Long-horizon task data: one procurement issue traced across 41 days and four systems

Short answer

Long-horizon task data is a record of work that runs through many dependent steps over days or weeks, where later steps only make sense in light of earlier ones. A procurement issue that moves from request to approval, shipment, inspection and settlement is a plain business example. For an introducer, the question is whether a company can link those steps across its systems for real cases, with outcomes, and whether someone can authorize a license discussion. Volume of transactions alone does not establish that.

Visual briefFull-size diagram
A 41-day timeline of one procurement case: request on day 0, a PO for 24 units, 16 accepted, 2 failed and returned on day 29, an invoice for 24, a quantity mismatch and payment for 22 on day 41, with an arc showing the payment depends on the day 29 return.
Illustrative: one replacement-parts order whose last step can only be checked against a step 12 days and three records earlier.

What does long-horizon task data mean in an ordinary business workflow?

It means the record of a whole case, not a single exchange: every step, the state it left behind, and how each later step depended on it. The test is whether the final step can be checked only by looking back several steps.

AI developers care because long chains are where agents still fall short. METR, an AI evaluation organization, measures capability as the length of tasks, in human expert time, that agents complete half the time, and reports that this length has been doubling roughly every seven months. On that measure, the longer the task, the less often agents finish it. Data vendors build tooling around whole attempts for the same reason: micro1 describes its flow product as letting reviewers inspect an agent's steps, tool calls and outputs and judge either the whole attempt or individual criteria. A whole attempt can only be judged if the record keeps the whole case. The live guide on long-horizon tasks and project records explains the general idea with a project example; this page works one operational case end to end and focuses on what links its steps.

Annotated case: replacement bearings for a packaging line

Illustrative. A 320-person food manufacturer orders 24 replacement conveyor bearings at $750 each, $18,000 in total. Every detail is fictional.

DayStepSystemDepends on
0Maintenance raises a work order and a purchase requisitionMaintenance system, ERPThe failed bearing recorded on the work order
2Plant manager and finance approve, since it exceeds the plant's $10,000 single-approver limitERP approval logThe requisition amount and the approval matrix
3Purchase order issued for 24 unitsERPThe approved requisition
9Supplier ships 16 units; 8 are backorderedSupplier portal, emailThe PO quantity
1216 units received and pass inspectionERP receipt, quality logThe shipment notice
29Remaining 8 received; 2 fail inspection and go back on a return authorizationERP receipt, quality log, return formThe backorder and the inspection standard
31Supplier invoices all 24 units, $18,000AP inbox, ERPThe PO, not the receipts
33Three-way match flags a quantity error: 24 invoiced, 22 acceptedERP match exceptionReceipts on days 12 and 29, and the return
41Paid for 22 units, $16,500; credit memo for 2 units, $1,500; replacements ordered on a new POERP payment, credit memoThe exception, the return form and the supplier's acceptance

The day 41 settlement is only checkable against day 29. A reviewer reading the invoice alone sees a supplier billing what was ordered; the return authorization on day 29 is what makes $16,500 the right payment. ERP matching does part of this automatically: Microsoft's documentation describes three-way matching as comparing invoiced quantity with the product receipt quantity and treating a difference as a matching error. What the match cannot do is explain why the quantities differ. That explanation sits in the quality log and the return form, often in different systems.

Evidence fields that hold the case together

A long-horizon case is not one record. It is a set of records joined by references. The fields below are what let a reviewer walk from day 0 to day 41 without guessing.

FieldWhat it linksIn the bearing case (Illustrative)
Work order numberThe original fault to the requisitionFailed bearing on the packaging line
Requisition and approval IDsThe request to the approval log and the approval matrix in forcePlant manager and finance, over the $10,000 limit
Purchase order number and lineEvery later receipt, invoice and creditPO for 24 units at $750
Receipt and inspection referencesEach delivery to its quality result16 accepted on day 12; 6 accepted and 2 rejected on day 29
Return authorization numberThe rejected units to the supplier credit2 units returned
Match exception IDThe invoice to the receipts it failed against24 invoiced, 22 accepted
Payment and credit memo referencesThe settlement to the exception that shaped it$16,500 paid, $1,500 credited
Timestamps on every recordThe order in which things happenedDay 0 through day 41

If any one join is missing, the case breaks into fragments. The invoice still exists and the return form still exists, but nothing says they belong to the same problem.

Why a case can break apart

Even a company with all these systems may not hold usable cases. Common reasons:

  • References typed by hand. A return form that names the PO in a free-text note can be joined only by fuzzy matching, and some links will be wrong.
  • Systems replaced mid-history. A move from one ERP to another often keeps open documents and drops closed ones, cutting cases that span the cutover.
  • Supplier-side steps held elsewhere. Shipment notices and backorder emails may sit in personal inboxes or a supplier portal the company cannot export.
  • Outcome never recorded. If the replacements on the new PO are not traced back to the original work order, nobody can say the fault was actually fixed.
  • Short retention. Quality logs and email are often purged sooner than finance records.

These checks change how many cases survive, not whether the idea applies. Only SourceX's review with the company can count them.

Does the company qualify, and who decides?

The introduced company needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license the records, and an authorized contact. 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.

Stay at metadata level: which systems hold each step (plain names such as SAP, a maintenance system or a quality database), how many years of cases, roughly how many purchase orders a year, whether returns and credits carry references, and who owns the records. Procurement records name suppliers and prices, so supplier contract terms may narrow what is licensable. Operations or finance leads usually know the systems; the CFO or owner decides on a licensing discussion. What stays unknown until SourceX reviews the opportunity: SourceX's current data needs, the rights review result, de-identification needs, the company's offer and any reward. Advisors working across a portfolio can fold this into a portfolio introduction plan. More record families are listed under data types.

Go deeper on long-horizon task data

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 client's operations lead confirms that receipts, returns and credits carry the original PO reference across several years, and can name who would authorize the discussion, register as a partner, confirm the basics on who qualifies, and read the program terms before you make the introduction.

  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

How long does a case need to run to count as long-horizon?

There is no fixed threshold. What matters is that later steps depend on earlier ones and that the records let a reviewer follow the dependency, whether the case takes five days or five months.

Is a large ERP export enough?

Usually not. Most of the value is in the joins: which receipt, return and credit belong to which request. A table of POs without the links to receipts, inspections and invoices shows volume, not long-horizon cases.

Do we need every system to be in English?

The 70% English condition applies to companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia. Companies in the U.S., Canada and the EU qualify without it.

Sources

  1. refer.sourcex.si/terms
  2. arxiv.org/abs/2503.14499
  3. learn.microsoft.com/en-us/dynamics365/finance/accounts-payable/t…
  4. micro1.ai/blog/introducing-flow

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