The readers are the people who hold a platform team to account: the fund's partners, its investment committee, its LPs. They have read enough pilot write-ups to discount most of them. A retrospective earns their trust the way a sound audit trail does, by letting every number be traced to something written before the number existed. For the internal review of a pilot whose companies are still moving, see what to do when a pilot has no completed transactions; this page is about the document that leaves the team.
Why investors discount most pilot write-ups
The same five faults recur: only the wins are described; the plan is written after the results; rates are quoted on denominators of six; companies still in review are counted as successes; and claims about SourceX's business, which the platform cannot see, are borrowed into a report on the platform's own work.
Research practice has a fix for the second fault, and it carries over. The Center for Open Science describes preregistration as specifying the research plan before the study and filing it with a registry, and asks that published results show which analyses were confirmatory and which exploratory, usually in separate sections; anything not specified in advance counts as exploratory (our paraphrase). A referral pilot is not a research study, but the habit of separating what was planned from what was noticed later is exactly what a skeptical LP looks for.
The methods-results-limitations outline
Ten parts, always in this order. Parts 1 to 3 are methods, parts 4 to 7 are results, and parts 8 to 10 are limitations and accountability. A part with nothing to report stays in the document with one line saying so, because a missing heading reads as a hidden one.
| Part | What it contains | What a skeptical reader checks | What fails it |
|---|---|---|---|
| 1. Question and dated plan | The question, hypothesis, unit, cohort rule, measures, stop rules and decision rule, with the date the plan was fixed and where it was filed | That the plan predates the first ask | A plan with no date, or one written after results |
| 2. Changes to the plan | Every deviation, with its date and reason | That changes are disclosed, not absorbed | A silent change of measure or cohort |
| 3. Cohort and screening | How companies were chosen, how many were considered and how many were excluded at screening, with reasons | Selection | Only the warmest relationships, unstated |
| 4. Company flow | Each company once, at its furthest dated stage, from screened to completed | That nothing dropped out of the count | Rows that vanish between stages |
| 5. Planned measures | Counts with denominators for the measures named in part 1 | That the headline is the measure planned | A headline measure chosen afterwards |
| 6. Patterns noticed later | Anything found after the data came in, labeled exploratory | That exploratory is not dressed as confirmed | Unplanned findings in the summary |
| 7. Negative results and effort | Declines in CEOs' words with their permission, misreadings, complaints, protocol incidents, and platform and founder hours | That the bad news is here at full size | An empty section |
| 8. Limitations | Sample size, selection, window length, open companies, what the design cannot show | Candor | Boilerplate |
| 9. What is withheld and why | Identities without consent, reward figures, SourceX's non-public deal details | That gaps are deliberate and explained | Gaps with no explanation |
| 10. Decision and next reads | The decision rule's output and the dates open companies will be read again | Accountability | "Promising" with no decision |
Proving the plan came first
A plan proves nothing unless someone outside the team could see it before the first ask. The cheapest proof is emailing it to a person not running the pilot, such as a partner or the fund's counsel; a registry like the one the Center for Open Science runs gives a stronger timestamp. Part 1 states the date and where the plan sits, and part 2 lists every later change with its reason.
The plan names what readers test the results against: a hypothesis with a number in it, the unit (one company named before outreach), the measures, the stop rules and a decision rule saying in advance what each result leads to. A pilot that skipped this still gets a retrospective, labeled observational in its title.
Every company at its last recorded stage
Clinical trial reporting has a standard answer to the vanishing-row problem. CONSORT 2025 pairs a 30-item checklist with a diagram of participant flow, and its item 22a asks for the numbers in each group who were randomly assigned, received the intended intervention and were analysed for the primary outcome (our paraphrase). Most referral pilots are not randomized, but the discipline transfers whole: every company that entered the cohort appears exactly once, at the furthest stage a dated record shows, and the open box is printed even when it holds every company.
The flow below is Illustrative, with invented counts that show the shape. Read across the rows and the arithmetic closes: twenty screened, seven excluded and two not asked leave eleven asked; four declines and one silence leave six who agreed; one duplicate leaves five verified applications; one decline in review leaves four open.
