Skip to content

Local data available for fork generation

The local holdings can supply substantial parts of the fork without downloading the sources again. The selected FEC generations are already sealed and audited. Other holdings include source releases, corrected cohorts, older public tables and bulk originals that need different preparation before publication.

This inventory was measured on September 21, 2026. It did not upload data or dispatch generation workflows. The fork generation plan owns the complete producer/output sequence; finding a local file does not close that work.

Locations and verification

The main receipt root, abbreviated R below, is:

/Users/mikewolfd/Work/corpora/supply-2026-09-02/receipts

The inventory evidence directory contains the discovery paths, local-prepared-table-inventory.json (current Parquet footers and lengths), local-source-release-inventory.json (exact release members) and its compact local-source-release-summary.json. Searches covered Work (including corpora, checkouts and preserved worktrees), Documents, Downloads, Desktop, application caches, Codex/Claude working directories and /tmp. Dependency trees and unrelated application caches were excluded. The first broad search returned an error status without retained error details, so this is not an exhaustive disk census. The separate temporary-directory search completed without errors.

Verification here used live file presence, byte lengths, Parquet footers and existing audit receipts. Large existing corpora were not rehashed or fully reread. The smaller temporary legislative files preserved below received digest and ZIP integrity checks. Prior raw/output comparisons remain dated evidence, not newly repeated audits. Sizes are apparent bytes; GB and MB below use decimal units. Do not add all rows in these tables: originals, outputs, repairs and preserved copies can overlap.

First publication candidates

Holding under R Verified local scope Preparation
fec-generation-readiness-2026-09-21/generations/ Two sealed families, 1,171,789,372 bytes including metadata: 13,717,161 source records, 649 collections, 183,390 reported relationships and 26 broad source-catalog rows. Reuse the exact sealed artifacts through managed publication, then verify remote downloads and MCP reads. The catalog describes source families; it does not claim all their records were acquired.
fec-census-delivery-2026-09-12/consumer/fec_committees.parquet 27,311 rows; 1,038,794 bytes. All 16 column names/types match the current registered schema. Seal and admit the existing table. Preserve the explicit cycle=2024&cycle=2026 traversal; this is not a fresh unfiltered census.

The FEC generation pins are:

fec-observations
sha256:11bcb620974dff661fe22a76b9e10f0d416ef54c334ecd010dccd0a6ba1898f8

fec-source-catalog
sha256:0570574b9ed0aff29880269f8d1e44b9ea5243014aec6cf8da71b2d7e8093d35

fec-generation-readiness-2026-09-21/seal.json identifies the exact directories and confirms byte identity with the audited delivery. completion.json records the successful local publisher → CLI download → stdio MCP roundtrip. Both still record that external data publication had not happened.

Regulatory source releases available for rebuild

supply-2026-09-02/campaign/catalog-A-inputs.json references 671 retained releases. Their manifests reference 65,745 distinct blobs totaling 13,797,370,204 bytes, plus 20,855,694 bytes of release-local members. Every referenced blob and local member exists and matches its recorded byte length.

Source selection Releases Native records in publication receipts Distinct referenced blob bytes
Federal Register, releases/fr-full-1994-2026 1 1,007,156 4,988,949,808
Regulations.gov dockets 335 278,607 723,799,895
Regulations.gov documents 335 1,943,108 8,084,620,501

The bytes include originals, acquisition evidence and source-native companion representations. They are not exclusively raw downloads. Counts come from publishedRecordCount, not the manifest's larger count across companion representations. The selected index contains no duplicate release names or artifact pins. These are retained snapshot counts, not current source totals.

The old receipts report no failed records and passing release verification; 139 docket releases and 19 document releases are empty selections. Those old verdicts do not supersede subsequently discovered field-recovery defects or establish full agency absence. Rebuild with the corrected readers, retaining the newer observations in the chosen public parent where the source snapshots are older. Replacing the public parent wholesale with these older populations would lose coverage.

