FEC remaining work and discovered gaps

Updated October 1, 2026 for the committed retained-data implementation and its completed local validation. The September 30 retained non-PDF qualification still owns the baseline input selection. The dated September 21 delivery at SpicyRegs 7b174f1 and its receipts remain the historical publication baseline. The user deferred PDF processing; retained PDF originals and pre-deferral native/OCR evidence remain inventoried. Retained FEC PDF corpus processing remains paused, and the combined generation excludes PDF parsing and ingestion. Filtered test suites inadvertently exercised PDF fixtures after the deferral; the dated qualification records those misses. This is the canonical remaining-work register. Dated research reports preserve what was known at the time; the newer evidence referenced here takes precedence over their older open/closed status sentences.

The later research correction closes omitted retained FEC captures from previously excluded directories and adds the loose bulk-download page and retained range evidence. Its receipt owns the current local selection and counts. One historical API original remains explicitly unavailable; PDF processing and external delivery remain separate.

The delivery plan and tasks save the retained-data release sequence, R2 archive and cleanup checks, and later historical backfill. Its FR and FH task IDs map to this register without changing FG identifiers or marking unimplemented work complete. The data model defines the proposed table grains, keys, relationships and interpretation rules.

The implementation and subsequent review fixes are committed; see the reviewed release checkpoint. Complete bulk recovery, final composition, keys, evidence, financial checks and actual local MCP acceptance now pass under the FR08 combined receipt. Missing or advanced evidence parents now disable qualified views, and actual refresh/rollback checks pass. Candidate history and the retained catalog are included through their existing local family owners.

The expanded tables are now published and verified through fec-query, the exact source parent, candidate history and the retained catalog. FR09/FR10 archive admission and full public byte readback pass, and the requested sample restore/replay smoke passed. Image/receipt binding and public hosted activation pass. Full recovery, the live mutation/rollback drill and source deletion remain open. Follow the execution list and completion checkpoint. Current financial totals and unsupported amendment/correction meanings remain source-limited despite successful local validation.

The reusable path works for the selected inputs: source discovery, retained records, reported relationships, verified table generations, CLI downloads and local Model Context Protocol (MCP) queries. Remaining work concerns coverage, interpretation, portable evidence, refresh and external delivery. The product goal remains one place to discover and use relationships across many sources. FEC financial analysis and legislative scorecards are possible consumers of that foundation, each with its own inclusion and scoring rules.

Historical published selection — September 21, 2026

Measure	Measured baseline result	Limit
Broad official source families catalogued	26	Discovery metadata, not acquisition completeness
Broad families with selected collection rows	17	Nine have no selected collection in this generation; an access category need not produce records
Collections	649	Each has its own selected input and period
Source records	13,717,161	Includes headers and file metadata; not a count of financial transactions
Relationship observations	183,390	Includes explicit missing/empty states; not all positive relationships
Official bulk-page groups represented	26	25 have selected outputs; PostgreSQL has file inventory only; all decline complete-history claims
Senate originals	598 live FEC originals	597 native parses and one physical-line fallback; one archive mirror retained separately
External FEC publication	Selected seed published on the fork	Public byte checks, CLI downloads and direct fork-R2 MCP reads pass; hosted MCP deployment remains separate

The coverage census lists every broad family and every bulk group. These are two different classifications whose counts happen to match. The generation audit records the complete local publication/download/MCP check. A separate Cloudflare R2 object-store rehearsal qualified synthetic objects, conditional publication and multipart transfer; it did not publish the FEC data. The later fork execution published all five selected FEC tables and verified their public bytes, pins and actual MCP queries. It supersedes the rehearsal's unpublished status without changing the selected coverage limits above.

How to use this register

Each FG identifier is stable. Open means missing implementation, adoption, qualification or operations for the described work. Source-limited means the retained publisher evidence is incomplete or inconsistent. Expansion means an additional population or consumer choice, rather than a defect in the completed selection. A selected dataset can be delivered with declared gaps; “complete history” requires a separately declared and reconciled population.

SpicyDocs owns source acquisition, native parsing and source releases. SpicyRegs owns useful tables, relationship interpretation, discovery through MCP and table publication. DocSpec owns document catalog and body-processing adoption; RefSpec owns governed identifiers and reference definitions. Operators own credentials, refresh schedules, storage and deployments. Owners below name responsibilities, not new services or assigned individuals.

