Starting-state fidelity evidence gaps: five records that blur the moment a task began

Short answer

Five gaps blur the moment an order task began: fields overwritten without a change log, shipment events with a date but no time, change requests that lived in an inbox instead of on the order, approvals with no record of who could grant them, and returns split from the original order line. Each one hides part of the starting state, so a correct response and a wrong one become hard to tell apart. Timestamps are the gap to ask about first. None rules a company out; each narrows the share of cases that can be rebuilt.

Visual briefFull-size diagram
Five rows of ten blocks each show expected versus kept evidence for a fictional electrical wholesaler: field change history 3 of 10, timed shipment events 8 of 10, request received time 2 of 10, approval records 5 of 10 and return linkage 4 of 10, with missing blocks outlined in orange.
Illustrative gap profile for a fictional 290-person electrical wholesaler. Shares are invented for this page.

This analysis is part of starting-state fidelity, which rebuilds an order at 10:42, the minute a customer asked to cut gloves from 200 cartons to 120 after 150 had already shipped. The job here is to compare what a review expects with what distributors usually keep, and to give honest wording for the difference.

What a rebuild of the 10:42 state requires

A reviewer must reconstruct the order as it stood at the request's received time, then judge the response against that state. Every row below supplies one piece of the reconstruction.

EvidencePiece of the starting stateWhy it matters
Field change history with old valuesThe quantities before anyone edited themShows the order said 200, not 150, when the request came
Timed pick and shipment eventsWhat had left the warehouse, and whenPlaces the 08:15 shipment before 10:42
Request received time, on the orderWhen the task beganFixes the moment to rebuild
Approval records with roleWho was allowed to offer whatShows whether a return offer was within authority
Returns linked to the original lineHow the case endedGives the response a checkable outcome

AI developers treat this as basic hygiene for their own simulations. Toloka's agent data page (checked 2026-10-18) lists versioned environments, deterministic resets and audit logs for reproducing agent runs exactly, and τ-bench keeps a hidden database state that changes only through tool calls. A business archive that can supply the same five pieces is offering a starting state real enough to test against.

Expected against usually kept

The pattern below reflects common distribution practice, not measured data.

GapHow it appears in the order caseUsually keptUsually missingMetadata question
Overwritten fieldsLine edited from 200 to 150 with no logCurrent quantityOld value, editor and edit time"Does the order system log field changes with user and time?"
Untimed eventsShipment 1 dated, not timedShipment dateTime of day"Are pick and shipment events stored with timestamps?"
Context outside the systemRequest lived in a buyer's inboxThe edited orderThe request and its received time"Are change requests attached to the order with their received time?"
Unrecorded authorityReturn offered with no reason codeThe return authorizationWho approved it, under what limit"Do returns and credits record who approved them?"
Open outcome30 cartons returned a month later under a new numberTwo separate recordsThe link between them"Are returns linked to the original order line?"

Why the overwritten-field gap is so damaging: order systems restrict edits by state. Shopify's help center, for example, says merchants can edit unfulfilled items but cannot remove fulfilled items or change their quantity. The right response depends on that state at 10:42, and an overwritten quantity erases it. Change logging is often a setting rather than a default; Microsoft Learn describes Business Central's change log as recording old and new values for the tables and fields a company chooses and activates.

Illustrative gap profile: a 290-person electrical wholesaler

Fictional company, invented shares. Each figure is how many of every ten change requests from the last three years have that evidence, as an operations lead might estimate.

EvidenceKept, out of 10What is missingWhat it hides
Field change history3Order-line logging switched on only last yearOriginal quantities on older orders
Timed shipment events8Will-call pickups recorded by dateWhether a pickup came before the request
Request received time2Requests handled by phone or personal emailWhen most tasks began
Approval records5Credits approved verbally by branch managersWhether offers were within authority
Return linkage4Returns booked under new numbersHow the case ended

The warehouse system does its part: 8 in 10 shipments are timed. The order desk is weaker, and the request row is the binding constraint. With 2 in 10 requests timed, most cases have a known end but no known start, so the usable set is the recent fifth where request, shipments and edits all carry times.

How each gap distorts the case

  • An overwritten field makes the outcome look like the request. A line reading 150 suggests the customer asked for 150.
  • An untimed event makes the order of events a guess. A shipment dated the same day as the request could fall on either side of it.
  • A request outside the system removes the start of the task entirely.
  • Unrecorded authority keeps the action but hides whether it was allowed, so a correct offer and an overreach look alike.
  • An open outcome leaves the response ungraded, because nobody can see whether the 30 returned cartons closed it.

Real partial-shipment cases with these details intact are hard to invent, which is part of the argument in synthetic versus real data for AI agents, and of why buyers want licensed data. Adjacent fulfillment checks are in the guide on fulfillment records with documented operational decisions.

Wording the gap in an introduction

One sentence per gap, naming the window that is complete:

  • "Shipments are timed for three years; will-call pickups are dated only."
  • "Order-line edit history has been logged since last year."
  • "Change requests are attached to orders only when they arrive through the portal."

Avoid "the ERP has everything" (overwritten fields usually say otherwise), "we can rebuild any order" (untested) and "we have backups" (a nightly snapshot does not show the state at 10:42). Stay with metadata: systems, years, counts, owners.

Eligibility first: 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. The starting-state fidelity scorecard turns these answers into a fit rating, and the starting-state fidelity review packet shows what the company prepares next. More record families are under data types.

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 operations lead can name a window where requests, shipments and order edits all carry times, register as a partner, check the company against who qualifies and read the terms, then introduce the decision-maker with that window and the gaps stated plainly.

  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 rules a company out?

None on its own. Missing timestamps rule out same-day change requests, but requests that arrive days after a shipment can still be rebuilt from dates. The gaps change the share of usable cases, which SourceX measures.

Why ask about timestamps before anything else?

Because it is the fastest test. If shipments carry only a date, nobody can tell whether a morning shipment left before or after a mid-morning request, and the rest of the record cannot repair that.

Can the warehouse send one order history to show what is kept?

No. Ask which events carry times and where they are stored. Any look at real orders happens under SourceX's agreement with the company.

Sources

  1. help.shopify.com/en/manual/fulfillment/managing-orders/editin…
  2. toloka.ai/agent-data
  3. arxiv.org/abs/2406.12045
  4. learn.microsoft.com/en-us/dynamics365/business-central/across-lo…
  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