| Stage (Illustrative) | Count | Out of |
|---|---|---|
| Screened on metadata | 20 | Companies named before outreach |
| Excluded at screening | 7 | 20 screened: 4 on full-time headcount at peak, 2 with client-owned records, 1 outside the geography sentence |
| Relationship owner declined to ask | 2 | 13 that passed screening |
| CEO asked | 11 | 13 that passed screening |
| CEO declined | 4 | 11 asked: 3 informed, 1 misreading |
| No reply | 1 | 11 asked |
| CEO agreed and was introduced | 6 | 11 asked |
| Returned as a duplicate | 1 | 6 introduced |
| Verified application | 5 | 6 introduced |
| Declined in SourceX review | 1 | 5 verified applications |
| Open at the cutoff | 4 | 5 verified applications |
| Completed handoff | 0 | 5 verified applications |
Results on the planned measures, apart from what turned up later
Part 5 reports only what part 1 named. For a referral pilot that usually means qualified next steps (a CEO's written yes, and verified applications SourceX did not return as duplicates) and founder trust (whether the CEO described the offer back correctly, whether they would hear another introduction, and any complaint). Each count carries its denominator in the same sentence. Below the floor the plan set, counts replace rates, and no rate ever pools closed companies with open ones. The reasoning for that floor is laid out in how to avoid declaring a winner from five replies, and the counts-first table for an internal committee is in reporting a small pilot without exaggerating percentages. A document that leaves the fund adds one layer to that table: the withholding rules in part 9.
Part 6 holds everything noticed afterwards, such as "owner-run companies agreed more often". It is labeled exploratory and phrased as a question for the next pilot's plan. Money gets the same restraint: a reward that has not been paid is described as a formula and conditions, never as a figure, the approach set out in describing referral economics without forecasting revenue. Section 5 of the terms is the line to keep in view: "A lead, meeting or signed agreement alone does not earn a payable reward."
What a retrospective withholds, and says it withholds
Part 9 exists because some true facts are not the platform's to publish, and readers trust a stated gap more than a silent one. Four kinds of fact stay out:
- Company identities. Only with the company's agreement, given by its CEO.
- Reward figures tied to a company. Under section 4 of the terms the reward is the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap. Below the cap, anyone who sees a company's reward R can work out its payout P, because R is the greater of $50,000 plus 5% and 15% of P. A reward figure is therefore a disclosure of what SourceX paid that company. Report paid or not paid, or nothing.
- SourceX's non-public deal details. Section 4.5 of the sample Channel Partner Agreement reads: "Neither Party may publicly identify the other as a partner or disclose non-public deal details without written authorization, except as legally required." Section 4.6 says the partner tracks its referral through generic stages or reference codes. Generic stages can appear; anything more waits for SourceX's written authorization.
- Anything about SourceX's business. The platform cannot see SourceX's records, which is why duplicates surprise it. A pilot of a few companies is evidence about the network, not about SourceX.
The trace test before publishing
One last check turns the outline into a design with a pass mark. The hypothesis is that every number in the summary can be traced by a reader outside the team to the dated plan and the flow table. The unit is one number in the summary. Two readers who did not run the pilot, for example a partner and the fund's counsel, each mark every number traced or not traced, working separately. The stop rules are absolute: any number that cannot be traced, any company named without consent, or any reward figure tied to a company holds publication until it is fixed. The decision rule is simple. Publish when both readers trace every number, and where they disagree, the stricter reading wins.
The screening rule readers will ask about
LPs will ask how companies entered the cohort, so part 3 states the published fit facts the screen used. Each company's full-time headcount reached 50 or more at its peak, contractors excluded, and it held years of operating records of its own, with the rights to license them and a CEO aware of the idea. 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.
Part 3 also says screening ran on metadata alone, with no files, samples or logins requested, that client-owned material is usually not a fit, and that exclusions are dated because each SourceX program applies its own criteria. Where offices screened differently, give each its own row, as the region-by-region results comparison does.
Limits no single pilot retrospective escapes
- Small numbers. A cohort of twenty can show whether the process works; it cannot estimate a rate anyone should plan on.
- Warm-first selection. First cohorts draw on the strongest relationships, so later cohorts may do worse with nothing else changed.
- A window shorter than the process. Completion sits after stages whose timing SourceX does not publish, so most pilots end with open companies.
- Self-reported trust. Answers gathered by people who hope the pilot works; a second coder reduces the bias, not removes it.
- Programs move. SourceX's criteria follow its current programs, so a decline describes a program at a date.
- Selective publication. If only good pilots get written up, the record misleads even when each report is honest. Commit in the plan to publishing whatever the result.
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.
Before the next pilot's first ask, date its plan, register as a partner under the entity that will sign, check each named company against who qualifies, and read the program terms so the retrospective's part 9 is settled before there is anything to withhold. A midpoint review for advisory firms is in the six-month partner program retrospective, and more designs are in the partner guides.
A retrospective on a referral pilot tells LPs how the platform worked; it is not a view on SourceX as an investment. Readers weighing that separate question can request materials from SourceX's investor page, where the company explains that it gives investors company and commercial metrics directly.
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.