A tracker reading and a wording test answer different questions
| Part | How it runs | What it can support |
|---|---|---|
| Trust core | Identical items, order and scale every wave | A description of that wave's respondents, by route; never a cause |
| Disclosure module | Each respondent reads one message, its ending drawn at random from three | A causal comparison of wordings, for that message and those respondents |
A shift in the core between waves is an observation: different people answered, the news changed, or nothing moved beyond noise. A gap between wordings within one wave is an experimental result, because chance decided who read which ending. The report keeps the two kinds of confidence apart and publishes no single trust index, which would hide disagreement.
The fixed trust core, asked the same way every wave
The OECD's Guidelines on Measuring Trust separate trust in people, including people the respondent knows, from trust in institutions, and recommend a numerical 0-to-10 scale with verbal anchors (paraphrase). The core follows both choices: 0 is "not at all", 10 is "completely", and every item offers "don't know".
| # | Item (Illustrative wording) | Measures |
|---|---|---|
| T1 | "How much do you trust business advice from the adviser or investor you have worked with longest?" | A known person, any topic |
| T2 | "If that person suggested you talk to a firm that licenses business records for AI training, how much would you trust the suggestion?" | A known person, this topic |
| T3 | "And if a professional you have met once suggested it?" | A one-time contact, this topic |
| T4 | "How much do you trust companies that license business records for AI training, as a group?" | The category, never a named firm |
| T5 | "When a professional recommends a company, how much does it matter to you whether they are paid for it?" | The weight a CEO gives disclosure |
The order never changes and no core item names SourceX, so waves stay comparable; rotating questions follow the core, never sit inside it.
What should a research lead know about referral disclosure trust research?
Every message here discloses. The useful question is which wording CEOs understand, so the arms differ only in what the disclosure adds.
FTC staff guidance on endorsements says a disclosure should be hard to miss, easy for ordinary people to understand and clear about the nature of the connection, without needing every detail; vague labels are unlikely to explain it (paraphrase). In FTC-commissioned research reported in December 2017, modest changes to disclosure wording, placement, size and color appeared to make paid search results and native ads noticeably easier to recognize as ads (paraphrase of the FTC's release). That was advertising, not referrals. Whether a rule reaches a partner's own message is for their adviser; do I have to disclose a referral fee is a start.
Base message, identical in every arm (Illustrative): "We've worked together for six years, so I'll be brief. SourceX licenses business records directly from companies for AI training and evaluation; its published criteria are at refer.sourcex.si/who-qualifies. If you'd like an introduction, tell me; if not, a one-line no is fine."
| Arm | Disclosure ending (Illustrative) | Adds |
|---|---|---|
| A | "If your company ever licenses data to SourceX, SourceX may pay me a referral reward." | Who pays, on what event |
| B | Arm A, then: "SourceX pays that reward separately, so your company would receive the same amount with or without me." | The company's payout is untouched |
| C | Arm B, then: "I'm writing on my own account. I don't speak for SourceX, I can't commit it to anything, and only SourceX decides whether a company fits." | Nothing is vouched for |
Arm B restates section 4 of the terms, under which the reward "is paid separately by SourceX and is never deducted from the supplier's agreed compensation." Arm C restates section 7.1 of the Channel Partner Agreement: "Partner is not an agent of SourceX and may not commit SourceX to prices, licenses, payouts or other obligations." No arm states an amount, suggests a fit, calls the writer an approved partner or mentions an outcome; section 4.6 bars telling a company that an estimate, interest, a signature or an upload assures a purchase or reward. The endings are test conditions, not templates; real wording lives in FTC disclosure examples for referral posts and the member newsletter disclosure guide, and the five-factor message study varies five features where this module changes one sentence. Section 4.5 says: "Neither Party may publicly identify the other as a partner or disclose non-public deal details without written authorization, except as legally required." A network fielding the module gets that authorization before its questionnaire or report names SourceX.
Five comprehension probes and the misreading key
The message does not return, so the probes measure what one reading left behind. An open question comes first ("Does the writer gain anything if your company goes ahead, and from whom?"), coded by two people blind to the arm. Five closed probes follow, each with "not sure".
| Probe | Accurate | Misreading |
|---|---|---|
| P1. Who could pay the writer: SourceX, my company or nobody? | SourceX | My company, or nobody |
| P2. Would your company be paid less because the writer is involved? | No (B, C); no or "doesn't say" (A) | Yes |
| P3. Does the message say SourceX approved or recommended your company? | No | Yes |
| P4. Is the writer speaking for SourceX? | No (C); no or "doesn't say" (A, B) | Yes |
| P5. Who decides whether to ask for the introduction: the writer, SourceX, or me or our CEO? | Me or our CEO | The writer, or SourceX |
Accurate, misread, not sure and skipped stay separate. A misreading is an answer the message or SourceX's published pages contradict, even on a point the arm left unstated; "not sure" marks a CEO who knows where their understanding stops, and is never added to it. The CEO then rates the message from 0 to 10 and picks a next step: reply yes, ask a question, check SourceX independently, ignore it, or ask not to receive such messages. A closing screen shows all three endings and their sources and says nobody will follow up.
