Data quality At the source

A record corrected at creation costs a second.

The cheapest correction is the one made before the record exists in four systems.

WHERE RECORDS ARE BORNSAP or Oracle screenvendor · material · customerCRM formaccount · contactLima app formnew supplier requestAPI or filepartner onboarding feedBEFORE SAVEcolumn rulesformat · list · rangerow rulescombinations within the recordduplicate checksearch before createverificationregistry · address · bankTHREE OUTCOMESAcceptedborn golden · no ticketCorrected inlinethe fix shown to the person, accepted in the formHeld for reviewsaved in flow · steward decides · consumers waitA record corrected at creation costs a second; the same record corrected after it has spread to four systems costs four fixes and a reconciliationthe centralised style puts creation in Lima; the coexistence style runs the same rules through the ERP's own validation exits and the CRM's API
The same rules run at every entry point before a record is saved, so a golden record is born clean.
5
entry points covered
product
1
rule set, shared with the golden record
product
3
outcomes: accepted, corrected inline, held
product
4×
the cost of a fix after the record has spread to four systems
estimate
The entry points

Where records are born, and how the rules reach each

Entry pointHow the rules runWhat the person sees
SAP or Oracle create screensThrough the ERP's own validation exits, which call Lima's rule service with the screen's fieldsThe ERP's own message, with the fix proposed; the record saves clean or is held
CRM formsThrough the CRM's validation rules calling the same serviceAn inline correction (“GB” for “United Kingdom”) and a duplicate warning with the existing record
Lima app formsThe rules are part of the form; the duplicate search runs as the person typesThe form corrects as it goes; a held record shows who will review it and when
APIThe headless API's create call runs the rules and returns the outcome with the findingsThe calling system receives accepted, corrected or held, with the reasons
Partner filesThe file is read in place; each row runs the rules; the outcome is returned per rowA file report with the rows accepted, corrected and held, and why
The checks before save

Four checks, in order

Column rulesFormat, list and range on each field: a tax id in the country's format, a country from the list, a lead time in range.

The rules are the same objects the golden record uses, so a change to a rule applies at the entry points on the next save. There is a single rule set, and the entry points inherit the packs' thresholds and the four-eyes marks.

The order matters for the person at the screen: format problems are fixed in the field, row problems on the form, duplicates before anything is saved, and verification runs while the form is still open. A held record tells the person who will review it and when, so nothing disappears into a queue unexplained.

The three outcomes

Accepted, corrected inline, held for review

OutcomeWhenWhat happens to the recordWhat the consumer sees
AcceptedEvery applicable rule passes at saveBorn golden; write-back to the other owners runs at once, acknowledgedA golden record on the next read
Corrected inlineA rule fails and the corrector has a fix above the confidence floorThe fix is shown in the form and accepted by the person; the record is born golden with the correction in its ledgerA golden record, with the correction visible in the history
Held for reviewA rule fails with no fix, or the field carries the four-eyes markSaved in flow; the ticket goes to the escalation route; consumers wait or read the in-flow viewThe state 'in flow' with the pending checks, through the API's status field
Styles and the source

Where creation happens decides how the rules reach it

In the centralised style, records are created in Lima first, so the form is Lima's and the rules are simply part of it; the record is published to the ERP and the CRM once golden. In the coexistence style, records are created in the ERP or the CRM and the rules reach them through the validation exits and the APIs; the record is read in place at once and checked again as part of the golden loop.

In both styles the same rule set applies, and the result is the same: a record is born clean, or born in flow with its reasons. The implementation styles page explains how the style is chosen per domain.

The cost argument is the reason this page exists. A country typed as “United Kingdom” and corrected in the form costs a second. The same value saved, replicated to three other systems, read by planning and finance, and found a month later costs four corrections, four acknowledgements and a reconciliation of the reports that used it. The estimate of four times is conservative; the reconciliation is usually the larger part.

Questions

Does this slow down data entry?

Column and row rules answer in milliseconds; the duplicate search and the verification run while the person is still typing the rest of the form. On the showcase estate a supplier form with verification completes in under two seconds.

What if the ERP's validation exit is locked down?

The API path covers it: the create call runs the rules and the ERP receives an accepted record. Where neither is possible, the record is read in place on creation and checked within the minute, and the finding goes to the entry point's owner.

Can partners be held to the rules?

A partner file is read in place, run through the rules per row, and reported back per row. Partners see the same outcomes and reasons the internal entry points see.

See quality run on your own domain

Bring one domain and two systems. In a week the sources are profiled, the ten costliest fields are under rules, the first corrections are written back with their acknowledgements and the publish gate is on.