just writing the plain sentence first, unlabelled, with the technical detail after it. This release strips those four labels without touching the sentences that followed them, and confirms (by re-grepping the whole package) that its prose already used British spelling throughout, so no spelling changes were needed.

  • README.Rmd: removed the leading **In plain terms:** label from the package summary; the sentence it introduced is unchanged. Re-rendered README.md via scripts/BUILD.sh readme.
  • NEWS.md: removed the leading In plain English: label from the 0.2.1, 0.2.0, and 0.1.0 entries; each explanatory paragraph is unchanged apart from the capital letter now needed to start the sentence on its own.
  • 4 labels removed, 0 spellings changed, across 2 files (README.Rmd, NEWS.md); README.md regenerated as a build artefact of the README.Rmd change; DESCRIPTION carries the version bump. Deliberately left alone: the "artifacts" wording in scripts/BUILD.sh and scripts/CLEANUP.sh (template-managed files, never edited inside a package), the # Determine badge color comment in the coverage workflow (it names the adjacent color variable and the shields.io field), and "labor force" inside a FRED series note in a test fixture (vendor text).

v0.2.1

Choose a tag to compare

@dereckmezquita dereckmezquita released this 18 Sep 19:27
56796fd

Fix the rendered README: a cross-reference to the promises package was showing up as literal escaped brackets instead of a link.

In plain English: the README described asynchronous calls using an R help-page cross-reference syntax that only resolves inside R's own help viewer. GitHub does not understand that syntax, so the rendered README on GitHub showed the literal text "[promise][promises::promise]" instead of a working link. This release replaces it with a plain markdown link, matching the fix already shipped in the hyperliquid and polymarket connectors.

  • README.Rmd: replaced the Rd-style [promise][promises::promise] cross-reference with a plain markdown link to https://rstudio.github.io/promises/, and re-rendered README.md via scripts/BUILD.sh readme.

v0.2.0

Choose a tag to compare

@dereckmezquita dereckmezquita released this 16 Jul 23:27
dfaa38b

In plain English: FRED reports when each data series was last updated as a timestamp string carrying its local timezone offset. The package used to parse that into a normalised UTC time — convenient, but the original string was unrecoverable, and the data-collection fleet archives sources byte-faithfully. This release keeps the parsed value and adds the raw string beside it.

Technical detail:

  • get_series_info()'s FredSeriesInfo gains last_updated_raw (character | NA) — the venue's verbatim offset-bearing string (e.g. "2026-07-03 07:48:03-05") alongside the parsed UTC last_updated. Additive; no existing column changes.
  • Typed via the @type shape and roxyassert contracts (regenerated; the doc-regen column-drop hazard checked); byte-verbatim round-trip tests on the fixtures.
  • Suite: FAIL 0 | WARN 0 | SKIP 7 (live-gated) | PASS 115; R CMD check at the pre-existing baseline.

v0.1.0

Choose a tag to compare

@dereckmezquita dereckmezquita released this 14 Jul 20:27
4d448c6

Initial release: the St. Louis Fed's FRED and ALFRED economic data in the fleet's connector idiom, built vintage-first.

In plain English: this package fetches official US economic statistics — GDP, unemployment, money supply, financial-conditions indices, the dollar, Treasury yields — through one typed, tested interface that works both synchronously and asynchronously. Unlike the general-purpose alternatives, it treats vintages (what each number was known to be at each point in the past) as a first-class surface, so a backtest can use the numbers a trader actually had then rather than today's revised history — the difference between an honest backtest and a look-ahead illusion.

  • FredSeries: the client, over the shared connectcore transport, with sync + async threaded from the constructor. Methods: get_series() (observations at any real-time window — the latest values are simply today's window, never a separate code path); get_series_vintages() (the headline — the ALFRED output_type=1 as-known-then long form); get_series_info() (metadata, including the revision-cadence notes); search_series(); get_releases(); and get_release_dates().
  • Vintage-first, point-in-time honest. get_series_vintages() returns one row per (observation date, real-time window), each carrying the value current during that window; a realtime_end of "9999-12-31" means still current. The whole-history matrix of a heavily-revised series exceeds FRED's 100,000-row response cap, so it is pulled in disjoint offset pages that stitch back to the un-paged whole with no seam clip, skip, or duplicate — the offset-pager logic proven live in the data-scraper's src/fred collector (NFCI's 906,899 rows reassemble byte-identical) is ported here into typed form, and its seam-stitching proof is a unit test.
  • Faithful typed shapes (FredObservations, FredVintages, FredSeriesInfo, FredReleases, FredReleaseDates) with every column documented as typed bullets — the observation value is numeric | NA because FRED serves . for a missing observation (which becomes NA, never 0); the observation date and real-time-window bounds are structural and strict.
  • Typed conditions from birth: fred_api_error layered in front of the connectcore transport chain (carrying status, FRED's own body error_code, a key-redacted url, and a body_snippet); fred_validation_error under the fred_error domain root. A test proves the API key is redacted from every stored URL.
  • Grounded against the live API's error envelope (a real non-2xx status plus a JSON {error_code, error_message} body — a missing/invalid key is a plain HTTP 400, no HTML redirect to disentangle) and fully-synthetic mock fixtures offline. Live tests (get_series, vintages, metadata, search, releases) gate on FRED_LIVE_TESTS and a working key; every FRED endpoint needs a key, so there is no keyless surface.