A platform team's early-warning system for changing data needs: a weekly, monthly and quarterly review cadence that weighs each signal by its distance from a SourceX decision

Short answer

Data needs change in layers, and an early-warning system should weigh each layer by its distance from a SourceX decision. SourceX's own program pages, the terms version and the who-qualifies criteria, are authoritative and cheap to check weekly. Your introductions' outcomes are direct but sparse, so review them monthly. AI research and regulation move first but say nothing about any program, so read them quarterly as hypotheses. Change founder outreach only on the first two layers, and never infer who is buying from news.

Visual briefFull-size diagram
Three swimlanes over one quarter: weekly checks of SourceX's program pages, monthly reviews of introduction outcomes and one quarterly reading of research and regulation; only the weekly and monthly lanes connect to a box that changes founder outreach, while the quarterly lane feeds a hypothesis list.
Illustrative cadence using this page's method; dates of real documents are cited on the page, other signals are invented.

This is for a VC platform strategist who wants to notice when the companies worth introducing are shifting, before founders hear stale criteria from the fund. Everything said here about the program reflects SourceX's terms (Version 2026-10-22.2, since replaced by version 2026-11-04) and who-qualifies page as they stood on 26 October 2026. The cadence, its triggers and the illustrative signals are this page's method.

Four distances between a signal and a SourceX decision

A signal is useful in proportion to how close it sits to the decision it is meant to anticipate. The closest is SourceX's own published text. Its screening page says each program is judged on its own criteria rather than against one universal minimum, and that a company outside one program's criteria may suit a different program later; the page carries its own update date, 26 October 2026 when this was written. The terms keep a running record: section 11 lists each change since version 2026-10-12, version 2026-10-22.2 redefined Actual Supplier Payout, and the current version, 2026-11-04, set today's reward structure; the October change also redefined Actual Supplier Payout in the words of the agreement's Schedule A.

One step out are the outcomes of your own introductions: duplicates, declines and the reason where one is given, and what introduced CEOs tell you. Two steps out is AI research and developer documentation. Anthropic's engineering post on agent evaluation, dated 26 November 2025 on its index, treats a task as one test with specified inputs and success criteria, and advises running several trials because model outputs vary. Three steps out is regulation. The EU AI Act's Article 10(3) says "Training, validation and testing data sets shall be relevant, sufficiently representative, and to the best extent possible, free of errors and complete in view of the intended purpose", and EUR-Lex's consolidated text dated 27 July 2026 carries an amended Article 10.

Every signal then has to be translated into a company question a relationship owner could put to a CEO, such as "does the company keep a record of whether each job's outcome was checked, or only of the conversation about it?" A signal that cannot become such a question is commentary. Funding rounds, press about AI companies' data deals and conference claims sit outside the layers altogether: they establish no supplier eligibility and imply no buyer relationship.

DistanceLayerExampleWhat it can tell youWhat it cannot
0SourceX's program textTerms version history; the who-qualifies criteria and update dateWhat the program says todayWhy it changed, or what comes next
1Outcomes of your introductionsA duplicate decision; two declines for the same reasonHow the criteria are being applied to your companiesAnything about companies you did not introduce
2AI research and developer documentationAnthropic's guidance on evaluation tasks and trialsWhat some developers say they measureWhether any program wants a given company's records
3RegulationEU AI Act Article 10 data-quality criteria for high-risk AI systemsObligations some AI providers faceDemand, price or any program's criteria
OutsideNoiseFunding rounds, deal rumors, conference claimsNothing about mandatesAnything you should act on

The mandate signal review cadence

Each layer gets a cadence matched to how often it changes and how much it can be trusted. Anthropic's advice to run several trials because outputs vary applies here by analogy: one decline is one trial. That is why the monthly trigger needs a reason to appear twice before anyone changes a sentence in outreach, and why a quarterly reading produces hypotheses rather than instructions.

When a program change does land, re-screen only the rows it touches, using the criteria change impact register, and recall old wording with the versioned criteria distribution process. The cadence below decides when you look; those two pages decide what you do once you have seen something.

