One end-to-end run · nine entities · zero answer key
No spreadsheet of expected answers. No pre-mapped chart of accounts. Just the messy paperwork a real practice actually receives — bank statements, dividend slips, trust deeds, a stale member map — and the instruction to prepare the group. Here is exactly what came out, figure by figure, with the workings shown.
Fully synthetic case. Every name, ABN, TFN, balance and address in "Callenbrook Group" is invented — a stress-test fixture built to mirror a real Australian family group. No real client data appears anywhere. Figures below are the numbers FediPilot derived from the source documents; the case ships with no answer key.
What we were preparing
A trust distributes to five beneficiaries and to a bucket company; an operating company charges rent to the fund and management fees to the trust; the SMSF leases a property back to the op-co. The money moves between the entities — so the figures only tie if you prepare them together.
Scroll the diagram sideways to see all nine entities →
The raw inputs · proof by paperwork
Twelve months of bank statements per entity, dividend and rental-agent statements, trust deeds, ASIC extracts, Div 7A loan agreements, PAYG summaries — plus the deliberate mess every real engagement carries. Nothing was pre-cleaned. Folders marked ! contain a trap the engine had to catch.








The part that matters most
We planted six traps a real practice would trip over. The test isn't whether FediPilot produces a number for each one — it's whether it produces the honest answer: raise the exception, hold the figure back, refuse the request. Every trap was caught.
The supplied map shows the trust's old individual-trustee arrangement and omits the SMSF's commercial property. The engine used the deeds and ASIC extracts as source of truth.
Ignored the stale map, flagged the conflictDaniel emails after 30 June asking to re-cut the trust distribution. Present entitlement crystallised on 30 June.
Refused — entitlement already fixedA $15,000 contribution was paid from Geoffrey's personal account, not the fund's — a characterisation question.
Excluded, not silently absorbedOnly 9 of 12 months of statements, a 9-month P&L, and three unexplained "sundry" deposits totalling $20,500.
Held out as NEEDS-REVIEW, not inventedInter-entity rent and management-fee charges appear on the invoices and bank statements but were never posted to the MYOB ledger.
Reconstructed from primary docsThe trust's CGT disposal reconciles two ways. The engine injected the computed $32,000, not the higher resolution figure.
Raised as a CRITICAL open itemEvery figure, derived · 9 of 9 entities rendered
Each entity ran the real workpaper pipeline end-to-end and rendered its own tax position from the source documents — five different entity shapes, one engine. These are the derived figures, not placeholders.
Cross-entity reconciliation · the group must tie
The trust's distributions have to land in the beneficiaries' returns and the bucket company's loan account. The engine reconciled each flow — and where a flow had no home, it said so rather than inventing an entity to balance the books.
| Flow | Amount | Result | Note |
|---|---|---|---|
| Trust → Thomas | $40,000.87 | PASS | Injected into Thomas's income_items; total rose accordingly. |
| Trust → Priya | $25,713.41 | PASS | Landed against her sole-trader return. |
| Trust → Marian | $32,000.00 | PASS | Discounted value used — not the higher $88k resolution figure. |
| UPE → Bucket company (Div 7A) | $55,546.39 | PASS | Recorded as loan principal; s109E minimum-repayment flag kept live. |
| Trust → Minor beneficiary | $400.00 | DOESN'T TIE | There is no individual return for the minor in this group. The reconciler flagged the gap rather than fabricate a return to make it balance. |
That last row is the point of the whole exercise. A system optimising for a clean-looking result would have manufactured a phantom return. FediPilot left the exception visible — which is exactly what an accountant needs to see.
What this proves
9 of 9 entities rendered through the real pipeline — five distinct shapes, one code path.
Every figure derived from 164 raw documents, with no expected-answers file anywhere in the case.
Group composition holds — cross-entity distributions injected and reconciled across the family.
Honesty under pressure — traps flagged, phantom entities refused, dubious figures held for review.
Cross-checked, not guessed — six deliberate traps in the paperwork, all caught and surfaced to the accountant.
Proven on the hard case — a nine-entity family group is the shape most tools can't handle at all.