Preregistration package: S6 "Right to Honest Government" ballot study (s6, s6-short, and the s6-verasight panel wave)

This is the complete, self-contained preregistration for the S6 ballot study, its trimmed sibling S6-short, and the S6-Verasight representative-panel wave. Everything needed to register and audit the study is in this one document: the hypotheses (section 2), the design (section 3), the sampling and AAPOR-style disclosure (section 4), the full verbatim instrument (section 5), the randomization spec (section 6), the exclusion rules (section 7), the analysis plan (section 8), the reproduction kit (section 9), the confirmatory-versus-exploratory split (section 10), the trimmed s6-short arm (Appendix A), and the s6-verasight panel wave (Appendix B). Nothing here depends on a separate companion file.

Companion materials, for convenience only (this document does not require them):

• The study packet, with every file and a downloadable ZIP: S6 study packet.
• The machine-readable instruments: rthg-s6.qsf and the panel-wave build rthg-s6-verasight.qsf, generated from the study definition rthg-study-definition.ts.
• The design rationale, in plain English: S6 methodology rationale.

All instrument text below is quoted verbatim from source. Question wording is not paraphrased. Markdown **bold** markers inside ballot and mechanism text are part of the rendered instrument and are preserved as-is.

Registration status

As of this registration, the confirmatory wave has not been fielded and no confirmatory S6 data has been collected or analyzed. Fielding to date is limited to small operational smoke tests (see section 4) used only to check survey plumbing and fair pay; no H1 through H4 estimate has been computed on S6 data, and the representative confirmatory wave has not yet run. No s6-verasight data has been collected and no HV or H estimate has been computed on s6-verasight rows (Appendix B). This document is therefore a preregistration, not an analysis of already-collected data.

This is the single canonical S6 preregistration. Other S6 prereg files in the repository are pointers to this document and should not be treated as independent versions.

OSF record (note added 2026-08-26, after registration). This package, with the canonical Qualtrics survey file attached, was registered on OSF at https://osf.io/x5ve6 on August 11, 2026, before any confirmatory fielding. The registered copy on OSF is the frozen version of record; this line is a website-only annotation recording that fact and is not part of the registered document.

Prior wave (S5). The earlier S5 wave was preregistered on OSF (osf.io/br3u2), posted before the S5 data were analyzed, and remains the public integrity anchor for S5. That public entry was deliberately kept minimal at the time to avoid exposing the legal-framework plans prematurely to the public or to parties who might prepare a strategy against the initiative. This S6 package is the full successor to that entry; the plain-English S5 record lives at docs/PREREGISTRATION.md.

1. Study identity

Source: lib/rthg-study-definition.ts.

• STUDY_ID = 'rthg-study-v4'
• STUDY_VERSION = '2026-08-04.s6.2' (the .s6.2 revision changes only the welcome subtitle wording; see the change log in Appendix B)
• STUDY_NEUTRAL_SPONSOR_LABEL = 'American Institutions Study 2026', the only sponsor label shown to participants before debrief.
• RECRUITMENT_TARGET_N = 1000; TOTAL_TARGET_N is derived from the active cells' target N. This constant is a non-binding planning default carried in the study definition, not the committed confirmatory target: the confirmatory wave's N is set per section 4 and recorded at close.
• Waves covered: s6, s6-short, and s6-verasight (the panel wave, consolidated in Appendix B). Each row carries an arm identifier (A-s6-mitigated-formatA-OFF for s6, A-s6-short-mitigated-formatA-OFF for s6-short, s6-verasight-mitigated-formatA for the panel wave). These are fixed data-partition keys, not a contrast: there is only one design, so the words inside the identifier do not name a chosen option against alternatives. The label vocabulary (mitigated, formatA, OFF) is inherited from the S5 design, where format and pre-framing were live experimental conditions; in S6 there is a single fielded cell and these words are retained only as stable partition keys, not as a hidden factorial. Rows are partitioned by arm and payload.meta.studyVersion ('s6' vs 's6-short') so the two never mix.

2. Hypotheses / research questions

These hypotheses are shared by the S6 and s6-short arms. They are estimation/description hypotheses, not causal-manipulation hypotheses. There is one fielded condition and no between-subjects manipulation.

• H1 (principle agreement): a majority of U.S. adults agree with the underlying governance principles, measured by the principles Likert.
• H2 (ballot support): a majority would vote for the ballot measure on first reading.
• H3 (support after mechanism): support is maintained, or moves in a stated direction, after respondents see how the citizen-review mechanism works (post-mechanism revote).
• H4 (mechanism preference): among four mechanism designs, the forced-choice distribution identifies which design respondents most prefer.

The s6-verasight panel wave adds three wave-specific hypotheses, HV1 through HV3 (rung-1 assent, trust bands, and the exploratory rung-1-versus-ballot gap), defined in Appendix B section B4.

3. Design summary

Single-condition instrument: there is no between-subjects manipulation, and every respondent on both arms sees the same design. That design is per-option Likert ratings of the four review designs plus one neutral forced choice, a built-in-protections summary on the mechanism page, and a separate panel-composition page.

s6 vs s6-short differences (consolidated in Appendix A):

• s6-short mechanism rating is opt-in (rate at least one) rather than forced rate-all-four.
• Panel composition is optional/collapsed in s6-short (skip or expand).
• s6-short revote drops the self-reported "did anything change" yes/no item, but keeps the merged one-line revote reason (required on both arms).
• The reflection page is dropped in s6-short.
• Per-mechanism open text and mitigation feedback are dropped in s6-short.

The demographic region and us-citizen items are dropped in all S6 arms (full and s6-short alike): state is retained and region is reconstructable from it, and citizenship is handled by the welcome-page screener, so neither item is re-asked in demographics.

Pre-divergence metrics (H1, H2) are measured under identical context in both arms and are poolable. Post-divergence metrics (H3, H4) are reported primary-from-S6 with s6-short as large-N corroboration, pooled only if the preregistered arm-effect check shows no material difference.

The s6-verasight panel wave fields the full S6 instrument with a small set of wave-specific changes (two pre-exposure lead pages, a reworded early trust item, demographics reduced to the political items, and no debrief interview opt-in), all stated verbatim in Appendix B.

4. Sampling and recruitment

The s6-short staged scale-up and fair-pay gate are detailed in Appendix A.

• Platform: an online survey panel (Prolific or Verasight). The instrument is fielded close to verbatim on whichever panel is used; the panel actually used is recorded at field time.
• Primary confirmatory wave: the headline H1 through H4 estimates are reported primarily from a representative or quota-balanced panel (for example Verasight), fielded close to verbatim to the locked instrument and reported with that provider's sampling targets and any provider weighting. Earlier and lower-cost Prolific fielding is treated as developmental and corroborative, not as the confirmatory headline.
• Sample size (planning estimates, not locked): the figures here are planning estimates, not committed targets; final N for each wave is recorded at close. The representative confirmatory wave is estimated at N around 1,000, and likely more (scaling toward 2,000 to 3,000 if needed for stable subgroup estimates by party). A large-scale s6-short corroboration on a low-cost panel such as Prolific is estimated at roughly 3,000 to 10,000, but may be substantially fewer. Because these are estimates, fielding more or fewer than a planning figure is not in itself a deviation; a change to the confirmatory-wave design or analysis plan would be.
• Operational pilots excluded: small operational pilots and smoke tests (for example about 30 respondents fielded on the project's own site to check plumbing) are not part of any confirmatory sample and are excluded from the analysis sample.
• No interim peeking for hypothesis tests: the staged checkpoints are operational (fair pay, data quality), not hypothesis-testing looks. Confirmatory analyses run on the full preregistered sample.
• Panel-type caveat: the sample is drawn from an online survey panel. On a nonprobability panel (such as Prolific), generalization to U.S. adults depends on demographic representativeness, not probability sampling, and Wilson confidence intervals are reported as precision summaries, not as classical margins of error implying a probability design. On a quota-balanced or representative panel (such as Verasight), the provider's sampling targets and any provider weighting are reported alongside the result, and the Wilson intervals are still reported as precision summaries.