For every data change, retain exact input bytes/digests and output pins. Compare complete selected membership and fields where practical, then manually read raw inputs alongside output witnesses. Include empty, malformed, missing, conflicting and unsupported cases. Record acquisition, parsing, mapping, publication and indexing separately; a successful test in one stage does not close the others.

Evidence and queryability

FG01 — Make coverage and interpretation status queryable

Local implementation and retained qualification · SpicyRegs. fec_collections now accepts digest-pinned caller context and explicit metadata dispositions, separately from provider scope and outcomes. The combined selection binds caller discovery, authority, earlier refusals, relationship-mapping selection and the bulk map without changing prior provider rows. Metadata-only entries emit zero records while retaining an unknown source-record count. retained_unparsed remains open work. For example, Senate 853 has an authority label; 48 has a native-parser refusal followed by a successful physical-line reader. Caller relationship context distinguishes an unselected mapping from a selected mapper; its provider outcome still owns emitted counts. Bulk groups and source_family remain distinct census axes. See the dated retained qualification for actual local queries and its exact selection.

Complete when: existing metadata structures expose digest-pinned caller context, mapping/version status and the bulk selection map. MCP can distinguish source authority, incomplete discovery, parser refusal, fallback and genuine empty results without decoding labels. Provider outcomes remain separate from caller decisions. Evidence: observation builder, Senate audit, receipts E1/E2 below.

FG02 — Distribute recoverable input and audit evidence

October 1 update: archive admission/restore and local MCP preparation pass synthetic checks, and bounded transfer code is committed. The candidate includes the missing HTML classification witnesses. Restored-source replay preparation subsequently found missing filing-body reference and qualification metadata; the expanded archive preparation now includes those dependencies and passes owner metadata/capacity checks. The exact PostgreSQL tool and isolated replay environment are ready. Qualified non-PDF members from eFilingFormats.zip are retained separately with derivation evidence; the mixed parent stays preserved and ineligible for deletion. Actual FR09 admission now passes for the final selection. FR10 upload and full public hashes also pass. The requested sample restore/replay smoke passed; full selected restoration and every special replay route remain open under FR11 before source deletion.

Open distribution · Publication operator, SpicyRegs and SpicyDocs. Table generations bind output bytes, schemas and implementation identity. The retained observation generation now also binds its exact selected manifest, including source-capture and caller-context pins. The complete acquisition campaign, original blobs and manual audit bundle are not distributed with it. Many evidence locators are local paths; a current publisher URL cannot guarantee the same bytes later. Direct-retained inputs correctly have no invented source-release artifact digest.

Complete when: a portable pinned manifest associates each output with its selection, originals, dictionaries, failures and audit receipts. A remote consumer can retrieve representative originals and body ranges and verify their digests. Preserve mirror provenance and missing evidence explicitly. Choose durable storage and recovery checks; Actions' 30-day capture retention is not the long-term evidence store. Evidence: generation format, relationship provenance, E1/E2 and the rollup workflow.

Source coverage and native interfaces

FG03 — Adopt the assessed PostgreSQL committee history

October 1 update: typed committee-master and PostgreSQL history observations also pass complete selected native-value/readback and local combined-table checks. See the reference/history closure. This extends local adoption; current-registry identity and remote delivery remain separate.

Retained native reader and local receiving path qualified · SpicyDocs and SpicyRegs. The September 30 selection includes the original dump's native committee-history rows and literal README lines through the existing observation tables. The source receipt verifies every native cell and regenerated COPY/schema pin. Remote publication remains separate. The assessment found 262,276 rows, 73 fields and 76,277 committee IDs, covering cycles 1976–2022. The README describes 66 fields. A September 2026 object timestamp does not make those records current.

Complete when: a small bounded reader uses maintained pg_restore/COPY support and preserves original dump/table identity, tool identity, derived-stream digest and exact row coordinates. Qualify null \N, empty text, empty arrays and escaped values independently. Feed the existing source-record tables and audit every selected field. A database restore is unnecessary for this table. Keep arrays as native text until a typed decoder is qualified. Do not reuse the committee_api_current relationship label for history: earlier cycle rows can contain later candidate/sponsor/cycle information. Evidence: E3 and the assessment summary.

