The tool here is the partner's own: a short question sheet, a spreadsheet tab or SourceX's company fit checker, used before a partner decides whether to raise SourceX with a CEO. A different question, whether a company that checks itself applies more selectively, is tested in the company fit checklist experiment.
What a fit tool may do, and what it may never touch
A fit tool turns the who-qualifies page into questions a partner answers from two places only: public information and what the CEO has said. Full-time headcount at the company's peak, contractors left out. Years of records the company created itself. Rights to license them. Where the company is, read against the published location sentence. Whether the CEO knows about the idea and wants the conversation.
| The tool may | The tool may never |
|---|---|
| Restate each published criterion as a question, in SourceX's words | Accept a file, a pasted export or a screenshot |
| Return one of three outputs: ask, ask first or not now | Return a score, a percentage or the word "qualifies" |
| Turn every unknown into a question for the CEO | Treat an unknown as a fail |
| Link to who-qualifies and the terms | Add a criterion SourceX does not publish, such as a guess about demand |
| Change the day SourceX changes its page | Keep an old version running |
"Ask" means every known answer passes. "Ask first" means at least one answer is unknown, and the unknown becomes a question for whoever holds the CEO relationship. "Not now" means a published criterion fails on a fact the company stated or that is public, and the log records which one and what would change it. Who-qualifies puts the rule for the middle case in one line: "Unknown is not a fail; it is a question for the relationship holder."
SourceX describes its own checker as "a preliminary, non-binding screening," and its page adds: "Nothing is saved unless you choose to, and no contact details are needed." Whatever a tool says, SourceX's review decides fit, including whether a company's records match a current program. Partners never collect, scan or score company records, as the no-upload screening form also shows.
The hypothesis and the count that decides
Hypothesis. Partners who run the tool before asking a CEO produce more qualified outcomes per ten companies considered than they would have without it, and set aside no company on an unknown or on a guess about demand.
Primary measure: qualified outcomes per ten companies considered.
- A company considered is any company a partner thinks of raising with its CEO, logged with the date at the moment it comes to mind and before any tool is run. Both halves follow the same logging rule.
- A qualified outcome is a considered company whose application SourceX verified and credited to that partner's introduction, still open on the read date. Copy SourceX's stage words exactly as the partner dashboard shows them; how it works lists the reward statuses as Estimated, Eligible, Approved and Paid.
Why companies considered, not CEOs asked? A tool that simply makes partners ask fewer CEOs would look brilliant per ask while losing fits. The considered log is written before the tool can influence it, so it stays fixed. The screening cohort analysis explains the same trap from the company's side. Why not clicks or tool sessions? The terms say an earlier link click alone does not decide credit, and that a lead, meeting or signed agreement alone earns no payable reward.
Baseline first, then a split by lot
| Element | Specification |
|---|---|
| Unit | A partner, because the tool changes how a person screens |
| Period 1, baseline | 90 days in which every partner works as usual and keeps the considered log |
| Split | Partners are paired by their period 1 volume, and one in each pair is drawn by lot to get the tool |
| Period 2 | 90 days in which the tool half runs the tool before every ask and the usual half carries on |
| Read dates | Each period is read 120 days after it ends, so the 90-day attribution window and some review time fit inside |
| Exclusions | Companies SourceX treated as already known, removed from both halves in both periods by the same rule |
| Estimate | Tool half's change from period 1 to period 2, minus the usual half's change |
The baseline is there because a partner team is small. With a dozen partners, the two halves can differ before anyone touches the tool, and subtracting each half's own baseline removes differences that stay constant. This is the difference-in-differences design. The World Bank's Impact Evaluation in Practice rests it on an equal trends assumption between the groups, and its DIME Wiki suggests comparing the groups' trends across several earlier periods and repeating the estimate with other comparisons (paraphrase). If your logs reach back another quarter, compare the halves on it before reading period 2.
A simple before-and-after on the whole team cannot do this job. Who-qualifies says requirements follow SourceX's current data needs and differ by program, so outcomes drift for reasons no tool caused. The usual half absorbs that drift.
The tool-assisted screening study log
One row per company considered. The study lead sets up the columns before period 1 and locks the half and period fields at the split.
| Field | Filled by | Rule |
|---|---|---|
| Company and legal name | Partner | Opened when the company first comes to mind |
| Date considered | Partner | Written before any tool run |
| Partner, half, period | Study lead | Never changed after the split |
| Tool output | Tool half only | Ask, ask first or not now, copied as shown |
| Partner's decision | Partner | Ask, hold or set aside |
| Reason for setting aside | Partner | The published criterion and where the fact came from, public or the CEO's words; "unknown" is never a reason |
| CEO asked and answer | Partner | Dates, and yes, later or no |
| Introduction submitted | Partner | Date |
| Stage at read date | Study lead | Copied word for word from the dashboard |
| Already known to SourceX | Study lead | As the dashboard shows it |
Columns that never exist: attachments, figures read from company documents, exports, screenshots, logins, and contact details for anyone at the company but the CEO.