AAPOR-style disclosure

• Mode: online self-administered questionnaire.
• Sample source: an online survey panel (Prolific or Verasight). Panel type (nonprobability vs quota-balanced or representative) and any provider weighting are recorded at field time.
• Eligibility: U.S. citizens aged 18+. Age and U.S. location are applied at panel recruitment; U.S. citizenship is confirmed by the in-survey welcome screener, which hard-blocks anyone who answers "No." (On Prolific the recruitment prescreen is U.S. residence; citizenship itself is enforced by the in-survey screener.)
• Weighting: none applied by the research team; any panel-provider weighting or quota targets (for example a representative panel's census-style targets on age, gender, region, education, and race or ethnicity) are reported as provided. Party post-stratification (Gallup 27D / 27R / 46I) reported as a robustness check.
• Questionnaire: the full verbatim instrument is reproduced in section 5 of this document.
• Incentive: participants are compensated by the panel provider; the per-complete reward is recorded at field time.
• Precision: proportions are reported with Wilson 95% intervals as a precision summary, not as a classical margin of error (see the panel-type caveat above).
• Field dates: to be recorded at close.
• Final N: to be recorded at close.
• Sponsor: Right to Honest Governance Institute (c3 nonprofit). The sponsoring organization's identity is withheld until debrief: participants see only the neutral label "American Institutions Study 2026" as the sponsor, never the Institute's name or its interest in the subject matter. The survey topic is not concealed. The recruitment listing describes a short survey about a proposed state ballot initiative on government accountability, and the welcome page describes a public report on how Americans think about government decision-making. What the neutral label hides is sponsor identity, not the topic.

Ethics and participant protection

Participation is voluntary and consented (welcome-page consent; 18+; respondents may stop at any time). The sponsoring organization's identity and its interest in the subject matter are withheld until debrief, a deception by omission limited to sponsor identity; the survey topic is disclosed at recruitment and on the welcome page, and every respondent receives a full debrief naming the sponsor and the publication plan. Data are de-identified before analysis or release, and only de-identified data and anonymized quotes are published. Participants are compensated by the panel provider at or above its fair-pay floor. No personally identifying information is required, and the optional paid-interview opt-in has no effect on compensation.

5. Full instrument (verbatim)

The machine-readable survey is also provided as a Qualtrics import file, qsf/rthg-s6.qsf (with qsf/rthg-custom-css.css and qsf/README-styling.md), included in the S6 materials packet. It is generated from rthg-study-definition.ts, so the verbatim text below, the live site, and the Qualtrics or Verasight fielding are the same instrument.

Page order for s6 (source: S6_PAGE_SEQUENCE, rthg-study-definition.ts): welcome/consent + citizen screener, section-a instructions, Section B principles, S6 ballot + vote, S6 mechanism, panel composition, revote, open reflection, Section G demographics, section-h-debrief.

s6-short trims (marked inline below; consolidated in Appendix A): mechanism rating becomes opt-in, panel composition collapsed/optional, revote trimmed, reflection page dropped, per-mechanism open text and mitigation feedback dropped. (The region and us-citizen demographic items are dropped in all S6 arms, not just s6-short.)

5.0 The instrument at a glance

The respondent moves through ten pages, one block per page. This table is the map; the verbatim wording for every page follows in 5.1 through 5.10.

5.1 Welcome / consent + citizen screener

Page id: welcome. Source: app/rthg-study/pages/WelcomePage.tsx; TRUST_WARMUP and sponsor label from rthg-study-definition.ts.

Page title: "A short research survey"

Subtitle (verbatim, with sponsor label interpolated): "About 12 to 20 minutes. Your answers are anonymous and will be used to produce a public report on how Americans think about government decision-making. This study is sponsored by American Institutions Study 2026."

(2026-08-04, study version 2026-08-04.s6.2: the earlier "combined with several hundred others" aggregation phrasing was removed; the consent bullet reserves the right to quote short written answers in anonymized form, so the subtitle must not imply aggregate-only reporting. S5 keeps its locked wording. See MATERIALS_CHANGELOG.md.)

Screener question (us-citizen-screener, required, hard-blocks on "No"):

Are you a U.S. citizen?

• Yes
• No

Block message shown on "No": "Thank you for your interest. This study is open to U.S. citizens only, so you are not eligible to participate. Please return the study on the recruitment platform."

"Before you begin" consent block (shown to citizens), verbatim bullets:

• Please answer based on what you actually think, not what you think we want to hear.
• You can take a break at any point; your progress is saved between pages.
• A few items are worded to check whether you're reading each statement carefully. Please take your time. Thoughtful responses are what your compensation is paid for.
• We publish the combined results and the analysis code so anyone can verify how the numbers were calculated. Any data we release is de-identified. We may quote short written answers in anonymized form.
• By continuing, you confirm you are at least 18 years old and consent to participate.

Consent checkbox (consent, required): "I'm at least 18 years old, I've read the above, and I agree to participate."

Trust warm-up Likert (trust-warmup, required, 1-7; endpoints "Not at all" / "Completely"):

How much do you trust the federal government to do what is right most of the time?

5.2 Section A instructions

Page id: section-a-instructions. Source: rthg-study-definition.ts (SECTION_A_INSTRUCTIONS_A_B).

Heading: "A quick note before the next section"

• You'll see a short list of statements about how government should work. For each one, tell us how strongly you agree or disagree.
• There are no right or wrong answers.
• Please rate each one based on what you actually believe.

5.3 Section B: principles Likert (14 items)

Page id: section-b-principles. Source: rthg-study-definition.ts: PRINCIPLES, B_BAL, B_ATT, B_REV, SECTION_B_ITEMS, SECTION_B_ORDER_CONSTRAINTS.

All 14 items use the 7-point scale: low "Strongly disagree", high "Strongly agree", with a separate no-opinion button: "I haven't thought about this enough to say" (stored as a distinct string, not a scale point). The 14 items are 1 cross-partisan-vocabulary item (B-bal), 11 governance principles (P1-P10, P6-narrow), 1 attention check (B-att), and 1 reverse-coded honesty check (B-rev). Of the 14 items, 13 are substantive (the 11 principles plus B-bal and B-rev, both of which are honest-government principles in their own right and counted in the agreement index). The 14th item (B-att) is purely an instructed attention check and is not counted in the agreement index.

Fixed slots: B-bal at position 1, B-att at position 7, B-rev at position 11. The 11 principles fill the remaining slots in stratified-random order (see section 6).

B-bal (balance item; vocabulary-balanced rephrasing of P2; required):

Government officials of both parties should explain their decisions honestly to the people they serve.

P1 (Article I §2(a) Loyalty):

Government officials serve the public, not private interests, and have a duty to act in the public's best interest.

P2 (Article II §2 Components of Honest Justification):

When the government makes a major decision, officials should have to honestly explain their reasoning, including the evidence they used, the alternatives they considered, and a response to the strongest objections to their choice.

P3 (Article II §1 What Must Be Justified):

Major government decisions that affect the public should be based on evidence of how the decision achieves the public's goals, not ideology, loyalty to a party, or the preferences of donors.

