What this record can—and cannot—tell us
Reproducible accounting, cautious identities, explicit uncertainty.
Source audit
The frozen snapshot contains 316,926 unique current transaction IDs from 1,888 filing committees. The raw acquisition contains 116 export workbooks. The audit checks their SHA-256 hashes, row counts, original 45-field schema, merged 46-field row widths, original-version IDs and complete nonoverlapping date partitions in each transaction family.
Matching the source’s reported count is one check, not proof that every required transaction was filed, every field is accurate or all recent activity is visible. Deleted records and superseded amendment versions are excluded.
| Dataset | Rows | SHA-256 |
|---|---|---|
| Contributions | 217,239 | af941a5a33884de19f47e56cfbaa04ad972062c67746da683f6679ed07a7bb53 |
| Other transactions | 99,687 | 129c673acff441295b47e6e99f52d6e1f9e350a57d8ab0137aca364869546790 |
Accounting rules
Amounts are parsed as exact integer cents. Cash contributions, cash payments, in-kind support, loans, loan payments, forgiveness, obligations, refunds, cancellations and adjustments have distinct definitions. The export’s Aggregate Amount is not added to transaction Amount.
An account payable and its cash settlement describe different stages of one purchase. We do not sum them as two purchases. Loan principal and interest are not silently separated when the source does not provide an allocation. Refunds remain explicit; without a verified association, they are not arbitrarily deducted from a particular donor.
Cash effects are not bank balances. Beginning cash is not zero by default. An official summary can refer to a calendar year and a later live state; it must be reconciled with the fixed snapshot before supporting a reserve or debt claim.
Official transaction filing manual · Campaign finance manual.
Identity and disclosure rules
ORESTAR committee IDs are authoritative identifiers. Other counterparties are grouped using strict matching rules by normalized name, reported type and address fields; this is a fingerprint, not independent identity verification. Original names, aliases and observed dates remain available. Similar names are review candidates and are not automatically merged.
Miscellaneous and anonymous categories are scoped to the filing committee. Blank counterparties receive transaction-specific unknown keys. Neither category contributes to unique-donor counts or bridges between campaigns. We publish geographic summaries, not residential-address maps.
The 69,613 groups in the earlier entity index are provisional and are not used as a verified population of people or organizations. Employer fields do not convert an employee’s personal contribution into company funding.
Transfers and network inference
A strict transfer requires reciprocal committee IDs, the same positive amount and the same transaction date, with exactly one candidate on either side within seven days. Wider or nonunique matches stay in a review table. Both source filings are retained. A committee outside the observed filer set is different from a counterparty with no reported committee ID.
Shared-donor tests use a selected 30-committee bipartite graph. Swaps preserve both sets of observed degrees. Two seeds each generate 199 draws after 10 edge-counts of successful burn-in swaps, with one edge-count of swaps between draws. The 435-pair hypothesis family receives Benjamini–Hochberg adjustment. Markov-chain mixing, entity resolution and unobserved donors remain limitations; p-values are exploratory.
Financial connectivity is not coordination. Commingled funds admit possible allocation paths, not unique donor-to-vendor tracing.
Uncertainty, comparisons and outcomes
Exact reported totals are a census of this snapshot, so we do not attach artificial sampling confidence intervals. Concentration estimates are conditional on entity rules and identified amounts. Disclosure sensitivity ranges are scenarios, not probabilistic confidence intervals.
Any election-result model requires certified completed outcomes, pre-election financial windows, complete race rosters and validation across races or time. Multi-seat ranked-choice races cannot be treated as ordinary single-winner contests. The November 2026 election is pending. This edition does not estimate causal effects of contributions on elections or government decisions.
