Intercompany Reconciliation — working demo
A reconciliation platform for a multi-entity group: it takes the general ledger
and the delivery system, rebuilds every intercompany relationship, and shows
where the two sides of the group disagree — down to the journal entry.
Open the live demo →
All data on this site is synthetic. Every company, project, client,
person and amount was generated for this demo. The model is real: the
figures reconcile, the controls run, and every drill-down ties back to the
ledger line it came from.
What it does
Seven pages over six monthly closes, all anchored to the same cut:
| Page | What it answers |
|---|
| Overview | Where does the group stand this close, across all three domains |
| AR / AP | Which intercompany balances disagree, by pair, entity and document |
| Loans | Which loan pairs do not mirror, and which entity books only one side |
| Revenue → Cost | The full cycle in ten perspectives: recognition, invoicing, WIP, cost |
| Elimination | Whether the consolidation entity absorbed everything measurable |
| Controls | The control panel: what was measured, what was expected, what was found |
| Help | The vocabulary, so the numbers can be read without a hand-over |
The ideas behind it
- One source, many views. Every figure derives from the same ledger, so two
pages cannot contradict each other. The tie-outs hold by construction, not by
patching each view until it agrees.
- Facturing is not recognising. Revenue accrues into WIP; the invoice
releases WIP. Keeping those two clocks apart is what makes the cost mirror
and the elimination reconcile.
- Every drill reaches the evidence. A number you cannot open is a number you
cannot defend, so a cell opens the documents, and a document opens the real
journal legs behind it.
- A control that cannot fail is not a control. Each one publishes what it
expected, what it observed and how many items it looked at — because zero
findings over zero items is a pass for not looking.
- Anomalies on purpose. A reconciliation where everything agrees proves
nothing. This dataset carries invoices the buyer never booked, amounts booked
differently, timing differences, a cost recognised by journal with no supplier
invoice, and a management recharge one side booked in the wrong account family.
The report finds all of them.
The data behind it
The dataset is generated, not hand-written: a synthetic general ledger is
posted entry by entry, and all 19 datasets the report reads are derived from it.
That is why the tie-outs hold — they cannot do otherwise. The generator is in
generator/, with its own notes.
6 monthly closes · ~2,500 journal entries · 19 datasets · 41 controls
Running it
It is a static site — no server, no build step, no database.
python3 -m http.server 8000
Then open http://localhost:8000.
Built by Engels Alvarez · [email protected]