CadenceTime boxWhat you checkTriggerWhat you doOwner
Weekly10 minutesThe terms version string and the who-qualifies update date, against saved copiesEither one changedRead the changed section within two business days; hold outgoing criteria sentences until they matchPlatform analyst
Monthly45 minutesYour introductions' outcomes: duplicates, declines with any reason given, CEO feedbackThe same unexpected reason twice in a month, or a reason the published criteria do not explainAsk SourceX through the partner channel before changing outreach; tag the affected rowsPlatform lead
Quarterly90 minutesThree to five primary research or regulatory documentsNone automaticWrite each as one hypothesis and one CEO question; add them to a pilot backlog, never to outreachPlatform strategist
On a program changeWithin 10 business daysRows on your map that the change touchesAny change to criteria textRe-screen only those rows; leave the rest alonePlatform analyst
Yearly2 hoursWhich signal sources preceded real changesA source that preceded nothing all yearDemote or drop itPlatform strategist

One quarter of signals, sorted

Four rows below are Illustrative and marked so; the other four are real, dated documents cited on this page. The sorting shows the habit the cadence builds: most signals end in a log entry, a few in a question, and only the closest ones change what the fund tells founders.

DateSignalDistanceDecision
22 Oct 2026Terms version 2026-10-22.2, since replaced by version 2026-11-04 redefines Actual Supplier Payout0Re-read section 4 and update the reward sentence in founder templates
26 Oct 2026The who-qualifies page shows a new update date0Compare with the saved copy; re-screen only if a criterion's wording changed
September, IllustrativeTwo declines in one sector citing client-owned records1Ask SourceX whether anything changed; sharpen the "who created the records" question for that sector
September, IllustrativeA portfolio CEO says an AI company approached her for dataOutsideLog it confidentially; it implies nothing about SourceX's programs or about buyers
26 Nov 2025Anthropic's guidance on evaluation tasks, success criteria and trials2Hypothesis: records that capture a checked outcome may matter more; one CEO question; no outreach change
27 Jul 2026Consolidated AI Act text with an amended Article 103Watch item; it governs high-risk AI systems, and no program change follows from it
August, IllustrativeA portfolio company announces a Series BOutsideNo action; funding never establishes eligibility
October, IllustrativeA conference speaker says developers pay more for one record typeOutsideAsk for a primary source; log it as unconfirmed until one appears

Relationship and ownership evidence no signal replaces

A signal can point at a kind of record. It never points at a company's eligibility. When a hypothesis survives and a company seems to match it, the company still faces the same tests as any other: a headcount that peaked at 50 or more full-time employees (contractors do not count), years of records of its own making, the rights to license them, and an eligible location. 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.

Who decides stays fixed too. Opening a licensing discussion and signing for the company both rest with the CEO, so a signal-driven idea goes first to the relationship owner who can put it to that CEO. Every question stays at metadata level, with no files, samples or logins, and whether a company's records match current demand is checked in SourceX's review, not by your signal log. Funding plays no part in SourceX's screening, which is why a funding alert belongs in the noise row; the event suitability test for funding announcements turns such alerts into review tasks instead of emails.

A two-quarter test of whether the system warns early

The cadence is itself a thesis: that some signals arrive before program changes and can be read in time. Test it for two quarters before trusting it. Log every signal with its date and distance, then, for each program change you see, look back for earlier signals that pointed the same way. Pair it with a narrow introduction pilot: write one CEO question from each quarterly hypothesis, ask it of up to ten watch-list companies through their relationship owners, and count how many describe the records the hypothesis expects.

Write in advance what would change the cadence. If no research or regulatory signal preceded any program change across both quarters, cut the quarterly reading to twice a year. If your monthly outcomes caught a shift before the program page showed it, give that layer more time. If the ten-company pilot finds the expected records at two companies or fewer, retire the hypothesis rather than stretching it. Testing a whole category narrative, rather than one hypothesis, is covered in the referral experiment behind a category-creation story.

