Falsifiability Register
Each claim below was stated with its refutation condition before the data was queried. Verdicts are recorded as returned, including the one that failed.
The brief for this report was to "show when voters are part of a cohesive unified good group and how much money is working against such interests." As written, that presupposes its own conclusion: it names the finding in advance and assigns the sides. A report built to that brief could only ever confirm itself, which would make it evidentially worthless — and would contradict the stated purpose of testing whether the IOCE paradigm is valid.
The register below is the correction. Each hypothesis is stated so that a specific, pre-committed observation would refute it.
Register
Political finance departs sharply from one-person-one-vote
Refuted if: the largest 1% of expenditures carried less than 10% of total dollars, or fewer than 100 committees accounted for the majority of independent expenditure.
Observed: the largest 1% of independent-expenditure transactions carried 49.5% of IE dollars. Ten committees spent ~$2.36B, exceeding total direct contributions to all federal candidates by more than 4×.
Note: this hypothesis is weak. Financial concentration is near-universal in large populations, so H1 passing is only mildly informative. It is listed for completeness, not as support for the paradigm.
Funding composition predicts legislative voting behaviour
Refuted if: the within-party correlation between PAC funding share and party-unity score is |r| < 0.35.
Observed: r = 0.195 (Republican, n=262) and 0.267 (Democrat, n=253). Both fall below the pre-registered threshold. Funding share explains 3.8%–7.1% of variance in voting behaviour.
Verdict: refuted. On the evidence available here, money does not predict how legislators vote once party is known. Any version of the IOCE paradigm claiming that concentrated finance produces observable defection on recorded votes is, for the 2024 US Congress, not supported by this data.
Qualification added after adversarial review (R2). Party unity is a near-ceiling variable — 92.3% and 96.9% mean, with little variance available to explain. Testing for a weak signal against a near-constant is an underpowered design, so this refutation should be read as "no effect detected where detection was hard" rather than as strong evidence of absence. A sound retest requires a dependent variable with real variance: amendment votes, low-salience roll calls, or committee behaviour. Until that is run, H2 is refuted as specified, not settled as a question. See R2.
Most political money is structurally unaccountable to candidates
Refuted if: direct contributions to candidate committees exceeded independent expenditure.
Observed: $545.4M direct versus $4.50B independent — a ratio of 8.25:1 against the refutation condition. The largest single category is money spent to defeat a candidate ($2.56B).
Legislators are financed by people who cannot vote for them
Refuted if: the median legislator's out-of-constituency funding share is below 25%.
Observed: across 519 sitting members matched to committee finance, the median share of itemised individual contributions originating outside the state they represent is 45.3%. By chamber:
| Chamber | Members | Median out-of-state | Median ≤$200 share |
|---|---|---|---|
| Senate | 96 | 71.8% | 18.6% |
| House | 423 | 40.1% | 3.0% |
The median senator draws roughly seven of every ten itemised dollars from people who cannot vote for them. Computed from 58,208,756 individual contribution records aggregated to committee × state × amount band.
Caveat: this is a share of itemised contributions. Contributions under $200 are not universally itemised, so the small-dollar column understates genuine small-donor participation and should not be read as a participation rate. The out-of-state share is not affected by this, since it is a ratio within the same itemised population.
Money operates on agenda control rather than vote-switching
This was the natural successor hypothesis after H2's refutation, and was recorded here as untestable — a deliberate guard against rescuing H2 by appealing to a mechanism nobody had measured. On 10 August 2026 the data needed to test it was obtained: the govinfo bulk bill-status corpus for the 118th House, 10,564 bills parsed to sponsor, committee, and whether each advanced.
Refuted if: the correlation between a sponsor's funding and the share of their bills that advance is |r| < 0.35.
Observed, across 349 sponsors with 5 or more bills:
| Party | Sponsors | Bills reported out | r (PAC share × reported) | r (total raised × reported) | r (PAC share × floor action) |
|---|---|---|---|---|---|
| Republican | 176 | 52.2% | 0.185 | −0.016 | −0.014 |
| Democrat | 173 | 45.0% | 0.117 | −0.076 | 0.030 |
Verdict: refuted as specified. Every coefficient falls far below the threshold, and three of six are effectively zero or negative. A better-funded member's bills are not measurably more likely to leave committee or reach the floor.
Both candidate mechanisms have now failed. Vote-switching (H2) and agenda advancement (H5) were the two observable routes by which concentrated money was supposed to become concentrated control. Neither shows an association worth the name. The structural findings — H1, H3, H4 — stand undisturbed; the behavioural claim that would connect them to governing outcomes does not.
Scope of the test: this measures bill advancement by sponsor. It does not test committee-chair power, leadership scheduling, or the content of bills as drafted. H5 is refuted in the form specified, not eliminated as a family of hypotheses — and that distinction is exactly what the qualification on H2 warned about.
What would refute the paradigm as a whole
Individual hypotheses are cheap to adjust after the fact. So the register also commits to conditions that would count against IOCE at the level of the framework:
- If OCDI-Gov turns out to be uniformly low across legislators — funding distributions closely tracking electorates — then the divergence IOCE claims to detect does not exist in this domain, and the framework does not transfer from firms to governments.
- If OCDI-Gov turns out to be uniformly maximal for every legislator regardless of circumstance, the measure is not detecting variation in the world but an artefact of its own construction, and should be discarded as non-diagnostic.
- If divergence scores fail to correlate with any independent indicator of capture — across multiple cycles and chambers — then OCDI-Gov measures something real but politically inert, and the normative claims attached to it are unlicensed.
H2 was refuted and is published as refuted. If later versions of this report report a finding that reverses it, that reversal must be accompanied by the specific data and specification change responsible — not by a silent edit to this page. Version history is retained.
A note on the logical form
The inference this report is tempted toward has the form:
(1) Finance is concentrated.
(2) Concentrated finance produces concentrated control.
∴ (3) Control of government is concentrated.
Premise (1) is measured and supported. Premise (2) is the load-bearing one, and H2 tested a specific observable consequence of it and failed to find it. The argument is therefore valid but not yet sound. What the data currently supports is the weaker claim: the distribution of resources that determine electoral viability is highly unequal — a structural fact — without an established behavioural transmission channel.
Conflating the structural claim with the behavioural claim is the single most likely way for this report to mislead. See the Assumption Audit and the Adversarial Review.