P4 (Article I §2(c) Good Faith + Article II §2(a) Clear Statement):

When the government makes a significant decision, the public has a right to see the reasoning behind it.

P5 (Article I §2(a) Loyalty + Article VI §1 Prohibited Actions):

Officials should be required to disclose conflicts of interest before voting on, or otherwise deciding, matters that could personally benefit them or their donors.

P6 (Article VI §2 Penalties):

When government officials commit fraud to benefit themselves or their donors at the public's expense, there should be real consequences, including removal from office for egregious cases.

P7 (Article III §1 Citizen Challenge + Article III §2 Remedies):

If a government decision is proven to harm the public for private gain, the public should have the ability to require the government use clear reasoning to make a better choice for the people they serve.

P8 (Article III Enforcement and Remedies):

When officials purposefully harm the public for private gain, if proven, accountability should be possible without waiting for the next election.

P9 (Article VII Application to All Branches):

Ethics rules for government officials should be enforceable, not just aspirational, and should apply equally to officials of every party.

P10 (Article III §1 Citizen Challenge + Article II §3 Response Obligation):

Americans should have a real way to inspect and challenge the government's stated reasoning for major decisions, and the government should have to respond directly, with reasons.

P6-narrow (Article VI §2 Penalties, narrow enforcement trigger):

When a government official puts false statements into the official public record of a major decision, and the evidence shows honest intent is unlikely, consequences should be proportionate to the harm caused and the evidence of intent.

B-att (attention check, position 7, expected answer = 4):

To confirm you are reading carefully, please select '4 - Neither agree nor disagree' for this item.

B-rev (reverse-coded acquiescence check, position 11). The S6 exclusion rule is: exclude when B-rev >= 5 AND B-bal >= 5 (genuine acquiescence, i.e., agreeing officials should be allowed to lie AND should explain honestly). See Section 7.

Government officials should be allowed to knowingly put false reasons into the official public record of a major government decision.

5.4 S6 ballot face + comprehension check + vote

Page id: s6-ballot. Ballot text source: S6_TIER1_BALLOT_TEXT, rthg-study-definition.ts. The S6 face differs from S5 by two locked changes: "and accountability" added to The right, and a recall sentence added to Courts and consequences. **bold** markers are part of the rendered face.

Heading: "Ballot measure". The term "major" carries a tooltip: "Decisions of significant public concern. Existing legal practices prevent the system from being overloaded by frivolous cases while keeping it open to legitimate ones. Specific options for how this is defined are discussed in a later section or in a separate study, depending on the version of this survey you are seeing."

The right:

The people of this state have the right to honest government. Public officials serve as fiduciaries of the people, owing loyalty, care, good faith, and accountability.

The requirement:

Every major government decision must be made with honest reasoning on the public record: the key facts relied on, the alternatives considered, and a response to the strongest objections.

When there is doubt:

When there is credible evidence of dishonest reasoning on the public record, state citizens may petition for review. Courts may promptly dismiss frivolous or duplicative petitions and combine petitions about the same decision.

Who reviews:

Who reviews each petition — and in what combination of courts, randomly selected citizens similar to a grand jury, and independent professional advisors — is the subject of the next page of this survey. There are several reasonable ways to design it. For now, we will call the reviewing body the Public Reasoning Panel. Different review designs appear on the next page. On this page, please rate the rule itself before seeing those design options.

How the review works:

The Public Reasoning Panel asks whether the reasoning was honest, whether it accounted for the key facts on the public record, and whether it was consistent with the official's duty to the public rather than to personal or private benefit. If the panel finds the reasoning failed any of these tests, the decision is sent back to be remade with honest reasoning on a corrected public record.

How findings are made:

Required procedures ensure that the panel's findings are grounded in evidence and reasoned in writing, so its decisions cannot be arbitrary.

Courts and consequences:

Courts keep final legal authority. The panel cannot write laws, cancel laws, or remove officials on its own. In the most serious cases, such as knowingly placing false information on the public record, destroying or concealing evidence, or willfully obstructing review, personal consequences may be imposed only after court findings under heightened proof standards. For those acts, this amendment sets constitutional minimum consequences of removal from current office and disqualification from future office. If anyone in public office fails this duty in a serious or repeated way, citizens could ask the courts to remove them from office, or, in the most serious cases, vote them out of office. Courts may impose additional consequences, including criminal referral. Officials acting in documented good faith are protected from personal penalty.

The goal:

To sufficiently discourage dishonest service of the people and refocus government on: what the people's goals are, how best to achieve them, how to represent them well, and how to bring them together on issues where they genuinely disagree.

Lasting protection:

No statute or official action may materially narrow, delay, or burden the rights and remedies created by this measure. Reasonable rules that speed review, combine duplicate cases, and protect fair process are allowed, but rules that substantially weaken access or remedies are not. Any conflicting law or action is void to the extent a court finds a conflict.

Implementation:

The legislature has two years to set up the system, consistent with the principles of honest governance in this measure, and subject to review. If it does not pass a qualifying law within two years, the default implementation contained within the amendment takes effect automatically until the legislature enacts compliant laws.

Comprehension check (smoke4-comprehension-check, required, placed between the ballot body and the vote). Any answer unlocks Continue; an incorrect answer is a downstream quality flag, not a hard block.

Based on what you just read, how long does the legislature have to set up this system before the default takes effect automatically?

• One year
• Two years (correct)
• Three years
• Five years

Initial vote (smoke4-vote, required, 1-7; endpoints "Definitely would not vote for it" / "Definitely would vote for it"):

If this were on your state ballot today, how likely would you be to vote for it?

Vote descriptor (smoke4-vote-describe, required; prompt "Which best describes your answer?"):

• Yes. I'd vote for this law.
• Yes, conditional on the mechanism: I support the principle, but my vote depends on how the review process is structured.
• Maybe. It depends on specifics.
• Probably not, or no.

(The definition file carries per-option open-text prompts for the "Maybe" and "Probably not" options; those conditional boxes are not fielded on any S6 arm. The own-words reason is captured by the smoke4-vote-reason item below, which is shown to every respondent regardless of descriptor choice. Decision of 2026-06-14, corrected in this document 2026-08-10; see the B8 change log.)

Vote reason (smoke4-vote-reason). In s6 required-status = optional; in s6-short emphasized as optional / de-emphasized (S6_SHORT section 4):

In your own words, what is the main reason you voted the way you did?

5.5 S6 mechanism page

Page id: s6-mechanism. Source: rthg-study-definition.ts: S6_MECHANISM_SHARED_CALLOUT, four options in S6_MECHANISM_OPTIONS, forced choice in S6_MECHANISM_FORCED_CHOICE, per-option open text, mitigation feedback.

Shared callout (verbatim, intro + 4 bullets):

Across all four designs, the same rules apply:

• Findings must rely on evidence that meets the Source Reliability Framework.
• Reviewers must follow the Honest Service Code.
• Advanced analytical tools are encouraged to speed research and analysis, but tools never decide; the human responsible for the public record verifies the evidence and reasoning personally and signs off on it.
• Courts keep final legal authority.

Source Reliability Framework tooltip (bullet format shown on hover):

A standard for vetting the evidence a review relies on.

• What it checks: track record, independence, methodology, transparency, expert consensus.
• What cannot support findings: sources with proven dishonesty or hidden conflicts.
• What it covers: the sources findings rely on, the sources officials cite in the decision under review, and any analytical tool used in the review.
• What this does not mean: weak or unreliable material can be examined; findings cannot rest on it.

Honest Service Code tooltip (bullet format shown on hover):

A standard for anyone serving on a panel.