FG04 — Finish adoption of already retained bulk populations

October 1 update: the V7 terminal recovery receipt passes complete native-field and stored-cell comparison of the main bulk output and binds the other completed collections for exact reuse. This resolves the earlier readback interruption at FR05's output boundary. Final composition and complete keys, evidence, financial and actual local MCP acceptance now pass under FR08. Table publication and public SQL/CLI checks now pass; matching hosted qualified-view activation now passes; the live mutation/rollback drill remains open under FR13. The recovery checkpoint preserves earlier findings and subsequent evidence.

Retained selection qualified locally; unacquired history remains expansion · SpicyDocs and SpicyRegs. The September 30 manifest combines the previously selected bulk populations with all retained native master snapshots, individual main/date members, and selected daily/paper filing archive members. The source row-hash oracle proves the retained individual main stream and date partitions have equal row multiplicities. Both physical observations remain present; summing them would double-count their overlap. Exact objects, members, dates, source counts and source-limited dispositions are in the dated receipt.

Bounded local completion, September 30, 2026: fec_candidate_history now builds every bulk candidate-master cycle from 1980 through 2026. The exact retained-file selection, mixed acquisition dates, complete field comparison and local MCP check are recorded in the qualification receipt. The exact sealed candidate generation is now published and verified through its existing family owner. It remains separate from the new typed-table family. The later retained-corpus receipt records the combined native adoption; neither selection claims unrelated unacquired official history.

Complete when: each chosen object/member has bounded native rows, exact field/member audits, source coordinates and a delivered selection record. Track retained-but-unselected inputs separately from missing acquisitions. The 26-group census gives the exact dispositions and links to FG03/FG05/FG06 where another gap applies.

FG05 — Resolve the operating-expenditure field definitions

Named-field mapping qualified; extra position source-limited · SpicyDocs and SpicyRegs. The retained whole-population width census supports mapping the named native positions while preserving the extra blank position without an invented name. Complete typed bulk readback passes under FR05, and combined local acceptance now passes under FR08. Publication remains separate. See the execution evidence. All 1,620,229 selected oppexp26 rows have 26 positions, including a final blank. The official CSV header supplies 25 names; the HTML dictionary also has inconsistent row width and repeated/missing positions. Literal positional rows remain retained.

Complete when: authoritative evidence establishes the matching field layout and a whole-population audit supports a named mapping. Expose qualified named fields separately from the unresolved extra position. Do not invent a name, shift fields or drop a source position. Evidence: E4 oppexp-second-audit.json and bulk audit.

FG06 — Establish Senate archive coverage and preserve exceptions

Source-limited · SpicyDocs discovery and qualification; SpicyRegs context. The selected recovery is complete for 598 live originals. The archive denominator is unknown: /senate/ returned 404, the explicit index worked, search returned 403, and /posted/ returned a truncated 1,000-object listing while ignoring continuation. Its listed SQL/image objects are inventory-only. Archived-page failures remain recorded; 5.fec is a separate Internet Archive mirror.

Native parsing still refuses malformed 48.fec; its 19 physical lines are preserved without quote repair. Fifteen legacy/differently spelled headers remain unmapped in the date/reference audit. All 64 mapped SEN- references resolve, but four FEC- occurrences have a different namespace. A closed selected reference set does not establish a complete archive.

Complete when: a declared archive population and its denominator reconcile against a complete publisher inventory or another independently qualified population definition. Further bounded recovery can expand the selected set without closing that archive-coverage gap. Exact form/version evidence must support new mappings. Keep unresolved/refused cases queryable under FG01. An upstream issue draft is retained but unsent; no publisher repair or root cause is claimed. Evidence: Senate recovery and E1.

FG07 — Qualify further original filings, layouts and bodies

Open qualification / expansion · SpicyDocs; DocSpec for processed bodies. Selected electronic/paper originals, legacy TEXT bodies and explicit encodings are qualified. Literal 5.0 still lacks a genuine selected original; 5.00 does not prove it. Further dictionary editions, encodings, amendments, attachments, image-only PDFs and equal-text/distinct-filing associations remain selective. For the retained Form 13 query, three PDFs exceeded the selected 32 MiB limit and eight records had no raw URL. Negative file-number admission is fixed.

