Adversarial Review

"Your central mechanism is unmeasured"

Objection: Having refuted the vote-buying hypothesis, the report retreats to agenda control — which it does not measure. This is unfalsifiable in its current form.

Response when the objection was first put: conceded. Agenda control was recorded as H5 with the status "untestable with current data" rather than "probable", and no confidence was added to it after H2's refutation — the prior was deliberately held at 0.60 rather than raised, precisely because raising it would have been the rescue this objection warns about.

Response now: the objection has been answered the only way it could be, by measuring. The govinfo bill-status corpus for the 118th House — 10,564 bills matched to their sponsors — made H5 testable, and it is refuted as specified: funding share correlates with bill advancement at r = 0.185 (Republican) and 0.117 (Democrat), and correlation with total money raised is negative. The prior fell from 0.60 to 0.25 — see the Bayesian Assessment. Agenda control is no longer an unfalsifiable retreat; it is a tested and failed hypothesis.

What survives of the objection: its general form, and it now bites harder than when it was made. Both observable transmission mechanisms have been tested and neither holds, so the report has no measured route from concentrated money to concentrated control. What remains — bill content as drafted, committee chair power, leadership scheduling, candidate selection before anyone files — is narrower, harder to observe, and still unmeasured. No prior is raised for any of it here.

"Party unity is the wrong dependent variable"

Objection: Party unity is near-ceiling (92–97%) with little variance to explain. Finding weak correlation against a near-constant is close to uninformative; the test was underpowered by construction.

Response: This is the strongest objection and it lands. A better design would test low-salience votes where discipline is weak, or amendment votes, or committee behaviour. H2's refutation should be read as "no effect detected in a place where detection was hard", not as strong evidence of absence. This qualification is now recorded in the register.

"One cycle proves nothing"

Objection: 2024 was a presidential year with idiosyncratic spending. No trend can be inferred from a single cross-section.

Response: Accepted for trend claims; no trend is asserted. The structural measurements are complete-population facts about 2024, which is what they are presented as. Multi-cycle comparison is scheduled.

"Aggregating donors destroys analytical value"

Objection: Refusing to publish individual donor records weakens the analysis.

Response: Rejected. Every measure here operates on distributions; none requires identifying a private individual. The aggregation is lossless for the questions asked and avoids constructing a targeting tool. The raw records remain public at the FEC for anyone wishing to verify.

"OCDI-Gov is an arbitrary construction"

Objection: The authorisation/finance substitution is stipulated. Different stipulations give different numbers, so the index measures a definition rather than a phenomenon.

Response: Partly conceded — the stipulation is explicit and defended rather than derived. The defence is that it is discriminating (varies across legislators), falsifiable (can come out low) and reproducible. Those are the properties that separate a useful construct from an arbitrary one, but they do not make it the only defensible choice.

"The title overstates the coverage"

Objection: "Federal, State, Local and Individual" describes a report that covers federal money and federal votes only.

Response: Conceded, and corrected on 10 August 2026. The report is now titled Control & Ownership of the US Federal Government, and the standfirst states the scope directly. This was the one referee objection that could be resolved by editing rather than by collecting more data — the remaining conceded objections (R1, R2) require new measurement and stand open.