• Follow the evidence where it leads. Even when it cuts against your first view.
• Use the same standards for every side. Apply them no matter who is being reviewed.
• Serve the public. Not yourself, a party, a donor, or an interest group.
• Disclose conflicts. Step aside when you cannot be neutral.
• No tampering or pressure.

Built-in protections across every design (verbatim, 6-bullet mitigation blurb):

Assume these protections apply.

• Capture protection. Conflict screening excludes anyone who benefits; tampering or intimidation is a serious crime.
• Gaming protection. Same Source Reliability Framework and Honest Service Code apply across every design, with written reasoning and penalties for bad-faith filings.
• Expertise support. Findings stay with the panel. Independent counsel runs analytical tools and advisors at the panel's direction, under human responsibility.
• Speed. Modern technology and a written record speed review.
• Weaponization protection. The same rules apply to every party and every official. Anyone filing or pushing a dishonest, retaliatory, or harassing case faces the same evidence-grounded review and can be penalized, including removal from office in serious cases.
• Uniform Fiduciary Duty. Public officials of every party and branch are held to the same fiduciary duty standard to act with loyalty, care, and good faith on behalf of the public. Courts follow enforceable ethics rules, and must explain any departure from the panel record in writing.

Four mechanism designs (option order randomized per respondent):

Option 1: s4-mech-courts-only "Courts only" Subtitle: "Courts review every petition under deadlines, with modern tools speeding analysis under human responsibility."

A petition that a major decision was made with dishonest reasoning goes directly to a court. The judge applies the three honest-reasoning tests: honest reasoning, key facts in the public record, and consistency with the public duty.

• How evidence is developed. The court uses normal procedures and may consult outside experts as it would in any case, plus modern tools that help sort filings, surface credible counter-evidence, and run consistency checks against the public record, under human responsibility.
• No rubber-stamping. A human lawyer or judge must engage with and verify the analysis on the record before any ruling. Courts cannot accept it thoughtlessly, and they cannot brush past credible findings without explaining why, on the same honesty standard officials face.
• Sources. The same Source Reliability Framework that governs the case also governs which sources the court relies on. No citizen panel.

Option 2: s4-mech-citizen-grand-jury "Citizen panel with source-credibility audit" Subtitle: "Random citizens run a source-credibility audit, with court review afterward."

A grand-jury-style panel of citizens is drawn from ordinary jury lists. The panel audits the official's published reasons using the same Source Reliability Framework and honesty standards the amendment requires of officials.

• How evidence is developed. Modern tools help the panel review the record, test whether sources and data support the stated conclusions, surface credible alternatives, and spot omissions, under human responsibility.
• Professional support. An independent legal officer (non-partisan, no stake in the matter) runs the analysis at the panel's direction, drafts summaries the panel reviews and revises, and helps with procedure. The panel can replace this officer for cause and can direct its own research. Outside advisors with subject-matter expertise may be summoned when the panel wants them.
• What the panel can do. Request additional records, send written questions, and require the official to answer surfaced credible alternatives on the public record. Courts review afterward and keep final legal authority.

Option 3: s4-mech-jury-chooses-path "Citizen routing panel chooses the review path" Subtitle: "Random citizens decide which review path fits the case: courts, a full citizen panel, or an expert-prepared review with citizen oversight."

A small grand-jury-style routing panel reviews each petition first. Its only job is to decide which review path fits this specific case.

• The choices. Direct review by a court, a full citizen panel like the one above, or an expert-prepared review with citizen oversight.
• Same standards. The routing panel uses the same modern tools described above (under human responsibility), the same Source Reliability Framework, and the same independent legal-officer support the downstream path will use, so its routing decision is itself reasoned and on the record. It chooses only among credible options. Routing a case to a path that would fail the three tests on its face is not allowed.
• What happens next. Whichever path it picks then applies the three honest-reasoning tests to the decision under review.

Option 4: s4-mech-audit-no-jury "Public audit with court referral" Subtitle: "State citizens may file an audit request; verified audits trigger court review."

State citizens may file a formal audit request alleging that the public record for a major decision is incomplete or dishonest. Citizen watchdog groups may help prepare the request.

• What the request must show. It must apply the same Source Reliability Framework to its own evidence. It may be prepared with help from the same modern tools described above, under human responsibility.
• Verification. An independent auditor (non-governmental, held to strict standards against deliberate gaming, with documented good-faith mistakes correctable rather than career-ending) verifies the audit.
• What happens next. A verified audit triggers court review of the decision under the three honest-reasoning tests. No standing citizen panel.

Per-option rating. In s6, each of the four designs is rated on a Likert (forced rate-all-four). In s6-short, rating is opt-in (rate at least one), exploratory only.

Per-option open text (s6-mech-open-{optionId}, optional). Dropped in s6-short.

Briefly: what makes this design work or not work for you? A sentence is plenty.

Forced choice (s6-mech-forced-choice, required). Confirmatory mechanism-preference measure (H4):

If these four designs appeared on a real ballot today, which one would you vote for? Helper: "Pick one. The order on this page is randomized; the order in this list mirrors the order of the four designs above." "None" option: "None of these four. I would prefer something else."

In s6-short, conditional voters (vote descriptor = conditional-on-mechanism) additionally answer a would-support / would-not selection across the four designs.

Mitigation feedback (s6-mech-mitigation-feedback, required). Dropped in s6-short. The 4-option multiple choice is kept; its follow-up open text is no longer conditional on the chosen option. A single always-shown, required follow-up (s6-mech-mitigation-feedback-open) with an explicit N/A escape replaces the per-option prompts.

After reading the protections above, how well did they answer the questions or concerns you had, if any?

• They answered the main concerns I had.
• Some concerns were addressed; others were not.
• They did not answer my concerns.
• I did not have concerns to begin with.

Follow-up (s6-mech-mitigation-feedback-open, always shown, required): "Which protections mattered most for you? What concerns were not addressed, if any? Or if no concerns, why did you feel confident already? You can choose not to answer by writing 'N/A' or 'None'."

5.6 Panel composition

Page id: s6-panel (same questions as S5, S5_PANEL_PAGE constants). In s6-short this page is collapsed/optional (skip or expand); the opt-in rate is itself recorded.

Heading: "How should the citizen review panel be set up?" Subtitle: "Now that you have rated the designs, please tell us how the panel itself should be set up." Intro reminds respondents that baseline rules still apply (conflict-of-interest screening, the Honest Service Code, anti-tampering protections, "just like a jury").

Q1 s5-panel-selection (required): "How should panel members be selected?"

• All seats filled randomly from ordinary jury rolls, with conflict-of-interest screening.
• Mostly random, with a minimum number of seats reserved for civic or professional members who pass independent screening on professional skills, ethics, and knowledge.
• All seats filled by nomination from professional or civic associations, with conflict-of-interest screening.
• Not sure.

s5-panel-selection-share (always shown, required, with a "Not applicable" option): "If you chose to reserve some seats above, what minimum share of seats should be reserved for screened civic or professional members? If you did not choose reserved seats, select 'Not applicable.'" (Promoted from an S5 conditional to an always-shown required item so it stays inline on the panel page; Qualtrics pushes true conditional text/choice items onto their own page.)

• Not applicable (I did not choose reserved seats).
• About 10% reserved (mostly random).
• About 25% reserved.
• About 40% reserved.
• About 50% reserved (an even mix).
• Not sure.

Q2 s5-panel-eligibility (required): "Beyond the minimum screening above, what additional screening should panel members go through?"