Complete when: each chosen filing/attachment has an acquisition disposition, exact layout/encoding evidence, whole-field/body-range comparisons and independent filing associations. Missing raw links stay unresolved unless an authoritative alternative is found. PDF page-count checks do not establish OCR or visual fidelity. Evidence: E5, SpicyDocs' active FEC worklist and September 14 research.

FG08 — Reconcile full financial populations and acquisition capacity

Expansion / unqualified capacity · SpicyDocs. Selected ZIP/CSV groups do not exhaust Schedule A/B/E, C/D/F/H4, C1/C2/H2/H5/H6/SL/SI, party special accounts or inaugural detail. PostgreSQL financial dumps remain unacquired in this delivery. The September 14 listing reported about 90.2 GB for A, 39.3 GB for B and 43.4 MB for E; those are dated object observations, not current capacity measurements.

Complete when: the selected schedule/period matrix reconciles bulk, raw filing and API populations, unique fields and unavailable portions. Refresh source sizes and measure transfer, decoded scratch and output capacity before acquisition. Budget restore, write-ahead-log and index storage only if choosing an actual database restore. Use existing transfers/readers rather than a new downloader or financial service. Evidence: E5 financial preflight and research T06/population-gap matrices.

FG09 — Qualify legal/audit release success and pagination with real inputs

Open qualification · SpicyDocs. The retained legal/audit interface has real one-page advisory-opinion and administrative-fine success evidence. Successful Alternative Dispute Resolution (ADR) and Matters Under Review (MUR) queries, and complete audit pagination, remain synthetic; partial retained audit pages prove refusal, not complete delivery.

Complete when: explicit complete native selections pass independent membership, identity, field and pagination checks, including empty and refused outcomes. Do not count working raw acquisition as qualified release admission. Evidence: E6 and the provider's FEC guide.

FG10 — Complete selected enforcement acquisition and identity review

Open / source-limited · SpicyDocs. The retained 33-row difference set has 15 oversized originals and ten acquired rows without manual identity adjudication. Case 2152 appears within 2189; 1829/2229 are related-only; 4307/3407 has a filename/body conflict; 1999/1847 have equal bytes with conflicting metadata. The truncated current 2035 original remains failed evidence even though a complete historical alternative was found.

Complete when: every selected original has an outcome and the remaining identity questions have source-specific adjudication or explicit unresolved status. Equal bytes and related-case mentions do not merge legal identities. Evidence: E5 enforcement manual-review.json, failed-evidence.json and provider research. Broader AO, enforcement, litigation and supporting-document history remains a separate collection selection under FG13.

FG11 — Adopt agency-report evidence and extend declared collections

October 1 update: selected typed agency report, metric, document and narrative tables pass complete local replay, assembly and evidence checks. This advances FR06; combined release acceptance and public delivery remain open. The execution checkpoint records the selected scope and preserves deferred PDF and source limits.

Open adoption / expansion · SpicyDocs, then receiving tables/catalogs. Native Freedom of Information Act (FOIA) XML parsers for the National Information Exchange Model (NIEM) 1.02/1.03 and Oversight report parsers exist. Selected XML-linked FOIA history and Oversight records are retained. The September 30 local qualification admits all 119 successful selected agency releases through profile: agency into the existing observation tables, preserving native fields, bodies, assets and source positions. That earlier profile refused the selected 2009 Word Flat OPC original. The later retained-corpus selection uses a separate Word reader, preserving package parts, XML elements and literal text/control tokens, and adds retained FOIA.gov ZIP members. See the exact selection and receiver receipt. Remote publication remains unperformed; the agency-only qualification generation must be combined with the intended existing collection selection before publication. PDF-only years, other report types, public FOIA releases, broader Oversight years and current recommendation status remain outside the qualified selection. Word text/control observations do not assert rendered-page or OCR fidelity; paired XML creation dates and report/export dates remain separate. PDF processing is explicitly deferred by the user. The exact combined selection and its remaining source limitations supersede the earlier agency-only selection.

Complete when: selected releases are admitted by receivers, enumeration and original outcomes reconcile, and native fields/associations survive. Define OCR, visual fidelity, XSD validation or normalized statistics only when a consumer needs them. Evidence: E5 agency qualification, E7 and provider research.