Failure and gap register
Collection summary and unresolved coverage are summarized publicly; detailed operational logs remain in the private research archive. Existing raw downloads include 19 recovered request failures in the non-contribution collection. Enrichment failures and incomplete coverage remain separate from zeros.
| Gap | Why the answer is unavailable | Required evidence |
|---|---|---|
| G010 · Small donors and participation | Aggregate rows are not people. A total person count or retention rate covering those rows is not identifiable from this snapshot. | Itemized source identities and longitudinal resolution; do not impute people from transaction counts. |
| G012 · Businesses, labor, organizations and self-financing | A complete reviewed organizational-affiliation crosswalk has not been established. Similar names or employment fields are insufficient to assign industry ownership. | Official organizational records, effective dates and reviewed entity links. |
| G013 · Businesses, labor, organizations and self-financing | Employment fields alone do not establish direction, reimbursement or coordination; occupation/employer missingness is material. | Separate individual giving from organizational funds; source-based case review and alternative explanations. |
| G015 · Geography and representation | City strings do not identify neighborhoods or districts. No population-normalized neighborhood measure is published without an audited geographic crosswalk. | Official boundaries, Census denominators, validated geographic joins and missing-location sensitivity. |
| G022 · Committee transfers and intermediaries | Commingled funds do not identify dollar-level provenance. Financial paths can be possible without being uniquely attributable. | Association-ledger review and bounded allocation scenarios; never present scenarios as tracing proof. |
| G025 · Campaign services economy | Campaign reports show committee payments, not the recipient’s cost structure or onward transfers. | Invoices, association details, reimbursement review and recipient financial records. |
| G027 · Independent spending | This comparison requires complete target-to-race allocation and a reviewed candidate-committee crosswalk, not just expenditure names. | Separate candidate-controlled spending from independent support/oppose allocations. |
| G029 · Portland public financing | Program participation is selected, not randomly assigned; the current matched race sample does not support a causal effect estimate. | Comparable-campaign descriptive comparisons, selection analysis, validated local-donor definitions. |
| G032 · Timing and campaign strategy | A change point does not identify intention. Election alignment and sufficient comparable pre/post observations are needed before calling a change strategic. | Prespecified event windows, reporting-lag checks and campaign-source corroboration. |
| G035 · Elections and outcomes | The election has not happened at the September 27 cutoff. No outcome or causal analysis is possible. | No outcome model for pending elections. |
| G036 · Elections and outcomes | A sufficiently complete certified-result and race crosswalk with pre-election finance windows has not yet been validated for this edition. A multilevel model without that foundation would be misleading. | Certified outcomes; race-appropriate single-seat versus multi-seat measures; grouped/temporal holdouts, regularization and uncertainty. |
| G039 · Disclosure and visibility | The current-version snapshot omits superseded amendments and deletions. Current Filed Date alone cannot reconstruct a past public view. | Historical version retrieval with original disclosure and amendment timestamps. |
| G040 · Connections to governing | Official lobbying/consultant disclosure sources are located, but complete dated and reviewed entity joins have not been established. Name-only coincidences are not published as connections. | Verified public-record joins, effective dates and case-specific source review. |
| G041 · Connections to governing | The transaction ledger cannot establish motive or causation. Even verified overlap needs plausible alternative explanations and defensible comparisons. | Dated decision records, policy context and a credible identification strategy; otherwise descriptive case studies only. |
| G044 · Unusual patterns and counterexamples | No. Exploratory anomalies require peer comparisons, multiple-testing controls and original-source review; normal reporting conventions can generate them. | Robust within-peer methods plus investigation log and counterevidence; no automated allegation labels. |
Reproduction and edition boundary
Acquisition, normalization, enrichment, analysis and publication are separate stages. The project-local Python environment pins DuckDB, numerical and network libraries. TypeScript browser collectors retain raw evidence, checksums, immutable acquisition partitions and bounded retries. Local analysis is limited to two numerical threads and a bounded-memory DuckDB configuration.
The explorer opens the checksum-verified database read-only, validates every filter and returns snapshot and metric versions. CSV exports stream in bounded chunks. No database or residential-address graph is sent to the browser.
This edition starts in January 2025. Earlier history and unresolved research questions remain future work; they are not implied to be complete.