Skip to content

Import reconciliation rules

Import reconciliation compares preserved source facts with the intended Toolora reconstruction. It asks whether the same objects are findable, whether their relationships agree and whether values retain their meaning. Matching created/skipped counts cannot establish those properties.

The current Bygglet reconciliation registry contains 27 rules. The accepted ADR records an earlier rule count as history; the current registry and operating runbook own today’s inventory. No rule count by itself establishes that the current dataset passes.

Dimension Inspected rule families Evidence that row counts miss
Identity Articles, employees, projects, quote threads/versions, invoices and customers A source object can be missing while an unrelated duplicate preserves total count
Relationships Project/quote, invoice/project/quote/customer, quote/customer and project/customer Correct objects can be linked to the wrong company or version
Commercial evidence Sent-version lines, acceptance records and customer-answer state A sent label can hide absent lines or fabricated acceptance
Amounts Invoice amount, line amounts, importer interpretation and quote subtotals Zero quantity can still carry a nonzero source fixed price
Time and presentation Imported time, project title and version-specific descriptions A populated target can contain a label from the wrong source field
Household-work source Dossier recipients, protected identity presence and year-plan totals Source information and native tax/application authority remain different
Payment state Status/date/amount together A paid date or local flag alone does not reconcile the invoice

The tool compares each rule’s actual/expected ratio with a committed baseline. Its tolerance is 0.005 in ratio, or half a percentage point. A regression beyond that tolerance fails; an improvement can be recorded through the explicit baseline-write path. A new rule is identified separately.

A baseline is an accepted measurement floor, not proof of complete correctness. A run can avoid regression while still showing substantial historical gaps. Updating the baseline to hide a failure would remove the information the check is meant to preserve. Review the evidence and resolve the owning mapping or exception instead.

The operational contract requires before/after reconciliation for changes to imported data. It needs the authorized source corpus and target database and is not the synthetic CI suite. Historical runbook references to production identify that infrastructure path; the current environment contract still governs pre-launch work.

The field probe checks the names importers read against actual source records. It distinguishes a missing source key, populated source information that is not read, and a field that is genuinely empty in the sampled source. An empty real field is not permission to invent a value.

A populated name can also be semantically wrong. A supposed signer label without the timestamp/evidence that would accompany a signature must not be promoted into acceptance. Version-aware source interpretation matters as much as key spelling. Source identity and agreement authority own those meanings.

Amount reconciliation checks the shared row mapper’s interpretation, including source price when quantity/unit-price are zero. Payment updates compare the status, date and amount together; a total mismatch needs the owning re-import/reconciliation procedure before status can be advanced.

An export has 100 invoices. Toolora also has 100 rows, but one source ID is absent and a different source appears twice. Identity reconciliation must detect the defect. Even after fixing those IDs, linking an invoice to the wrong project remains a separate relationship failure.

For another synthetic row, quantity and unit price are zero while source price is 2,000. Importing zero can preserve both document and row counts while losing 2,000 of value. These examples explain the rules; they are not measured facts about the shared tenant.

The 27-rule count can be checked from the source AST without opening private records. Full source/database comparison requires a named scoped receipt. Synthetic parser/key tests do not prove corpus parity, and a green technical import job does not approve operational cutover.

This pass changes documentation only. It does not run a reconciliation against shared data, rewrite the baseline, repair historical records or copy private corpus material into the handbook.