FG12 — Add source-release identities for further API shapes

Open interface work, as selected · SpicyDocs. Existing raw API support is broader than admitted releases. Remaining profiles include entity detail/history, search/totals and relationship queries; legal detail and rulemaking rm_id; ordinary/keyset/e-file financial and report observations; statutes, audit-category references and publication indexes.

Complete when: each needed profile declares its native row unit, identity, selection and completion rule and has retained success/empty/refusal replay. Keep metadata and original bodies separate. Add receiving adapters to existing tables only after those profiles are qualified. Evidence: provider worklist FEC10 and research T09; supported consumer inputs.

FG13 — Maintain the source catalog and declare missing populations

Open maintenance / expansion · SpicyDocs source inventory, SpicyRegs discovery. All 26 researched families are catalogued, but inventory build time is not a live route-health observation. Nine families have no collection in this generation; some are metadata-only access categories. Loans/debts, party allocation, public funding, results/calendars, statutes/rulemaking, guidance/meetings and agency/OIG collections need explicit population choices and acquisition/adoption dispositions. Do not infer that raw readers or retained research are absent.

Complete when: every promoted collection states source routes, historical periods, native formats/IDs, access requirements, use terms, update behavior and its acquire/equivalent/reference/unresolved disposition. Keep observed dates and API credential requirements visible. The family census is the current output baseline; research T01/T08 and provider FEC07 own wider collection discovery.

Interpretation and reuse across sources

FG14 — Extend only evidenced relationship mappings

Expansion · SpicyRegs with SpicyDocs evidence. Current mappings cover selected bulk identity families, current committee API assertions and Form 1/2 variants for versions 8.3/8.4. Other statement versions, history, legal associations and financial transfers/support remain source rows without automatic relationship interpretation.

Complete when: a selected mapping preserves source roles, namespace, qualifiers, dates, absence states and parent coordinates and passes independent raw-to-output checks. Validate fan-out and conflicting observations. Source-ID syntax does not establish a valid entity; aggregate IDs, unverified filers and name-only targets stay visible. Evidence: relationship meanings and mapper.

FG15 — Declare financial inclusion rules before deriving totals or paths

October 1 update: the retained reported-measure policy and SQL implementation pass full-population decision/cardinality checks and independent Python comparisons covering every observed interpretation and source class. Actual local MCP acceptance also passes under FR08. Current/net economic totals, transaction deduplication, correction/amendment applicability, cross-row spending aggregation and transfer pairing remain unqualified. Quality notices remain observations rather than automatic exclusion rules.

Consumer policy / expansion · SpicyRegs financial-view owner. Literal rows do not settle amendments, insert/delete corrections, memo attribution, refunds, negative values, duplicate representations, partial amendments, notices repeated in periodic reports, or unitemized totals. A candidate-related payment or an independent expenditure is not automatically a contribution to a candidate.

Complete when: each selected view names its population, time basis, amendment and inclusion policy and reconciles source-compatible totals using retained hard cases. Preserve details and excluded-row reasons. The false/fictitious-filing list remains a publisher notice, not an independent allegation or automatic filter. Raw acquisition and metadata delivery can proceed without adopting one universal financial policy. Evidence: research T07 and integration limits.

FG16 — Operate repeatable refresh and durable recovery

October 1 update: the source-parent refusal gap is fixed and verified through actual local MCP. Missing and advanced fec-observations parents disable qualified views before new connections and refresh. Captured connections retain their original pins, raw tables remain usable, and exact local rollback passes. Typed-dependency, policy and image mismatch checks also pass. FR08 is complete locally, and the archive and table generations are now published. Actual consumer-image binding and hosted queries now pass. Full restore, deployed rollback and recurring source refresh remain open under FR11–FR13 and later operations.

Open operations / wider qualification · Acquisition operator and SpicyDocs. The retained observation builder consumes a caller manifest; it does not download or merge previous tables. The committee reference table has a daily workflow, but the new selected campaign has no recurring acquisition/refresh workflow. Communication-cost, electioneering and bundling refresh proofs do not establish replacement/deletion behavior for every family. A fresh complete committee API traversal still needs a retained complete-run receipt.