The set-aside audit: fits the tool may have cost
Every company the tool half set aside in period 2, plus an equal random sample of the usual half's set-asides, goes to a second partner who is not told which half a row came from. The reviewer reads only the logged reason, against SourceX's published wording.
| Class | Example | How it counts |
|---|---|---|
| Published criterion, stated fact | The CEO said full-time headcount never reached 50 | Justified, under section 1 of the terms |
| Unknown treated as a fail | "Not sure what language the records are in" | Wrongful set-aside |
| Guess about demand | "SourceX probably has no use for freight records" | Wrongful; relevance is decided in SourceX's review |
| Relationship reason | The person who knows the CEO declined to ask | Not a fit question; left out of the audit |
The target is zero wrongful set-asides. A tool that raises qualified outcomes by teaching partners to drop every company with an open question has bought its gains with lost fits, and the comparison alone would never show it.
Side effects counted alongside
- The tool quoted as a verdict. Check one in five messages the tool half sends to CEOs for lines such as "the tool says you qualify." Target: none.
- Company material in the tool or the log. Target: none.
- CEO complaints, including a CEO who felt pre-judged.
- Asks per ten companies considered, read beside outcomes; fewer asks is fine only if outcomes hold.
- Minutes per company, so a team can weigh the tool's cost.
A partner who delegates screening to a colleague keeps the judgment calls; delegating screening without delegating judgment shows where those lines fall.
Floors, duration and stop rules
Floors. At least six partners in each half and 40 companies considered per half in each period before any per-ten rate is published. Below that, report counts and audit classes only.
Duration. Two 90-day periods, each read 120 days after it ends: the final read comes about ten months after the start.
Stop rules.
- A file, export or record is pasted into the tool or the log: stop the tool half, delete it, fix the tool and restart period 2.
- A partner tells a CEO the tool says the company qualifies or is approved: pause, correct it with that CEO and brief the partner again.
- Two of the first ten audited set-asides from the tool half are wrongful: stop and rewrite how the tool handles unknowns.
- SourceX changes who-qualifies or the terms on a criterion the tool restates: update the tool that day, and report outcomes before and after the change separately.
- A CEO complains of pressure: pause that partner's asks and review the message.
Illustrative arithmetic, not results. Invented rates to show the calculation; SourceX publishes no conversion rates.
| Half (Illustrative) | Period 1, per 10 considered | Period 2, per 10 considered | Change |
|---|---|---|---|
| Tool half | 1.0 | 1.6 | +0.6 |
| Usual half | 1.0 | 1.2 | +0.2 |
| Tool half minus usual half | 0.0 | 0.4 | +0.4 |
The last row is the difference in differences. The usual half's 0.2 is drift no tool caused, so only the extra 0.4 is credited to the tool, and only if the audit found no wrongful set-asides. With 40 companies considered per half, 0.4 per ten is under two companies, well within what chance moves. A gap that size would justify continuing the study, not rolling the tool out.
Reading the comparison and the audit together
| Comparison | Audit | Reading | Action |
|---|---|---|---|
| Tool half gains more | No wrongful set-asides | The tool helps | Give it to everyone and keep auditing a sample |
| Tool half gains more | Wrongful set-asides found | Gains bought with lost fits | Fix the unknown handling and run period 2 again |
| No difference | No wrongful set-asides | Harmless, not helpful | Keep it only if it saves time, and claim nothing about fit |
| Tool half does worse | Any result | It discourages good asks | Withdraw it |
| Below the floors | Any result | No comparison | Report counts and audit classes only |
Where a partner-level comparison misleads
- A few partners carry the result. One productive partner can swing a half. Pairing by baseline volume helps but does not cure it.
- Spillover. Partners talk, and the usual half can absorb the tool's questions. Ask them not to share it, and note any leak.
- Logging drift. The tool may prompt its users to log more companies, inflating their denominator. Compare each partner's considered counts across periods.
- Being watched. The tool half knows it is being studied and may screen more carefully for that reason alone.
- The tool is not the review. Even a perfect restatement of published criteria cannot see demand, rights or record quality; those stay with SourceX.
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.
When the tool returns "ask" and the CEO wants the conversation, register as a partner, compare the tool's questions with the live who-qualifies page and the current program terms, then submit.