Negative demonstrations evidence gaps: seven proofs a reviewer looks for, and the ones error archives usually lose

Short answer

Four gaps make negative demonstrations hard to assess: missing context (the integration log that shows the timeout is gone), missing permissions (nobody can say under whose authority the second request ran), versions that moved (the date the integration was fixed is unknown), and unrecorded final outcomes (the story stops at "canceled"). Each removes a different piece of evidence a reviewer needs. None rules a company out; together they decide how many failure cases are usable.

Visual briefFull-size diagram
Seven horizontal bars, one per piece of evidence for a failed retry, each showing the expected length as an outline and the typically held share as a filled bar; the timeout signal, actor permission, change date and final cost rows show large missing segments outlined in orange.
Illustrative pattern, not measured data: which evidence for a failed retry error archives tend to keep and which they tend to lose.

This gap analysis belongs to the guide to negative demonstrations in an ordinary workflow, where an order integration sent the same approved requisition twice and the supplier received both PO-55120 and PO-55121 for one $18,400 requisition. That case is the reference below. The question here is narrower: what would a reviewer need to prove the second send was the wrong choice, and what is usually missing when they look?

The seven pieces of evidence a review expects

A failed action is useful as a negative demonstration only if a reviewer can show what was intended, what was done, what the actor knew, whether the action was allowed, which version of the process ran, what it cost, and why it was wrong. For the duplicate order, that means seven items.

  1. Intended state. The approved requisition: one order for $18,400.
  2. Action taken. The second create request, with its time and payload reference.
  3. Signal at the time. The lost response on the first request, which left the outcome unknown.
  4. Actor and permission. Which account sent the repeat and whether it was allowed to create orders unchecked.
  5. Process version. Which integration logic ran, and the date it was later changed.
  6. Consequence and final cost. Two orders sent, $36,800 committed, and what happened after cancellation.
  7. Written reason. A ticket or note saying the retry should have checked for an existing order.

Item 7 can be checked against published rules, which is what makes this failure type testable. The HTTP standard (RFC 9110) and Stripe's API documentation both warn against repeating a request that is not idempotent when its outcome is unknown, and NetSuite recommends external IDs to prevent duplicate records. With items 1 to 6 in hand, a reviewer can apply those rules to each case. Without them, the reason in item 7 is an assertion nobody can test.

Expected versus held: the gap table

Error archives tend to keep what lives in the system of record and lose what lives around it. The pattern below is drawn from how these systems typically store data; it is not a measurement of any company.

Evidence itemUsually held?Where it livesWhat goes missingEffect on the case
Intended stateUsuallyProcurement tool or approved requisitionRarely lost unless the requisition is overwrittenWithout it, nobody knows what correct looked like
Action takenUsuallyERP order record with creation timeThe link to the request that caused itThe duplicate is visible, its cause is not
Signal at the timeOften missingIntegration run logLogs roll over after weeksThe retry looks deliberate rather than blind
Actor and permissionOften missingService account settings, access reviewsShared accounts hide who actedCannot judge whether the action should have been allowed
Process versionOften missingChange tickets, deployment notesNo dated record of the fixOlder and newer failures get mixed together
Consequence and final costPartly heldSupplier statement, cancellation, invoiceRestocking charges, shipped goods, double invoicesThe harm stops at "canceled"
Written reasonPartly heldTicket linked to the reversalFixes made by phone, no noteTurns a labeled failure back into a bare error

The first two rows are rarely the problem. Almost every company that reverses orders keeps the order and the requisition. The middle rows are where archives thin out, because they sit in technical systems nobody thinks of as business records.

Four gaps and what each does to the usable set

Missing context. If the integration log is gone, PO-55121 looks like a buyer who changed their mind. The timeout at 14:02 is the fact that explains the duplicate, and it lives in exactly the records most likely to be purged. Our page on integration error records covers where those logs tend to live. Cases without the signal may still be usable as examples of reversals, but not of a blind retry.

Missing permissions. The record shows the retry ran, but not under whose authority. If the integration used a shared service account, nobody can say whether it was allowed to create orders without a check. That matters when the lesson is about which action should have been taken, because "the system was allowed to do this" and "the system should never have done this" teach different things.

Versions that moved. On day 10 the integration was changed to look up existing orders first. Cases before and after that change behave differently, so a reviewer needs the change date. Without it, a duplicate from the old logic and a clean run from the new logic sit side by side with no way to explain the difference, and the reasons attached to each become unreliable.

Unrecorded final outcomes. The ticket says PO-55121 was canceled. It may not say whether goods shipped, whether a restocking charge applied or whether the supplier invoiced twice. A negative demonstration is strongest when the cost of the wrong action is visible, and weakest when the record ends at a status change. Invoice holds tell a related story; see the guide on purchase order mismatch resolution.

How to describe the gaps in an introduction

Your introduction note should give SourceX an accurate starting picture. Name each gap, how far back it reaches, and who told you. Illustrative wording, all fictional:

GapNote that helpsNote that misleads
Missing context"Integration logs kept 90 days; tickets back to 2021""We log everything"
Missing permissions"Integration runs under one shared service account""Access is controlled"
Versions that moved"IT owner thinks the duplicate check was added in 2024; no change ticket confirmed""The integration is stable"
Unrecorded outcomes"Cancellations have reasons; restocking charges are in AP, not linked""Finance can see the impact"

"Unsure" and "not confirmed" are acceptable answers. A note that admits a gap saves a review cycle; one that papers over it costs credibility when the records are opened.

What only SourceX's review can measure

Some gaps can only be sized by looking at records, which partners never do. How many reversals carry a full chain from intended state to written reason, whether logs survived an ERP migration, how supplier names and employee identifiers would be de-identified, and whether integration vendor contracts limit use are all settled in SourceX's review under an agreement with the company. So is buyer interest. Appen lists failure-mode annotation among its agent data services, and there is public evidence that AI companies pay for data, but neither says anything about one company's error archive. The broader case for failure records is in failure data for AI training.

Procurement, finance operations and IT usually share these records, and the CEO authorizes a licensing discussion. The company needs 50 or more full-time employees at peak (contractors excluded), documented operating history, rights to license, 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.

To rate a company before describing its gaps, use the negative demonstrations qualification scorecard. After an introduction, the company assembles the negative demonstrations review packet.

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 the gaps you found shrink the usable set rather than empty it, register as a partner, confirm the basics on who qualifies, and read the terms before you introduce the CFO with your gap notes attached.

  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

Which gap is hardest to fix?

A missing timeout or error signal. Once integration logs roll over, nobody can show what the system knew when it retried, and a duplicate order then looks like a buyer changing their mind. Tickets and cancellations can sometimes fill other gaps; deleted logs cannot be recovered.

Should the company reconstruct missing reasons for old reversals?

No. Reasons written years later are opinions, not records of what people knew at the time. Describe the gap and the date from which reasons were recorded instead.

Does a gap mean the company should not be introduced?

Usually not. Gaps change how many cases are usable and what they can show. Only a failed eligibility gate, such as headcount, rights or authority, stops an introduction outright.

Sources

  1. rfc-editor.org/rfc/rfc9110.html
  2. docs.stripe.com/api/idempotent_requests
  3. docs.oracle.com/en/cloud/saas/netsuite/ns-online-help/sectio…
  4. appen.com/agentic-ai
  5. refer.sourcex.si/terms

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