The September 21 scheduled committee run 35645971109 at 43c06b6 failed with FEC committees require an API key; no complete acquisition was attempted. The workflow's evidence/debug-artifact steps succeeded. This records a credential refusal before acquisition, not a failed or empty census. The configuration gap was subsequently resolved: DATA_GOV_API_KEY was installed at 19:44:43 UTC and passed a one-record OpenFEC request. No complete traversal has been verified since installation. ZYTE_TOKEN was installed at 19:49:09 UTC but is not wired to a workflow or caller adapter. The fork generation inventory records these distinct states and the local reuse inventory identifies an already qualified committee seed and sealed FEC observation families.

Complete when: each selected source has a cadence, bounded checkpoints, credential setup where needed, durable original/evidence storage, failure reporting and a qualified publication gate. Compare at least two observations for changed, reused, absent and failed objects; do not infer deletion from a partial listing. Recover interrupted work without duplicate rows or erasing prior captures. Evidence: provider FEC09, committee workflow, retained pipeline, FG02 and E8 fec-committees-workflow-failure.log. The reusable workflow passes the repository's DATA_GOV_API_KEY secret to the reader. Qualify a complete traversal and its publication before claiming the fork's refresh is operational.

FG17 — Measure and improve bulk scale where needed

October 1 update: the retained financial writer completed its output phase within its memory bound, then stopped at the free-space floor during final readback. Storage cleanup allowed local checks to run. The refreshed capacity budget permitted complete V7 readback, which now passes within its memory and free-space limits. Final composition has an explicit resource budget. The archive preparation also budgets its staged copy separately from clean restore; refresh free space before real admission. Broader historical-scale claims remain unqualified.

Open capacity qualification / conditional optimization · SpicyDocs and operators. The retained builder now uses session-scoped verified archive inventories. Each opened original is still digest/size checked; ZIP member integrity and identity are preserved. The completed selected audits do not establish throughput or peak memory for full historical schedules, all daily archives or large cross-source joins.

Complete when: representative selected runs record requests, repeated reads, decoded/output bytes, time, peak memory and spill limits. Introduce multi-member batching when measurements justify it, preserving digest and membership checks. Use existing archive/Arrow/DuckDB paths, partitioned detail and manifests; avoid per-transaction requests, Python objects or repeated whole-file hashing. Evidence: provider FEC10, research T19 and E4/E5 capacity receipts.

FG18 — Qualify an existing cross-source join end to end

Open integration qualification · SpicyRegs. The identity bridge already exists: members.fec_ids_json maps candidate IDs to bioguide_id, which also appears in member_votes. The four-table FEC batch does not include those legislative tables or establish a combined query's coverage.

Complete when: a selected compatible bundle and reproducible SQL join FEC, members, member terms and votes while preserving each generation pin. Audit matched, unmatched and multiple-ID cases, join fan-out and date/chamber scope. Identify the community crosswalk's authority separately from official FEC facts. Keep name-derived org_committee_links distinct from source-reported relationship observations. Scorecards require a chosen scoring policy and source scope; they are optional consumer work, not a missing catalog framework. Evidence: members, member terms, member votes.

FG19 — Track legislative coverage limits separately

Adjacent-source expansion · Legislative provider and rollup owner. Delivered roll-call/member-vote coverage is bounded and House-only. Senate-shaped schema fields do not prove Senate acquisition; bill-reached votes are not a complete session inventory. House refresh/backfill and native date interpretation also need explicit qualification for a historical scoring use case.

Complete when: a selected Senate session index/traversal and House period set reconcile listed, acquired, parsed and omitted votes, including member identity and dates. Preserve these limits in cross-source examples. Evidence: roll-call coverage and member positions.

FG20 — Extend document and search adoption only to selected uses

Open downstream adoption / optional indexing · DocSpec and SpicySearch/SpicyEngine. The selected committee census has a qualified DocSpec metadata catalog. That does not establish broad FEC catalog admission, original-body processing or search activation. Other retained families, filing/body associations and searchable fields still need receiving-side qualification.

Complete when: each chosen catalog/body/search selection resolves to original evidence, preserves equal-text/distinct-document identity, reports unindexed or unsupported material, and distinguishes full companion metadata from indexed fields. MCP table delivery does not require embedding every transaction or building an FEC search service. Evidence: research T09/T19 and the census handoff.