Who is invited, by which route, with what consent
Frame. CEOs, owner-CEOs included, of companies whose full-time headcount reached 50 or more at its peak, contractors excluded, located where SourceX's published sentence applies: "Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English." Respondents pick a size band and country group; anyone outside is screened out and counted. The questionnaire is English-only.
Two routes, never pooled. In the network route, an association, investor network or advisory firm sends the researcher's invitation through its own channels. Those CEOs already know someone who may be paid, so an independent business-leader sample with no partner tie is reported beside it where the sponsor buys one. No network passes on a member list, name or company detail, or sees an answer. The invitation names the sponsor, says the network may receive referral rewards from SourceX, implies no advantage for taking part and uses a plain survey link; under section 2 of the terms a link click alone does not decide credit, and a response is not an introduction.
Consent and limits. The first screen states the purpose, a target under ten minutes, that no company name, records or financials are wanted, group-only reporting, withdrawal until the data freeze, and any thank-you, unrelated to answers. No item asks whether the company would license data, no answer reaches SourceX's review or a partner, and nobody is contacted about licensing because of an answer. Nothing in the survey bears on any company's fit, which only SourceX's review assesses.
Pre-specified analysis, sample arithmetic and stop rules
Core. Per item and route: the distribution, median and "don't know" share. Waves are compared, with intervals, only if frame, routes, wording and order held; a shift inside the interval counts as no detectable change.
Module. By assigned arm, with CEOs who left after the message counted as missing and breakoff by arm reported. Two questions are fixed in advance: does B lower P2 misreadings against A, and does C lower P3 and P4 misreadings against B? Message trust is compared two-sided; Holm's correction covers the two primary comparisons; the rest is exploratory.
Sample arithmetic (Illustrative). To detect a drop from 30% misreading a probe in one arm to 15% in another, the standard two-proportion formula at 80% power and two-sided 5% needs about 121 completed answers per arm, 363 in all; at 2.5%, the strictest Holm step, about 147 per arm. The 30% is invented and a pilot sets real inputs; a network below the floor runs the core alone and pools the module.
Stop rules.
- A sentence an arm restates changes in the terms or agreement: stop, rewrite the arm, restart the wave.
- An open answer names a company or describes records: delete it and add a warning to the question.
- A network presses members or ties taking part to licensing: close its route and report why.
- Do-not-contact requests pass a threshold set before launch: pause and review the invitation.
- Arm results stay unread until the window closes, and the window never moves because of them.
Every invitee stays in the denominator
AAPOR's Standard Definitions sort cases into interviews, non-respondents, unknown eligibility and ineligible, offer several response-rate formulas depending on how partial and unknown cases are treated, and expect the formula named and a table of final dispositions published (paraphrase). Forwarded invitations leave many cases of unknown eligibility, so each network's send count is reported as an upper bound.
| Disposition, per wave | Network route | Independent route |
|---|---|---|
| Networks that agreed, and any that then sent nothing | Not fielded | Does not apply |
| Invitations sent, as reported | Not fielded | Not fielded |
| Declined consent or screened out | Not fielded | Not fielded |
| Left during the core, or after the message (by arm) | Not fielded | Not fielded |
| Completed | Not fielded | Not fielded |
| Asked not to be contacted again | Not fielded | Not fielded |
Any cell under 10 answers is shown as a count, never a percentage.
How results will be reported, and why this page has none
Results: not published. No wave has been fielded, and no number here is a finding. Each wave report will carry the sponsor and each network's tie to SourceX, field dates, the terms versions behind the arms, all wording verbatim, the disposition table, core distributions by route, every pre-specified comparison with its interval (nulls included), the codebook and dated departures from this plan.
If an arm cuts misreadings without lowering trust, a network may suggest that sentence to members; a published sentence that confuses readers goes to SourceX as a wording question. The survey fits the research blueprint for company data introductions; results from permissioned program records follow partner results research.
What this design cannot tell a partner
- Stated trust is not a decision. A CEO who rates a message 8 may still ignore it, and the survey never measures licensing.
- One message, one sender. A colder note or another writer may read differently.
- Recall, not study. A misreading may be memory, since the message is gone before the probes.
- Selection and drift. Network invitees may be warmer, and a higher later reading may reflect a different mix of people.
Until a wave reports, nothing changes for partners: disclose plainly, as SourceX's live pages describe, and leave the decision with the CEO. Monetizing your professional network covers the wider practice.
Next step
Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.
If a CEO you know would want the conversation once they understand who pays and who decides, register, sign the Channel Partner Agreement before submitting anything, see whether the company meets the published fit criteria and go through the terms yourself; further study designs sit in the guides library.