How it works

From raw logs to a reconciled day in about an hour.

You install a small anonymisation module, run it on one day of your orchestrator logs and provider statements, and hand over the anonymised package. Reconora translates every provider’s format into one model, reconciles it line by line and returns a result you decode locally. Measured on a real day: about an hour of machine time.

Machine time to the first result

63.1 min

  • Anonymise the logs40.3 min
  • Anonymise the statements2.9 min
  • Engine run17.3 min
  • Reconciliation0.1 min
Measured on one anonymised merchant day, ordinary laptop for anonymisation.

Four steps

  1. Install the module. One folder, Python 3.11+, standard library only. It works offline and sends nothing over the network. You receive its full source code — how the anonymiser works.
  2. Anonymise a day. 1.5 million log records took 40.3 minutes on an ordinary laptop; a full statement workbook took 2.9 minutes. The encoding table and secret are created at your site and stay there.
  3. Hand over the package. Anonymised files plus a verification report. Nothing personal, no provider names.
  4. Get the result. Reconciliation, discrepancies with reasons, claim candidates and blind spots — in codes you decode with your own table.

What happens inside

  1. Read in native format. Each provider’s callbacks, responses and statement lines are read as they are — own fields, statuses, units, time formats. Providers are recognised by the structure of their messages; nobody on the provider side has to change anything.
  2. Translate into one model. Every message becomes an event in one canonical model: the same status, amount, time and identifier, whatever the provider called them.
  3. Fold into transactions. Many events describe one payment. 790,948 events folded into 318,804 transactions in one day; repeated observations of the same fact are counted, not duplicated.
  4. Reconcile three sides. Your log, your notifications and the provider’s statement are matched line by line, using several keys — your ID, the order ID, the provider’s ID.
  5. Name the outcome. Every line ends in one of four outcomes, with a reason.
  1. 01Read as receivedCallbacks, responses and statement lines in each provider’s own format.1.5 million records
  2. 02Recognise the shapeThe provider is recognised by the structure of the message, not by its values.1,198 message shapes
  3. 03Extract the primitivesIDs, status, amount and its role, currency, event time, error, message type.8 primitives
  4. 04Translate into one modelThe same status, amount, time and identifier, whatever the provider called them.318 statuses → 9 meanings
  5. 05Fold into transactionsMany events describe one payment; repeated observations are counted once.790,948 events → 318,804 transactions
  6. 06Reconcile three sourcesYour log, the notifications and the statement, matched on several keys.35 statement formats
  7. 07Name the outcomeEvery payment ends in one outcome with a reason and its source rows.Four outcomes
Counts from one anonymised merchant day (full day); the eight primitives are the model’s concept; statement formats as configured.

Nothing is lost silently

Every record is accounted for: 1.5 million records = 1,187,215 parsed + 329,997 skipped by rule + 9,635 quarantined with an address and a reason. The balance is checked on every run.

Measured, not promised

Every figure below was measured on one anonymised merchant day; none is an estimate or a vendor average.

What was measuredResult
Log records anonymised on an ordinary laptop1.5 million in 40.3 minutes
A full statement workbook anonymised2.9 minutes
Engine run17.3 minutes
Reconciliation0.1 minutes
Machine time to the first full result63.1 minutes
Events folded into transactions790,948 into 318,804
Message shapes recognised1,198, from 96 providers
Raw status values translated318, into 9 meanings
Statement formats configured35
Records lost without a reasonNone: every record is parsed, skipped by rule or quarantined

How every number is produced

What we need — and what we don’t

We needWe don’t need
Orchestrator log export (CSV)Access to production systems
Provider statement workbook (xlsx)An API integration to start
A computer with Python 3.11+Any personal data
Changes to your finance team’s processes

Try it on one day of your data

One day of your data, reconciled, and the report stays with you whether or not you continue.

Updated