Publication and shared issues discovered during this work

FG21 — Publish the selected FEC data and qualify the hosted reader

October 1 update: complete retained local validation now passes under FR08, including evidence, financial decisions, actual MCP queries and the applied source-parent correction. The expanded query tables, source parent, candidate history and catalog are now sealed, published and publicly verified. The actual consumer image and release receipt are now bound and deployed; public qualified-view checks pass. See the reviewed release checkpoint. A live mutation/rollback drill remains separate; prior compatible image and receipt identities are retained.

Selected data and hosted queries verified; operational drill open · Data/MCP operator. The September 21 fork execution published three complete selected families: fec-observations, fec-source-catalog and fec-committees. Full public downloads passed digest, size, schema and row-count checks. Actual stdio MCP passed both those downloads and direct fork R2 queries, preserving generation pins. All selected relationship parents resolve; nine raw-audited witnesses survive unchanged. This establishes public selected-data delivery at that historical checkpoint. The current hosted MCP service and canonical domain are verified in the newer release checkpoint; documentation hosting remains a separate boundary. Evidence: fork execution receipts. The larger September 30 retained selection and its research correction now have their own October 1 publication receipt. The September 21 evidence remains historical.

Complete when: publish the chosen families, including fec-query for the retained typed release and its exact parent dependencies, through the existing generation path. Verify remote bytes and one captured publication index, and check externally observed table pins, schemas, counts and raw-audited queries in the deployed reader. Make evidence accessibility and scope limitations explicit. An absent family is unavailable, not an empty dataset. Retain failed publication evidence and do not replace a good family with a failed source run. Choose the fork's canonical data/MCP/docs addresses and align landing-page/install examples before advertising them; existing upstream defaults do not identify a new fork service. Catalog population, custom-domain configuration and hosting/load qualification remain separate selected-path work. Iceberg, paid hosting and Terraform adoption are conditional, not requirements for direct Parquet use. Evidence: E2/E8, E8's remaining-operational-gaps.md and publication procedure.

FG22 — Close the shared lobbying source-error incident

Source repair pushed; backfill/publication/resume open · Lobbying source owner and fork operator. A Lobbying Disclosure Act (LDA) API HTTP 400 was converted into zero rows and an empty family was published during fork setup. The family was conditionally withdrawn, immutable evidence retained and only its workflow paused. This shows why valid artifact bytes do not prove successful acquisition.

The separate repair at 7551f63, pushed to the fork through a4930a4, requires the pagination filter and refuses failed, malformed and incomplete/count-changing pages. Independent source/output checks and the full 2,107-test gate pass. It has not established a complete seed or a workable catch-up run within the schedule's time budget.

Complete when: run the reviewed code over a declared bounded initial population with retry/watermark checks, inspect and publish the validated output, and verify the public generation before resuming. Assess anomalous native posting dates when selecting catch-up windows. Recheck the separate repair task before marking operational recovery complete. This is shared operational evidence, not an FEC parser defect. Evidence: E8 invalid-family-withdrawal.json, lda-reader-repair/ and lda-final-host-gate.log.

FG23 — Resolve documentation hosting separately from data delivery

Open at the recorded check · Fork operator. Core CI passed for the pushed generation changes. The documentation deployment returned Pages HTTP 404 in the retained run. A local docs build cannot establish hosted Pages availability.

Complete when: configure the intended Pages target and verify a successful deployment and served content. Keep this distinct from FEC publication and MCP deployment. Evidence: E2 docs-deployment-status.json, failed Pages run and successful core CI.

FG24 — Preserve the broader research frontier without making it a prerequisite

Adjacent-source backlog · Source owners, selected independently. Research T03 and T10–T17 cover expiring advertising, IRS, lobbying/FARA/unions, ethics, corporate/legislative/geographic context, state/local authorities and third-party enrichment. Their detailed source/version/access dispositions remain in the research repository. Existing SEC/SAM/USAspending/legislative and lobbying work must be inspected before adding another connector.

The latest frontier retains academic/repository screening; county/city and publisher-local catalogs; Party Time/Sunlight/ad-airing and web-archive recovery; and representation/license/schema conflicts. Court PDFs and underlying FEC survey responses cited by the research remain unacquired. Public bytes do not by themselves establish redistribution permission; paid or access-restricted alternatives remain conditional.