Prepared non-FEC data

Holding Local scope Use and remaining qualification
R/data-validation-sprint-2026-09-21/public/ 22 whole non-FEC public objects and one comments sample; 449,754,377 bytes. Includes 279,085 dockets, 2,001,222 documents, 112,861 comments-index rows, 885,266 SAM entities and 287,606 lobbying filings. Reusable retained parents. Read reconciled-assessments.json before selecting any table; these are not uniformly qualified final outputs. Preserve acquisition pins from public-acquisition.jsonl and public-documents-refresh.json.
R/remediation-sprint-2026-09-21/fr-identity/verified-replay/public-plus-native/federal_register.parquet 803,997 rows; 123,857,394 bytes; 23 columns. Corrected retained public parent with a bounded dated-identity repair. Wider historical source-field recovery remains.
R/remaining-gaps-wave1-2026-09-21/sr1/independent-second-review/native-bill-final/ Five corrected 118th-Congress HR/S tables: 129,718 rows; 4,658,963 Parquet bytes. Reuse in a complete bill-family build. The sealed cohort is named qualified-bill-status-118-hr-s; publishing that family unchanged would claim five table keys needed by bill-family. Reconcile all 18 outputs before production publication.
R/d1-measured-run-2026-09-19/output/ Measured members/terms, House votes, current committee rosters, laws, amendments, meetings, nominations, record issues, treaties and press releases. Current schema admission and complete-family sealing; preserve measured Congress/date limits. These are bounded runs, mostly for Congress 119.
R/rollups-0-24-0-adoption-2026-09-20/ Newer house communications: 4,969 rows. Four print-citation outputs: 41 / 29 / 12,700 / 49,935 rows. Prefer these later prepared outputs over the corresponding older measured copies; qualify the current schema and family members.
R/correction-lifecycle-2026-09-21/senate-final/run2/replay-output/senate_expenditures.parquet 3,272 rows; 234,172 bytes. Bounded retained corpus with one 357-row Part I correction. Other unavailable packages remain unrepaired.
spicy-regs/output/court-data-2026-08-22/court_opinion_clusters.parquet 10,070,727 rows; 3,939,704,769 bytes. Retained materialization; paired raw cluster-dump agreement is not qualified.
Same directory, court_opinion_bodies.parquet 250,000 rows; 1,735,931,994 bytes. Rebuild the old 19-column output: demonstrated native text variants were omitted. The corrected 25-column replay contains only 284 rows.
Same directory, docket_courts-2026-06-30.parquet 71,677,647 rows; 332,460,420 bytes. Auxiliary two-column join input, not an additional hosted rollup.
Same directory, court_cluster_scope.parquet 10,070,727 rows; 101,258,062 bytes. Five-column companion to the retained cluster population, not additional source coverage.
spicy-regs/output/bill_subjects.parquet 20,013 rows; 403,212 bytes, Congresses 116–119. Older enrichment input; 5,624 overlapping rows matched a later native audit. Reconcile the selected bill population and publication schema.

The five-table bill cohort's sealed pin is sha256:2bb61c8c8b4af24270a036d9d9e5a031b7803c56cb17bf2cbabfd6e0a51cbd73. Its review records 1,189,950 native-field comparisons and 2,206,611 stable-key cell comparisons. This supports the selected cohort, not broader historical coverage or the other thirteen family outputs.

Retained Congress originals

Two official 118th-Congress BILLSTATUS ZIPs remain under R/cbo-routes-2026-09-20/blobs/; the adjacent requests.jsonl records successful acquisition from the official URLs:

Scope Blob name XML members Compressed bytes
118 HR 8e7ca7dab50a7b9b977f021ec1b3231f8fedf82c33494553857b892fadfdba98 10,564 35,522,726
118 S 269261c0989db3ced789680ee2202747df9a7298f1ac8d2b074d3356b06e399c 5,649 14,410,894

The temporary-directory search found four additional ZIPs: /tmp/bs119s.zip and hr.zip, hres.zip, sres.zip under /tmp/claude-501/-Users-mikewolfd-Work-spicy-docs/8a5a1a5d-bd44-4a09-a525-c269c5837b3d/scratchpad/. They contain 18,366 XML members / 50,395,149 compressed bytes: 5,428 S, 10,503 HR, 1,566 HRES and 869 SRES members, all named for Congress 119. They do not cover the four other resolution types.

These four archives and the scratchpad's 40 bill XML bodies with 40 acquisition sidecars (17,686,151 bytes together) were copied to the inventory directory's retained-legislative-inputs/blobs/sha256/. Its manifest.json records original paths, preserved paths, byte lengths, digests and archive member scopes. Copy digests and all ZIP CRCs passed. Every body matched its acquisition sidecar's digest, size and HTTP 200 status. No source was downloaded for this preservation.

Archive integrity and member naming do not prove archive source authenticity or freshness; qualify their acquisition provenance before production use. The body sample does have retained acquisition evidence, but remains a 40-body selection. The measured D1 run's eight BILLSTATUS ZIPs were streamed rather than retained; its archive-listing table is metadata, not another set of source bytes.