• No additional screening.
• Civic-knowledge screening for all panel members (basic understanding of how laws and government work).
• Civic-knowledge screening for all panel members, plus professional-skills screening for any reserved seats (the screened civic or professional slots in the mixed model above).
• Not sure.

Q3 s5-panel-compensation (required): "How should panel members be paid and protected while serving?"

• Treated like jury duty: small daily stipend plus job protection.
• Treated like skilled labor: professional-grade pay plus job protection.
• Volunteer only: no pay and no required job protection.
• Not sure.

Open s5-panel-open (optional): "What, if anything, about panel setup still feels missing, unfair, or worth changing?"

5.7 Revote

Page id: s6-revote. Source: SMOKE4_REVOTE_VOTE, SMOKE4_REVOTE_DESCRIBE, SMOKE4_REVOTE_CHANGED, S6_REVOTE_REASON in rthg-study-definition.ts.

Revote Likert (smoke4-revote, required, 1-7; same endpoints as initial vote):

Now that you have considered how the review might work, how likely would you be to vote for this measure?

Revote descriptor (smoke4-revote-describe, required): same four options as the initial vote descriptor (section 5.4).

Changed checkpoint (smoke4-revote-changed, required). Dropped in s6-short.

Did anything change for you? Helper: "It's fine to say no. We just want to know whether the designs moved you at all, in either direction."

• Yes, something shifted, even a little.
• No, my view did not really change.

Revote reason (s6-revote-reason, always shown, required on both arms). Merges the former "main reason" and "what shifted" prompts into one. Kept in s6-short (only the smoke4-revote-changed yes/no above is dropped there).

What's the main reason for the way you voted? If anything changed, what specifically changed when you saw the mechanism options?

5.8 Open reflection (s6 only; dropped in s6-short)

Page id: reflection page. Source: S6_OPEN_PAGE_PROMPTS in rthg-study-definition.ts.

• s6-recall (required): In one or two sentences, how would you explain this proposal to a friend in your own words?
• s6-pull-pause-mitigation (required): What pulled you toward your vote, what gave you hesitation, and what change would move you the most? Even a rough idea is valuable.
• s6-attention (optional): What stood out most as you were reading? Any word, sentence, or idea that drew your attention.

5.9 Section G: demographics

Page id: section-g-demographics. Source: SECTION_G_ITEMS in rthg-study-definition.ts. In all S6 arms the region and us-citizen items are dropped (state retained; citizenship is handled by the welcome-page screener).

party (required): "Which of the following best describes your political affiliation?"

Democrat / Republican / Independent / Libertarian / Green / Other / Prefer not to say

party-lean (shown if party is Independent/Libertarian/Green/Other; optional): "If you had to lean one way, which would it be?"

Closer to Democrats / Closer to Republicans / Truly independent / Prefer not to say

vote-2024 (required): "Which candidate did you vote for in the 2024 U.S. Presidential election?"

Kamala Harris (Democrat) / Donald Trump (Republican) / A third-party or independent candidate / I did not vote / I was not eligible to vote / Prefer not to say

age (required), age buckets:

18–24 / 25–34 / 35–44 / 45–54 / 55–64 / 65–74 / 75 or older

gender (required): "How do you describe your gender?"

Woman / Man / Non-binary / Prefer to self-describe / Prefer not to say

state (required): "Which U.S. state do you live in?" (50 states + District of Columbia + "I do not live in the U.S." + "Prefer not to say").

(The us-citizen and region demographic items defined in SECTION_G_ITEMS are dropped in all S6 arms: citizenship is gated by the welcome-page screener and region is reconstructable from state. They remain in the shared definition only for the locked S5 instrument.)

race-ethnicity (multi-select, required): "Which of the following best describes your race or ethnicity? (Select all that apply.)"

White or European / Black or African American / Hispanic or Latino/a / Asian / Native American or Alaska Native / Native Hawaiian or other Pacific Islander / Middle Eastern or North African / Other / Prefer not to say

education (required): "What is the highest level of education you have completed?"

Less than high school / High school diploma or GED / Some college, no degree / Associate's degree / Bachelor's degree / Graduate or professional degree

final-attention-check (required; correct answer marked): "To confirm careful reading, what was this survey about?"

• How grocery prices have changed over the last year
• Principles for how government should explain major decisions (correct)
• Tips for improving personal productivity
• Opinions about professional sports

5.10 Debrief

Page id: section-h-debrief. Source: DEBRIEF_TEXT, DEBRIEF_FOLLOWUP_OPTIN in rthg-study-definition.ts.

Thank you for completing this survey.

This research was conducted to better understand how Americans think about the principles underlying government accountability and about the specific legal mechanisms that would give those principles practical effect. Your answers will help refine those mechanisms before any public advocacy or ballot campaign.

The survey was sponsored by a nonprofit research effort at the Right to Honest Governance Institute (righttohonestgovernance.org). The name of the sponsor was not shown at the start of the survey because revealing it would have risked biasing responses toward or against a specific political framing. Now that your answers are locked in, we can share this context with you.

We will publish the combined results from this study along with the analysis code used to produce them, so anyone can verify how the findings were calculated. Any data we release is de-identified. We may quote short written answers in anonymized form.

If you're interested in a paid follow-up interview to discuss your answers in more depth, please indicate your interest below. Participation in follow-up research is entirely optional and has no effect on your compensation for this study.

Follow-up opt-in (followup-optin, optional): "Are you interested in a paid follow-up interview (approximately 30 minutes, USD 25)?" S6 uses provider-neutral opt-in copy (S6_DEBRIEF_FOLLOWUP_OPTIN) so the same instrument can run on Prolific, Verasight, or any panel; S5 kept Prolific-specific wording.

Yes, you may contact me through the survey provider / No, thank you

6. Randomization spec

Source: orderSectionBItems(seed, { clusterConstraint }), lib/rthg-cell-assignment.ts; fixed slots in SECTION_B_ORDER_CONSTRAINTS, rthg-study-definition.ts.

• Section B order: stratified-random per respondent, seeded deterministically from the respondent assignment seed. The 11 principles are shuffled (Fisher-Yates). S6 uses a plain shuffle: the S5 no-three-consecutive-same-cluster redraw constraint is dropped for S6 so the live (Azure) instrument and the Qualtrics/Verasight instrument randomize identically (Qualtrics cannot reproduce the cluster rule). In code the constraint is opt-in; S6 call sites pass clusterConstraint: false. S5 keeps the constraint.
• Fixed slots: B-bal at position 1, B-att at position 7, B-rev at position 11. The shuffled principles fill positions 2-6 (5), 8-10 (3), and 12-14 (3).
• Mechanism options: the four mechanism designs are presented in a per-respondent shuffled order; the forced-choice list mirrors that same order.
• Platform difference (reproducible seed vs recorded display order). On the Azure/live-site instrument, both the Section B order and the mechanism-option order are generated from a single deterministic per-respondent seed stored on each row, so any respondent's exact ordering can be regenerated and audited from the seed. On the Qualtrics/Verasight instrument the ordering is produced by Qualtrics' own question randomizer (configured by hand to hold the three fixed slots and shuffle the rest), not by our seed; Qualtrics records each respondent's realized order in its export (the display-order fields), so order is still captured per response and auditable, but via stored display order rather than a reproducible seed. The two platforms yield equivalent randomization with different audit trails.

7. Exclusions

Full derivation and rationale are self-contained in this section. One shared computeResponseFlags() pass scores every response. We report two samples that differ only in how they treat the reverse-coded acquiescence check (B-rev).