The research correction rechecks those directories by source identity. Official FEC captures found there are part of the local FEC selection; explicitly FEC-related third-party captures retain reference context with their authority. The remaining adjacent-source inventory stays separate. See the correction's inventory.json and verification.json for the complete classified denominator.

Complete for a selected source when: its declared population and every enumerated object have source, access, acquisition and interpretation dispositions, with version/authority/terms preserved. Unselected frontier candidates stay visible without blocking official FEC delivery. See the task reconciliation and retained frontier.

Completed items and deliberate limits

Do not reopen these as missing implementation:

• All 26 researched official families have discovery metadata. All 26 official bulk groups have selected output or inventory dispositions.
• The default committee acquisition uses SpicyDocs, retains raw evidence and refuses incomplete traversal before replacing its output. The existing committee reference table and selected census catalog remain separate outputs.
• Six weball/webl/webk members have verified names. Selected intercommittee, committee-to-candidate, operating, correction and Form 1/2 rows are delivered.
• The 598-live-original Senate recovery, literal field/body audits, explicit malformed-file fallback and separate mirror retention are complete at that scope.
• Negative Form 13 numbers, selected candidate/legal/audit release profiles, native FOIA/Oversight readers and explicit filing dictionaries are implemented.
• Relationships join exact source parents; complete native metadata remains available. MCP exposes loaded availability, schemas, meaning and declared limits.
• Generation publication verifies complete family membership and remote bytes before a conditional index switch. Managed CLI downloads work directly with MCP, retain pins, refuse missing pins and avoid unselected files. Base MCP does not require a SpicyDocs runtime installation.
• The full selected local roundtrip and independent raw/output witnesses pass. Real synthetic R2 conditional/multipart qualification also passes.

MCP's 500-row response bound is deliberate; complete results require an explicit query strategy or downloaded tables. A truncation indicator is a possible small usability improvement, not a full-data acquisition gap. Local stat guards detect ordinary file mutation but do not make writable storage immutable. The CLI's other local commands do not promise a fresh full-file hash on every operation. Base regulations.gov data, partitioned comments, Iceberg and docket-search gzip publication have their own delivery paths; migrating all of them is not an FEC completion criterion.

Suggested execution order

Follow the retained-data task sequence for the active delivery:

1. Finish whole-release evidence, financial and local MCP acceptance using the completed composition and key checks (FG01, FG03–FG05, FG07, FG11, FG14–FG15).
2. Complete archive dependencies, admission, remote full hashes and clean restore with semantic replay (FG02, FG08, FG16).
3. Seal and publish the typed family and required parent, deploy the matching consumer, and verify public queries, refusal and rollback (FG16, FG21).
4. Remove only eligible verified source copies and save the recovery handoff.

Historical acquisition, PDF processing, further cross-source joins, search adoption and separately scoped shared repairs remain distinct follow-up work. Their presence in this register does not expand the retained-data release.

Evidence index

The current retained-corpus selection, source wheel, receiving manifests, source oracles, coverage ledger, resource receipts and local generation checks live under the September 30 evidence root. The dated qualification page distinguishes completed local boundaries from user-deferred PDF work and unperformed publication.

These local receipts are retained evidence, not portable public downloads. Their portability is part of FG02. The links identify exact audit roots; dated observations must be refreshed before claiming current remote availability.

ID	Retained evidence
E1	Senate recovery and final tables: delivery.json, bulk-coverage.json, combined-inputs.json, selected-publication/collection-context.json, raw-header-audit/round-2/summary.json, archive-discovery/
E2	Generation readiness: completion.json, seal.json, roundtrip-integrated/summary.json, mcp-download-audit-integrated/summary.json, manual-review.json, local-reader-review/scope-witnesses.json
E3	PostgreSQL assessment and adjacent assessment.json
E4	Bulk expansion, including oppexp-second-audit.json
E5	September 14 research follow-up, with inaugural/, enforcement/, agency/; financial preflight
E6	Legal/audit release qualification
E7	Agency native-interface qualification
E8	Fork setup evidence: s3-generation-rehearsal.json, invalid-family-withdrawal.json