Skip to content

FEC candidate and agency local qualification — September 30, 2026

This is the earlier bounded selection. The subsequent retained-corpus qualification reconciles it with the prior observation family, adds native Word and archive readers, and records the user's PDF-processing deferral. The measurements below preserve this earlier run's exact scope.

The candidate-history rollup and retained agency-report receiver passed local source-to-table checks. These results establish the selected local generations; no remote publication, hosted MCP availability or deployment was performed.

The source wheel is spicy-docs==0.53.0+fec.3fc8388b368b, built from 3fc8388b368b204a57f6ea4029e26eadb132cfad. Its SHA-256 is 9b1b944f1dce879310d1bf37cbb4c9ae560dceb70d7eb9816d0862eb11b700b8. Independent archive builds produced identical bytes. The local version identifies this candidate separately from released 0.53.0; see vendor/README.md in the repository.

Candidate master

The exact input manifest covers every even-year bulk candidate master from 1980 through 2026. The header and 2024/2026 files reuse retained September 12 captures. The previously unretained 1980–2022 files were captured through free official requests on September 30. All original acquisition dates, requested and resolved URLs, byte sizes and digests remain in the manifest. The PostgreSQL dump's older cycles are outside this selection.

The standard rollup ran with HTTP responses supplied from those verified retained bytes. Its evidence artifact retains both the input manifest and response bodies; replay transport timestamps are explicitly identified as local replay times. The sealed generation is:

sha256:dc9634bae8e1e0d881e34487c4d7115915ba281972dfff98a41426b657ec9ec0

An independent Python CSV/ZIP reader compared every field by candidate ID and cycle: 130,562 rows and 2,088,992 cells, with no missing, extra or duplicate identities and no differing cells. Only empty nonidentity source fields map to NULL. Every generation member and source-evidence digest verified. The ordinary local MCP server described the expected schema and returned matching SQL row counts for every cycle.

This is complete coverage of the selected files, whose observation dates differ. It does not assert that the current live source bytes are unchanged. The scheduled producer will acquire every cycle afresh. A missing, empty, malformed or duplicate cycle, changed header, or redirect to another cycle stops the build before a new generation is published. Complete successful refreshes replace prior rows, including retiring rows that disappear from the selected files.

Agency reports

The installed source wheel processed the retained selection of 17 XML originals and 103 Oversight pages. The FEC 2009 XML-named original is Word Flat OPC and was an explicit native-parser refusal in this run. The later retained-corpus selection uses a separate native Word reader. The receiver admitted every successful source release: 119 collections and 9,191 records, covering 16 FOIA XML originals and 103 Oversight pages. No candidate or committee relationship was inferred. The local generation is:

sha256:2f2123013c3902c6b426064a3920beb712c86d9483e26821a50a96d483d8f356

Independent comparison matched every received record against the retained native report maps, including assets, bodies, fields and capture facts. A separate standard-library XML traversal also matched all 7,743 XML elements.

The receiver accepts profile: agency through the existing retained-input manifest, either as a directly selected original or a pinned source release. Source-release identity, original digest, acquisition date, ordinal, format, native metadata, bodies and declared assets survive in the observation tables. The html dependency extra is pinned for Oversight parsing. Missing or altered bytes, a wrong release pin and an unsupported representation refuse the whole selected output before installation. The later Word reader preserves package and text/control facts without claiming rendered or OCR output.

This agency-only local generation is a qualification selection. Before updating the existing public fec-observations family, prepare and qualify the intended complete manifest that preserves the existing selected collections alongside these additions. This local generation alone is not a replacement for that larger published family. PDF-only years, other report types, broader enumeration and current recommendation status remain open under FG11.

Retained evidence

Receipts live under fec-publish-completion-20260930/ in the retained corpora:

  • source/wheel.json and source/build_wheel.py: exact build identity and recipe.
  • consumer/candidate-inputs/manifest.json: original capture facts and provenance for the reused and newly acquired inputs.
  • consumer/candidate-replay.json and consumer/candidate-mcp.json: sealed generation, evidence pin and actual local MCP replies.
  • independent/candidate-output-comparison.json: independent whole-table comparison.
  • source/agency-release-manifest.json, source/agency-refusals.json and source/installed-replay/qualification.json: every agency outcome and source pin.
  • consumer/agency-replay.json: receiver selection, counts, member digests and verified local generation.
  • independent/agency-output-comparison.json: complete receiving-table comparison.
  • consumer/gate.json: repository checks and their logs. Initial failures and corrective checks remain retained alongside the final results.

Repository validation

The installed-wheel full test run completed with 3,759 passing tests, two skipped and 14 live integration tests deselected. Its sole failure found that the inherited candidate schedule shared 20:00 UTC with Unified Agenda. The candidate schedule now uses 20:10 UTC; the targeted hosted-workflow suite then passed all 93 checks. The full suite was not rerun after this schedule-only fix. Lint, type checking and dictionary checks passed on the final files. The original nonzero gate result remains in consumer/gate.json; the corrective results are recorded in consumer/final-verification.json.