Both samples drop a response for any clear quality failure, all fixed before the confirmatory data is collected:

• not a U.S. citizen (the study measures U.S. adults, so non-citizens are excluded from the official counts);
• failed the instructed attention check (B-att, the item that tells the respondent which answer to pick);
• failed a ballot comprehension check (the "Two years" recall);
• failed the end-of-survey attention check (final-AC);
• answered implausibly fast (under 45 seconds on the 14-item battery, or under 45 seconds on the mechanism page);
• did not finish, duplicate Prolific ID, or otherwise ineligible.

All of these are scored at analysis time and are non-blocking, with one exception: the instructed attention check B-att is enforced in-survey in S6. A respondent who does not select the instructed answer ("4 - Neither agree nor disagree") stays on the same page, with all answers intact, and sees an inline notice ("Please check your answers. One item on this page requires a specific response to confirm careful reading, and it was not answered as instructed. Please review your answers before continuing. You have one remaining attempt."). A second submission that still misses the instructed answer ends the survey as a recorded screen-out. Because B-att screens on a content-neutral instructed answer rather than a substantive opinion, gating it removes non-readers without coaching the ballot or principle result. The post-hoc B-att rule above is retained as a backstop for any response that reaches analysis without passing the gate. All other failures (including the comprehension checks) remain recorded-not-blocking, so the survey never makes a respondent retry a substantive item until they "get it right."

The two samples then differ on a single rule:

• Headline sample (the public "X% of Americans" figure): drops every failure listed above but keeps responses whose only problem is B-rev. B-rev catches yea-saying, not inattention, so a "failure" can be a sincere if unusual view. The headline therefore excludes only on clear evidence of non-engagement. This is the broadest sample we treat as valid.
• Confirmatory sample (the strict-clean figure): everything the headline drops, plus acquiescent responders, defined narrowly as people who agree with two contradictory statements at once (B-rev >= 5 AND B-bal >= 5). Someone who agrees officials may lie but disagrees that they should explain honestly is not dropped, because that is a coherent position.

Both figures are reported side by side. On the locked S5 data, applying the stricter acquiescence rule versus not changes the headline support figure (the percentage of respondents voting 5 or higher) by under 0.3 percentage points, so the headline does not hinge on the rule.

s6-short retains the full Section B quality battery (B-att and B-rev both sit on section-b-principles, which is kept in s6-short). It also retains the final demographics attention check (final-attention-check on section-g-demographics). On top of those checks, s6-short additionally relies on Prolific's attention infrastructure, a completion-time floor (threshold set after the smoke median is measured), and standard straight-lining / duplicate checks.

8. Analysis plan

The headline figures are the principle-agreement rate (H1) and the ballot-support rate before and after the mechanism (H2, H3), each reported as a proportion with a Wilson 95% interval on the confirmatory sample, with the headline sample shown alongside for comparison. Everything below is the full estimation plan.

• Support (H2/H3): proportion voting >= 5 on the 1-7 vote/revote Likert, with Wilson 95% confidence intervals; conditional branch reported separately.
• Principle agreement (H1): per-item top-2-box proportion and mean Likert with Wilson CIs, plus an overall agreement index across the 13 substantive Section B items (the 14-item battery minus the instructed attention check B-att).
• Mechanism preference (H4): forced-choice distribution across the four designs with CIs; opt-in per-option ratings reported as exploratory alongside their opt-in denominator.
• Pre/post movement (H3): net movement = post-mechanism support minus pre support at the aggregate level, with a paired / McNemar-style test on the support indicator.
• Robustness: party post-stratification reported as a robustness check.
• Age: age crosstab reported descriptively, not weighted.

Registered claim wording

The public headline claim is registered as one of two fixed sentences, chosen by the panel actually fielded; bracketed values are filled at close. Reporting follows the panel-type caveat in section 4.

• Representative / hybrid panel (for example Verasight): "In a preregistered study of U.S. adults fielded on [panel], a hybrid probability and non-probability online panel, after reading a full description of the proposal as a ballot would present it (its principles and its built-in protections), [NN]% rated their support at 5 or higher on a 1 to 7 scale (Wilson 95% interval [LL]% to [UU]%), using the provider's census-balanced quotas and weighting."
• Non-probability panel (Prolific): "In a preregistered study of U.S. adults fielded on a non-probability online panel (Prolific), after reading a full description of the proposal as a ballot would present it (its principles and its built-in protections), [NN]% rated their support at 5 or higher on a 1 to 7 scale; because the panel is non-probability, the Wilson 95% interval [LL]% to [UU]% is a precision summary, not a classical margin of error."
• Principle agreement (either panel): "In the same study, [NN]% of respondents had a mean of 5 or higher across the 13 honest-government principle items."

9. Reproduction kit

• scripts/rthg-reproduce.ts: end-to-end recompute of published numbers from the de-identified dataset.
• scripts/rthg-snapshot.ts: builds the analysis snapshot.
• lib/rthg-study-flags-s6.ts: S6 exclusion / quality flag definitions.
• lib/rthg-study-flags-router.ts: routes rows to the correct flag set by arm / study version.
• Codebook export: the machine-readable instrument codebook (item ids, prompts, options, scale labels) generated from rthg-study-definition.ts.

10. Preregistered vs exploratory

Confirmatory (preregistered): H1 principle agreement (top-2-box + index), H2 initial support, H3 post-mechanism support and net movement, H4 mechanism forced-choice preference, and the party post-stratification robustness check, all on the strict-clean confirmatory sample. Everything else is exploratory: opt-in per-option mechanism ratings, panel- composition distributions, self-reported vote-change and qualitative reasons, mitigation feedback, reflection/recall, and the age crosstab. The arm-effect check that gates any S6 / s6-short pooling is preregistered; pooling itself happens only if that check shows no material difference.

Appendix A. The s6-short arm (consolidated)

s6-short is not a separate study. It is a trimmed, low-cost arm of S6, registered as part of this package. This appendix gathers everything specific to that arm so it can be read in one place; the verbatim instrument, exclusions, and pooling rules above already mark every point at which s6-short diverges from full S6. No s6-short fielding is currently scheduled; the arm is retained in this package as an optional future fielding route.

A.1 Purpose

s6-short is a shortened version of the S6 ballot study, designed for low-cost fielding to a large Prolific sample. Its job is to estimate, at high precision, the two headline quantities of the ballot study (principle agreement, H1; and ballot support, H2/H3) while cutting the deep-qualitative content that those headlines do not need. It tests the same H1-H4 as full S6 (section 2), under the same single fielded condition (no between-subjects manipulation).

A.2 What s6-short keeps and what it drops

s6-short keeps the full Section B principles battery and its quality checks: the instructed attention check (B-att) and the reverse-coded B-rev both remain on the principles page, and the final demographics attention check is retained. The principle-agreement measure and the core attention and acquiescence screens are therefore identical to full S6.

It trims content the headline estimates do not need:

• mechanism rating becomes opt-in (rate at least one) rather than forced rate-all-four; the forced choice is kept;
• panel composition is optional/collapsed (skip or expand), and the opt-in rate is itself recorded;
• the revote drops the self-reported "did anything change" yes/no item, but keeps the merged one-line revote reason (required on both arms); net movement is still computable from vote versus revote;
• the post-mechanism reflection page (recall, pull-pause, attention open text) is dropped;
• per-mechanism open text and mitigation feedback are dropped.

(The demographic region and us-citizen items are dropped in all S6 arms, not just s6-short: state is retained and region is reconstructable via a fixed state-to-region crosswalk, and citizenship is gated by the welcome-page screener.)

