Source documentation · record 01
Federal Register candidate-publication corpus.
The upstream inventory draws on the FederalRegister.gov public API and is retained as 26 bounded annual source captures covering 2000–2025.
- 788,470
- Federal Register publication records
- 2000–2025
- Complete publication years
- 26/26
- Annual candidate audits passed
- 144,849
- Records routed for full-text work Dedicated audit not started
- 43.6 GiB
- Current private working archive Measured 30 July 2026
- 594,727
- Local archive files Not for public distribution
Unit and sampling frame
One retained API record per publication instance.
Each year was reconstructed from retained FederalRegister.gov API pages and reconciled against the project’s candidate files. The audit checks publication identities, record-key uniqueness, dates, assigned years, provenance, and annual totals.
Because every retained record’s date matched its assigned year, the 26 annual sets cannot overlap. Independently summed annual counts equal 788,470 exactly.
Source documentation:Federal Register API documentation. FederalRegister.gov distinguishes its informational XML renditions from the official PDF edition on GovInfo; the project preserves that source distinction in its provenance.
Candidate-layer audit
Checks applied to every annual inventory.
- Twenty-six annual inventories reconstructed
- Record keys checked for blanks and duplicates
- Publication year and date reconciled
- Identity and provenance fields reconciled
- Annual counts independently summed
- Frozen source capture retained for replay
Figure 01 · Source corpus
Annual coverage of the candidate publication universe, 2000–2025
788,470publication records
Federal Register publication records retained before sanctions-related screening. Bar lengths share a 0–34,000 scale; exact annual counts are printed.
Table 01 · Exact annual values
Candidate publication records by year
Values are generated from the selected PASS ingestion audit at build time.
| Year | Records | Year | Records |
|---|---|---|---|
| 2000 | 33,325 | 2013 | 30,771 |
| 2001 | 32,003 | 2014 | 30,168 |
| 2002 | 33,094 | 2015 | 29,977 |
| 2003 | 32,805 | 2016 | 31,127 |
| 2004 | 32,377 | 2017 | 27,530 |
| 2005 | 32,233 | 2018 | 27,750 |
| 2006 | 31,505 | 2019 | 27,168 |
| 2007 | 30,818 | 2020 | 28,308 |
| 2008 | 31,817 | 2021 | 27,712 |
| 2009 | 30,663 | 2022 | 28,033 |
| 2010 | 32,471 | 2023 | 28,317 |
| 2011 | 33,103 | 2024 | 30,830 |
| 2012 | 30,875 | 2025 | 23,690 |
| Total | 788,470 | ||
Important limitation
Frozen-capture integrity is not live-source completeness.
This verifies internal completeness and consistency for the frozen local capture; it is not a current live-source completeness audit. A separate live-source audit would be needed to test whether the Federal Register API returns the same complete universe today.
Operational archive snapshot
43.61 GiB across 594,727 files.
This is a changing local working archive containing raw captures and intermediate files. The eventual public dataset will be a smaller, curated, versioned research package.
What comes next
From a broad publication pool to governed observations.
Compact Codex and Claude Code source packages are now retained in the project repository. They become reader-accessible when the repository is made public; no open-source grant exists yet.
- Route
High-recall metadata rules identify records for complete-text work.
- Review
Complete text is evaluated against the substantive research boundary.
- Verify
Accepted decisions and targets are tied to exact official evidence.
- Transform
Only governed, released observations enter the future index.