MeasureHow to compute itWhat would change the cadence
Lead timeDays between the earliest pointing signal and the program changeNo lead from distances 2 and 3 in two quarters: read them twice a year
False alarmsHypotheses written that no change or pilot supportedMore than three of four unsupported: tighten what enters the backlog
Outcome warningsProgram changes first visible in your monthly outcomesOne or more: lengthen the monthly review
Pilot matchWatch-list companies, of up to ten, describing the expected recordsTwo or fewer: retire the hypothesis
Time spentHours a quarter across all cadencesOver eight: drop the least useful source

Where an early-warning system fails

Method note. The layers and cadence rank signals by their distance from a SourceX decision, using SourceX's terms and screening page, Anthropic's evaluation guidance and the EUR-Lex consolidated AI Act text, all read on 26 October 2026. They measure when a signal appeared and whether it preceded a change. They do not measure demand for any company's records, prices, payouts or SourceX's pipeline, none of which is public; SourceX shares company and commercial metrics with investors directly. The quarter of signals is a teaching example, not a forecast.

  • Overreaction. A single vivid paper can pull a whole map toward one record type. The two-occurrence rule and the hypothesis backlog exist to slow that down.
  • Sparse outcomes. A fund with a handful of introductions a quarter will rarely see the same reason twice; its monthly layer may stay quiet even when something has changed.
  • Unshared reasons. SourceX may not give a reason for every outcome, so a quiet monthly review is not evidence that nothing moved.
  • Regulatory scope. Article 10 of the AI Act concerns high-risk AI systems; reading it as a statement about all AI data demand stretches it past what it says.
  • Programs change by design. Criteria follow each program and can move quickly, while signed agreements are never silently replaced and keep their own rates and exceptions, so what governs an introduction already made can differ from today's page.
  • Confirmation bias. A team that wrote a thesis will read signals as supporting it. Record the signals that point the other way with the same care.
  • Synthetic substitutes. If synthetic data replaces real records in a domain, research signals pointing at that domain would mislead; treat any such finding as a reason to retire a hypothesis.

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 weekly check shows your criteria sentences match SourceX's current pages and a company's CEO wants the conversation, register for a partner link and sign before submitting, screen the company against who qualifies, and keep the program terms version you read on file. The portfolio-side overview is for venture platform teams, the record-event memo is the quarterly brief on network signals, and more essays are in the guides hub.

Watching the program for changes is referral work; judging SourceX as a possible investment is a separate task, and investor materials, including the company and commercial metrics SourceX shares with investors directly, can be requested through the SourceX investor overview.

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.

  1. Step 1Share your linkSend your personal link to a company you know.
  2. Step 2Company appliesThe company applies itself at /apply.
  3. Step 3Company signs and is paidThe company signs, delivers its data and is paid by SourceX under its agreement.
  4. Step 4You get your rewardthe greater of $50,000 plus 5% and 15% of the actual supplier payout, with no cap. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details.

Common questions

Can a research paper or a new regulation change which companies we introduce?

Only indirectly. It can suggest a question worth asking a CEO and a hypothesis worth testing, but it says nothing about any SourceX program. Change outreach when SourceX's own criteria change or your introductions' outcomes show a pattern, not when a paper appears.

A founder says an AI company asked her for data. Is that a mandate signal?

Not for SourceX's programs. Log it as a note on that company, keep it confidential and draw no conclusion about who buys what. It tells you about one approach to one company, nothing more.

How much time does the cadence take?

About ten minutes a week, forty-five minutes a month and ninety minutes a quarter, plus re-screening when a program change touches your list. If it takes much longer, the team is reading too many indirect signals.

Sources

  1. refer.sourcex.si/terms
  2. refer.sourcex.si/who-qualifies
  3. anthropic.com/engineering/demystifying-evals-for-ai-agents
  4. eur-lex.europa.eu/legal-content/EN/TXT/HTML
  5. sourcex.si/investors

Free resources

Know a company with valuable proprietary data?

Become a referral partner, get your link and introduce the company’s CEO (in an owner-run company, usually the owner). Referral partners can join from most countries. We can't work with partners in countries or regions under U.S. sanctions or on U.S. restricted-party lists. The companies you introduce should be based in the U.S., Canada and the EU, or in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia with at least 70% of their data in English.

Refer a company →

I own a business

Explore licensing your company's data directly to SourceX for AI training and evaluation. Start a short assessment; no uploads needed.

Start an assessment