A.3 Sampling, reward, and staged scale-up (preregistered)

This section describes the Prolific low-cost staged path for s6-short specifically. A representative-panel read (for example Verasight at N around 1,000) is an alternative fielding route for the full S6 instrument and uses that provider's pricing, quotas, and turnaround rather than the staged Prolific reward schedule below.

• Platform: Prolific. Eligibility: U.S. citizens, 18+.
• Reward: set at field time and confirmed by the fair-pay gate below.
• Fair-pay gate: field a pilot wave of 10 to 30 participants first and confirm the median completion time against Prolific's fair-pay floor (about $8/hr) before scaling. The pilot is the gate on both the completion-time exclusion threshold and scale-up.
• Staged stopping / scale-up rule: (1) an s6-short smoke of n = 30, alongside a parallel full-S6 smoke of n = 30, fielded as separate Prolific studies so they do not contaminate each other; (2) on a clean smoke and fair-pay confirmation, scale to a mid-size read; (3) then to a large-scale s6-short corroboration estimated at roughly 3,000 to 10,000, which may be substantially fewer. These figures are planning estimates, not committed targets; the final N is set at the mid-size checkpoint based on the precision needed for H1-H4 subgroup reporting and is recorded at close.
• No interim peeking for hypothesis tests: the staged checkpoints are operational (fair pay, data quality), not hypothesis-testing looks. Confirmatory analyses run on the full preregistered sample.

A.4 Data quality for s6-short

In addition to the Section B battery and the final attention check it shares with full S6, s6-short relies on Prolific's approval and attention infrastructure, a completion-time floor (threshold preregistered after the pilot median is known), and standard straight-lining and duplicate checks. Exclusions: non-completes, failed time-floor, duplicate Prolific IDs, and non-eligible respondents (non-citizen or under 18) where detectable. Every rule is fixed before the confirmatory sample is run.

A.5 Pooling rule (how s6-short and full S6 combine)

• Pre-divergence measures are poolable. Principle agreement (H1) and initial ballot support (H2) are measured under identical context in both arms (everything before the mechanism page is the same), so the two arms may be pooled for these headlines, with the arm reported as a subgroup and a preregistered arm-effect check run first.
• Post-divergence measures are pool-with-check. The forced-choice mechanism preference (H4) and especially the post-mechanism revote (H3) are preceded by different mechanism and panel engagement across the arms. These are reported primary-from-S6, with s6-short shown alongside as large-N corroboration, and pooled only if the arm-effect check shows no material difference.
• S6-only measures (per-mechanism rate-all-four means, full-sample panel composition, self-reported vote change and qualitative reasons, mitigation feedback, reflection and recall) are reported from full S6 only.
• Arm-effect check (preregistered). Before any pooling, each headline is reported by arm and tested for an arm difference; if material, the arms are reported separately rather than pooled.

A.6 Open items

• Final reward and the completion-time exclusion threshold are confirmed after the pilot median is known.
• Final target N is a planning estimate set at the mid-size checkpoint and recorded at close.

Appendix B. The s6-verasight panel wave (consolidated)

The Verasight-panel edition of the S6 ballot study: the full S6 instrument (sections 1 through 10 of this document) plus two pre-exposure items asked first and a small set of wave-specific changes, all stated verbatim below. Everything not listed in this appendix is identical to the main registration.

B1. Wave identity

• Data partition: STUDY_ID = 'rthg-study-v4'; rows carry arm = 's6-verasight-mitigated-formatA' and payload.meta.studyVersion = 's6-verasight'. These are internal storage-partition and instrument-variant labels carried from the study platform's development history and are never shown to respondents; "v4" numbers the storage partition, not a wave, and "mitigated-formatA" names the S6 mechanism-page variant this wave inherits. The flags router and the analysis wave map (WAVE_CONFIGS['s6-verasight']) scope these rows so no s6 or s6-short analysis absorbs them, and vice versa.
• Study definition version at registration: 2026-08-04.s6.2.
• Canonical fielded instrument: the Qualtrics survey file rthg-s6-verasight.qsf, SHA-256 54ce168fbd53e2d814ea1b5b8fa43d4a681e02106a0df1a752666172d88700b3, generated 2026-08-11. The build is byte-deterministic, so regenerating from the same source reproduces this exact hash. The QSF is generated from and checked against lib/rthg-study-definition.ts; if the two could ever disagree, the fielded QSF governs what respondents saw and this registration's verbatim text governs what was registered. The file is attached to the OSF registration so the hash is self-verifying. The file's Survey Flow carries the two in-survey gates (the section 5.1 citizenship termination, shown as a "Not eligible" page with the verbatim block message, and the screen-out branch of the instructed-attention-check gate of section 7; the gate's same-page warning rides as question JavaScript on B-att, which requires the importing account to permit survey JavaScript, standard on licensed brands), so an import needs no branching configured by hand. The file also embeds the survey stylesheet (published separately as rthg-custom-css.css); styling is cosmetic and changes no wording, order, or logic.

B2. What this wave is

1. Representative-panel confirmation. Section 4 names a nonprobability panel's frame as the main generalization caveat. This wave fields the locked instrument on a representative or quota-balanced panel (planned: Verasight), reported with the provider's sampling targets and weighting. H1 through H4 keep their definitions from section 2.
2. Bridge to the project's earlier website surveys. The project's earlier public-website surveys (Prolific-recruited studies run on the campaign's public website) headline on agreement with the sentence "The American people have a right to honest government service." (pooled result at lock: 912 of 917 unique participants, 99.45%, cited publicly as 99.4%; receipt dated 2026-08-04). Asking the identical item, identically worded and scaled, as a clean pre-exposure measure lets the panel wave anchor that number on a representative sample by a direct join on the same question, not a fuzzy match across instruments.
3. Trust moderator. The trust-in-government item from those website surveys, asked immediately after, lets us report whether ballot support and rung-1 assent hold across trust levels on a panel whose trust distribution is not Prolific-shaped.

B3. Instrument: deltas from the main registration (and nothing else)

Source of truth in code: S6_VERASIGHT_LEAD and S6_PAGE_SEQUENCE['s6-verasight-mitigated-formatA'] in lib/rthg-study-definition.ts. Wording is verbatim-locked: the study code asserts at module load that these strings are byte-identical to the website-survey constants, so the two instruments cannot drift apart silently.

B3.1 Two new pages, inserted after the instructions page and before Section B

Page order: welcome/consent + citizen screener, section-a instructions, s6-lead, s6-lead-trust, Section B principles, then the S6 sequence (ballot, mechanism, panel composition, revote, reflection, demographics as amended in B3.3, debrief).

Page s6-lead (rung-1 agreement, asked alone). Heading: "A quick question before we begin". Subtitle: "Your first reactions are what we want here. There are no right answers." One required item, key rung1.agree; no "No opinion" option (the website-survey item offers none, and adding one would break the distribution comparison); locked read-only on back.

Statement (displayed as its own stimulus card): The American people have a right to honest government service.

Question: How strongly do you agree or disagree?

Scale: 1-7 agreement, low label "Strongly disagree", high label "Strongly agree".

Page s6-lead-trust (trust in government). Heading: "One more quick question". One required item, key trust.gov; locked read-only on back.

Question: How much do you trust government to do what is right most of the time?

Scale: 1-7, endpoints only, low "Not at all", high "Completely".

The two items were originally designed as one page; they were split onto separate pages (2026-08-04, before any fielding) so the rung-1 statement is answered with nothing else on screen. Item order is fixed, rung 1 first: rung-1 assent is the number this wave exists to bridge, so it is answered before the trust item can prime it.

