The point is interpretable learning even if no transaction completes. A paid outcome depends on SourceX's review, the company's agreements, the Acquisition Partner's acceptance and payment, all outside a platform team's control and on no published timetable, so an experiment that waits for paid deals cannot say when it will end.
What ten introductions can and cannot tell you
They can reject a large claim, find a broken step and test whether the team's screening judgment beats guessing. They cannot measure a rate precisely, compare two segments or say anything about payouts.
| Term | Meaning in this plan |
|---|---|
| Row | One company that enters the test, identified by a code |
| Ask | The relationship owner's request to the CEO for a conversation |
| Primary outcome | The CEO agrees to a conversation within 30 days of the ask |
| Cutoff | The date on which every row's status is read, fixed before the first ask |
| Pre-registration | The dated, unchangeable plan written before any ask is made |
The Center for Open Science describes pre-registration as specifying the research plan in advance of the study and submitting it to a registry, so planned tests can be told apart from exploration afterwards. A shared document with a locked, dated version history does the same job for a platform team.
How wide ten rows leave the answer. The interval below is the exact binomial interval, the method the NIST/SEMATECH handbook works through for a proportion. We computed the table with that method and checked it against the handbook's own example, 4 defectives in 20 units, which gives a 90% interval of about 0.071 to 0.400.
| CEO conversations out of 10 | Observed share | 95% exact interval |
|---|---|---|
| 0 | 0% | 0% to 30.8% |
| 1 | 10% | 0.3% to 44.5% |
| 2 | 20% | 2.5% to 55.6% |
| 3 | 30% | 6.7% to 65.2% |
| 4 | 40% | 12.2% to 73.8% |
| 5 | 50% | 18.7% to 81.3% |
| 6 | 60% | 26.2% to 87.8% |
| 7 | 70% | 34.8% to 93.3% |
| 8 | 80% | 44.4% to 97.5% |
| 9 | 90% | 55.5% to 99.7% |
| 10 | 100% | 69.2% to 100% |
Three of ten is consistent with anything from about one in fifteen to about two in three. Anyone who reports "30% conversion" from ten rows is reporting the middle of a very wide range.
The pre-registered ten-intro test plan
Fill every field, date the document and lock it before the first ask.
| Field | What to write before the first ask |
|---|---|
| Thesis | One sentence that could be false, for example: mapped companies with a willing relationship owner and four known fit facts usually produce a CEO conversation |
| Sampling frame | The full dated shortlist the ten are drawn from, with its row count |
| Selection rule | The top ten by a ranking written in advance, or ten drawn at random from rows meeting the entry criteria; every excluded row listed with its reason |
| Entry criteria | A relationship owner who agreed to ask; 50+ full-time employees at peak, operating history, rights and location believed to pass; the submitting partner has signed the SourceX Channel Partner Agreement |
| Consent | Relationship owners agree to be recorded by code; each CEO decides freely; no records, samples or logins requested |
| Primary outcome | CEO agrees to a conversation within 30 days of the ask |
| Secondary outcomes | Verified application submitted; review outcome by the cutoff; a reason code for every no |
| Forecasts | Before each ask, the growth lead writes the probability, 0% to 100%, that the CEO agrees |
| Comparison | Those forecasts scored against a flat 50% forecast |
| Decision rules | The three bands in the next section, copied word for word |
| Stop criteria | The four listed below |
| Cutoff and report | 90 days after the tenth ask; all ten rows published to the team, whatever happened |
| Changes | Any change after the first ask dated, explained and reported |
The 90-day cutoff is chosen to match the program's attribution window: the terms credit the first valid introduction within a 90-day window once the company submits a verified application, so the application stage has had its full window by the time the rows are read.
Decision rules a ten-row test can actually reach
Pick claims that ten rows can separate, and write down in advance what each count means.
- 0 or 1 conversation: reject "usually". If the true rate were 50%, the chance of seeing one or none out of ten is about 1.1%.
- 2 or 3 conversations: inconclusive. Both claims survive. Two or fewer of ten would happen about 5.5% of the time even at a true 50% rate, so two is not treated as a rejection.
- 4 or more conversations: reject "almost never". If the true rate were 10%, the chance of four or more out of ten is about 1.3%.