Additional FEC originals and prepared sources

Holding under R Local size/scope Next use
fec-cycle-2024-2026-09-12/tables/ 15 prepared files; 8,915,803 bytes / 120,280 rows. Much is already represented in the sealed selection. Adopt only missing populations.
fec-cycle-2026-2026-09-12/tables/ 12 prepared files; 8,661,529 bytes / 101,432 rows. Masters, links, candidate/committee summaries and leadership remain available for builder adoption. Field audits are retained beside the tables.
data-validation-sprint-2026-09-21/public/fec_committees.parquet Older wider public table: 89,643 rows; 2,680,520 bytes; 16 columns. Retain for population reconciliation. The independently source-qualified 27,311-row census above has a narrower explicit selection; it does not replace or prove this wider population.
fec-bulk-expansion-2026-09-21/blobs/sha256/ Five complete financial ZIP originals; 2,477,585,475 bytes. indiv26.zip alone is 2,196,803,354 bytes and has 32,034,987 main-member rows still awaiting expansion. oth26, oppexp26, insert and delete streams already contribute selected rows to the sealed generation.
fec-senate-discovery-2026-09-21/ 598 live URL observations, 597 distinct stored originals / 122,507,816 bytes. Already adopted in the selected seed; one physical-line fallback remains explicit. The separate mirror is excluded.
fec-postgres-inventory-2026-09-21/ 14,190,177-byte committee-history dump, plus README; assessed 262,276 rows / 73 fields, cycles 1976–2022. Reader and table adoption still needed; current seed describes the file inventory.
fec-research-integration-2026-09-14/inaugural/ 51 originals; 211,723,750 unique stored bytes. Reuse qualified retained originals; oversized PDFs and missing raw URLs remain recorded gaps.
Sibling enforcement/ 23 acquired URLs; 21 distinct originals; 203,444,543 bytes. Identity conflicts and wider legal/body adoption remain.
Sibling agency/ 13 acquired URLs; 11 distinct originals; 9,455,299 bytes. Selected FOIA/Oversight qualification exists; agency/OIG collections have not entered the sealed FEC tables.

Also retain fec-source-expansion-2026-09-13/bulk/, fec-filings-2026-09-12/ and fec-next-collections-2026-09-14/historical/ for additional selected periods and filing originals.

The separate research repository at /Users/mikewolfd/Documents/Codex/fec-data-research-2026-09-11/ has 16,707 physical files and 1,535,832,195 apparent bytes excluding Git/bytecode/OS metadata. Its 47,915 bulk-object and 134,739 legal-object inventory entries are discovery records, not that many downloaded files. Some captures are prefixes or failed responses. Do not upload that directory wholesale.

Files that must not become full-corpus replacements

  • The retained public comments.parquet is a 212,733-row sample from a 23,889,661-row published object. It cannot supply the full comments mirror. No full local comments corpus was found in the searched locations. The large catalog/Iceberg import contains documents and dockets; its catalog policy has commentInput: null. Older 50,000-row comments experiments do not fill the gap.
  • The corrected regulatory cohort has 392 dockets, 547 documents and three comments. Its 10,969 passing raw/output comparisons prove the repair path, not a full base dataset.
  • The corrected court-body replay has 284 rows; the corrected report-section replay has 11 rows. Rebuild the wider selected populations with those fixes.
  • The older CFR table contains the demonstrated part-ancestry defect. Older report sections have the defective representation, and older bill-family outputs precede the cosponsor repair; their three model tables are empty.
  • Retained SAM, court dockets and USAspending tables lack paired source proof for the producing generation. Nonzero rows alone do not resolve that limitation.
  • Search indexes, DocSpec working stores, preserved worktrees and catalog exports often duplicate these sources. Their combined disk size is not new coverage.
  • ~/.claude/jobs/fcb997e0/tmp/ contains inventory JSONs pointing to receipt files, not additional Parquet data.

Execution tasks

The consolidated fork plan owns task status, dependencies, owners and completion checks. This dated document retains the measured inputs and qualification evidence. Use these task groups:

  • T03–T05: transfer selected inputs, publish the audited FEC seed, then expand separately from the remaining retained FEC populations.
  • T06–T07: repair the regulatory parents and resolve the missing full comments population without substituting samples.
  • T08–T14: admit, audit or rebuild the prepared legislative, regulatory, court and organization families using the source-specific limits above.
  • T15–T20: build dependent outputs, qualify workflows and consumer access, and reconcile completion across every intended output.

Uploading source releases for runner access and publishing queryable rollups are separate steps. Transfer selected manifests and their referenced bytes with provenance; do not turn an evidence-directory copy into an active dataset.