B3.2 One wording change: the trust item, this arm only

The S5/S6 welcome page asks "How much do you trust the federal government to do what is right most of the time?". On this wave the trust question is instead asked early on its own page (B3.1), worded "How much do you trust government to do what is right most of the time?" to match the earlier website surveys exactly, and the welcome-page version is not asked again: two near-identical trust questions minutes apart would prime each other. The s6 and s6-short arms are unchanged. Consequence, stated ahead of time: s6-verasight rows carry trust under the trust.gov key, and any trust comparison to S5/S6 rows crosses a wording change ("the federal government" vs "government") and must say so.

B3.3 Demographics reduced to the political items, this arm only

The demographics page keeps only its political items, worded identically to S5/S6 so party subgroup comparisons stay direct: party affiliation, the independent-lean follow-up, and 2024 presidential vote. The census-style items that duplicate the panel provider's profiling (age, gender, state and its non-U.S.-resident citizenship follow-up, region, race and ethnicity, education) are removed on this arm only: the provider samples to be representative on those variables and delivers them with the data export, and the welcome-page citizenship screener already covers eligibility.

The end-of-survey recall check ("what was this survey about") is also not fielded on this arm. The panel provider runs its own respondent quality control, and the instrument already carries the instructed attention check with its in-survey gate, the reverse-worded item pair, and the ballot comprehension check; a fourth quality item adds length without adding discrimination on a professionally managed panel. B6 states the effect on the exclusion rules.

B3.4 Debrief follow-up interview opt-in not fielded, this arm only

The S6 debrief page ends with an optional opt-in question about a paid follow-up interview, plus a debrief paragraph introducing it. That opt-in is a recruitment feature of the project's Prolific fielding and does not apply to a panel wave, so on this arm the opt-in question and its introducing paragraph are removed. The rest of the debrief text (study purpose, sponsor reveal, data-release disclosure) is unchanged. The item is optional and collects no study measure, so its removal touches no hypothesis, no exclusion rule, and no analysis.

B3.5 Welcome/consent page adapted for the panel platform, this arm only

The welcome page keeps the section 5.1 title, sponsor label, screener question, screener block behavior, and consent structure. Four wording adaptations, all made because the wave runs on Qualtrics through a managed panel rather than on the campaign-adjacent study site:

1. The subtitle states the arm-specific duration, "About 13 to 21 minutes" (this arm adds the two lead pages and removes most demographics, so its computed range differs from the website arms' "About 12 to 20 minutes").
2. The bullet "You can take a break at any point; your progress is saved between pages" is not shown. Qualtrics session resumption depends on provider-side settings this project does not control, so the instrument does not promise it.
3. The attention-check bullet reads "A few items are worded to check whether you are reading carefully. Please take your time." The sentence "Thoughtful responses are what your compensation is paid for." is dropped because compensation is administered by the panel provider under its own terms, and "each statement" is dropped with it.
4. The consent checkbox reads "I am at least 18 years old, I have read the above, and I agree to participate." (contractions expanded; substance identical to section 5.1).

None of these items is a study measure; no hypothesis, exclusion rule, or analysis touches the welcome wording.

B3.6 Everything else is unchanged

From welcome to debrief the instrument is the fielded S6 full arm: same Section B battery and randomization (plain shuffle, fixed slots B-bal at 1, B-att at 7, B-rev at 11, no cluster constraint), same ballot face, mechanism page, panel composition, revote, reflection, and debrief text (as amended in B3.4), same neutral sponsor label ("American Institutions Study 2026"), same B-att in-survey gate. The welcome subtitle uses the 2026-08-04.s6.2 wording (the "combined with several hundred others" aggregation phrasing removed; see the change log in B8).

B4. Wave-specific hypotheses

H1 through H4 are unchanged (principle agreement, ballot support, support after mechanism, mechanism preference), now estimated on the representative panel. New, specific to this wave:

• HV1 (rung-1 assent): a large majority of U.S. adults agree with the rung-1 statement. Primary estimand: share of rung1.agree in 5 to 7, Wilson 95% interval, headline-tier sample, with provider weights when the panel provides them. Reported alongside the pooled rung-1 share from the earlier website surveys as a bridge, with the population and context differences stated (paid pitch-page funnel on the campaign website vs representative research-survey panel).
• HV2 (trust does not gate support): ballot support (H2, share voting 5 to 7) is majority-level within each trust band, including the lowest. Bands, fixed in advance on trust.gov: low 1-2, middle 3-5, high 6-7. This is the confirmatory form; the full support-by-trust curve is exploratory.
• HV3 (rung-1 vs ballot gap, exploratory): the within-respondent gap between rung-1 assent and ballot support (who assents to the right but does not support the measure, and what their open-text reasons say) is exploratory description, not a test.

B5. Sampling and recruitment

• Panel wave (confirmatory): a representative or quota-balanced online panel, planned as Verasight, target N recorded at field time (planning estimate around 1,000), fielded to the hashed QSF. The provider's sampling targets, quotas, and any weighting are reported with the results, along with the provider-delivered profile variables used in place of the removed in-survey demographics (B3.3). Field dates and final N recorded at close.
• Operational smoke tests (excluded by rule): any fielding used solely to verify survey mechanics (completion code, embedded data, export) is excluded from all confirmatory and headline samples.
• The AAPOR-style disclosure block, eligibility rules (U.S. citizens 18+, in-survey citizenship hard block), neutral sponsor label, debrief, and fair-pay commitments carry over from section 4 unchanged.
• Nonprobability language discipline carries over: Wilson intervals are precision summaries; classical margin-of-error claims are made for no Prolific-fielded data; the panel provider's methodology statement covers the panel wave.

B6. Exclusions

Unchanged from section 7, on purpose: the two-tier rule set (headline tier excludes on every quality check except B-rev; confirmatory tier adds the B-rev acquiescence rule with the B-bal guard), the B-att in-survey gate with one retry, the citizenship hard block, and the completion-time checks.

One quality item does not exist on this arm: the end-of-survey recall check is not fielded (B3.3), so it drops out of the quality-check set that the two-tier rules operate on. Every other check applies as registered.

The lead pages add no exclusion item and no timing rule: two Likert taps are too short to carry a meaningful speed check. Dwell time on the lead pages is recorded for description only.

B7. Analysis plan additions

• rung1.agree: top-3-box share (5 to 7) with Wilson 95% interval; distribution and mean as description. Same coding convention as the ballot vote.
• trust.gov: distribution; ballot support (H2) crosstabbed by the three preregistered trust bands; party post-stratification per section 8, using the in-survey party item (kept on this arm; B3.3).
• Bridge table: rung-1 assent on this wave vs the pooled earlier website surveys (same wording, same scale), each with its own interval and its own frame disclosure; no pooling across the two frames.
• All existing S6 figures computed by the same published pipeline (scripts/rthg-reproduce.ts); s6-verasight rows enter via WAVE_CONFIGS['s6-verasight'].

B8. Design change log (all before any fielding)

The panel wave was designed after the s6/s6-short instrument was drafted; these are the design decisions that produced its final form, all made before any s6-verasight fielding and before this registration was frozen:

Any deviation after this registration freezes will be added to this log with its date and reason, and material deviations will be flagged in reporting.

B9. Ethics and participant protection

Carried over from section 4 unchanged: voluntary consented participation (18+, may stop at any time), sponsor identity withheld until debrief (topic disclosed at recruitment; every respondent receives a full debrief naming the Right to Honest Governance Institute and the publication plan), de-identified data only, anonymized quotes only, provider fair-pay floor respected, no PII required. The optional paid-interview opt-in is not offered on this arm (B3.4).