These are binomial probabilities for ten independent asks. Asks made through the same relationship owner are not fully independent, so the plan should cap each relationship owner at two rows.
Did screening judgment beat a coin flip?
The comparison that works at ten rows is forecast against outcome. Score each row as the square of the gap between the forecast and what happened, counting a conversation as 1 and anything else as 0, then average the ten scores. Lower is better. A flat 50% forecast scores 0.25 on every row whatever happens, so a team averaging above 0.25 forecast worse than no judgment at all.
Illustrative. Three invented rows to show the arithmetic, not results: forecasts of 80%, 60% and 20%; outcomes yes, no and no. Scores are 0.04, 0.36 and 0.04, an average of about 0.15, better than the flat 0.25. The 60% row is the lesson: the team was confident about a CEO who declined, and the reason code says why.
Declined, missing and unsuccessful rows stay in the denominator
| Code | Meaning | In the denominator |
|---|---|---|
| Yes | CEO agreed to a conversation | Yes |
| Declined | CEO said no | Yes, counted as no |
| No reply | No answer within 30 days | Yes, counted as no |
| Owner withdrew | The relationship owner decided not to ask after entry | Yes, counted as no and reported separately, because it is a finding about the path |
| Known to SourceX | SourceX already had a relationship with the company, so it is a duplicate under the terms | Yes, flagged; it tests how new your map really is |
| Not eligible | Review found the company outside current programs | Yes, for secondary outcomes |
| Open | No review outcome by the cutoff | Reported as open, never as success or failure |
Ten rows describe what happened to ten companies. They do not show that the network caused it: a CEO who would have applied after reading about SourceX elsewhere looks the same as one persuaded by the ask. Report counts, not causes.
Stop criteria
- Relationship harm. A relationship owner or CEO objects to how the ask was made: pause and rewrite the script before the next ask.
- Consent breach. Any company contacted without its CEO's knowledge: stop. SourceX's screening guidance says an introduction the decision maker was never told about is usually not a fit.
- Screening failure. Three rows in a row fail a fit fact the plan recorded as passing: stop, repair the screen and start again under a new dated plan.
- Terms change. If the program terms version changes mid-test, record the version on every row and continue only if attribution and eligibility are untouched.
A failure the plan should expect (Illustrative)
Illustrative. Six of ten CEOs agree to talk, which rejects "almost never" and looks like success. The reason codes then show that five of the six companies keep their most useful records inside their clients' accounts, and SourceX's screening guidance treats material that belongs to clients as usually not a fit. The thesis about conversations survives; the thesis about licensable supply fails. Without secondary outcomes and reason codes the team would have scaled the wrong thing.
Small samples, honest reports and the next step
Publish the full ten-row table, the forecasts, which rule fired, every change and what happens next. A two-of-ten result, written up, saves the next team a quarter; a silent one teaches nobody. If the result is inconclusive, run ten more under the same plan: twenty rows narrow the interval, though six conversations out of twenty still spans roughly 12% to 54%.
Limitations. Ten rows from one fund's network say nothing about another fund. A 90-day cutoff will leave most review outcomes open. SourceX applies the criteria of the specific program, so a company declined today may fit a later program, and if synthetic or public data covers more of what model developers need, demand for real records could narrow regardless of how the test turns out. Picking the top ten tests the best case, not the typical one. Forecast scores only teach if forecasts are honest.
The live first-month plan for venture platform teams is an execution calendar; this test is what tells you whether to keep running it. To see whether growth in the network adds rows worth testing, use the company-level cohort study, and for what happens after the application, the conversion benchmark. If one of the ten is a software company, its own testing archive may be a record type in its own right, as the page on experiment logs as training data explains. More in the guides hub and the overview for venture platform teams.
The entry criteria lean on four published fit facts: 50+ full-time employees at peak, not counting contractors; a documented track record of operations; the rights needed to license; and an authorized contact, who for this program is the CEO. 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.
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.
Once your plan is dated and locked, register for a partner link before the first ask, check each row against who qualifies and read the program terms so the attribution rule is written into the plan.
Evaluating SourceX as an investment is a different decision from running this test, and the SourceX investor overview is where to request investor materials for it.
This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.