# Financial audit report — 16 November 2026 (what we tested, what we found, what we learnt)

Prepared by the owner with Claude (AI engineer) and recorded for TravelClip. Scope: the whole money side of the platform —
sales, the ledger, the general ledger (books), VAT, agents, corporate clients, the owners' split, the screens that show money,
and the rules that stop losses. Method: the audit procedure in handbook 25 (risk-based, ISA-style), run on copies of the database.

## 1. Opinion (in plain words)
The accounting ENGINE is sound: every sale's money adds up, the double-entry books balance, cash ties to bookings, and every
money screen shows the same figures. The DATA in the books so far is test-period data (August–early November 2026) made before
margins and the gateway rate were set, so its margins and gateway costs are not representative — the go-live tool clears it.
No sale made since the money settings were completed (8 Nov 2026) can lose money or miss its gateway cost.

## 2. What was tested and the results
| Test | What it proves | Result |
|---|---|---|
| Books proof (53 identities, re-counted from raw tables) | Every rand accounted for: each sale's lines add back to the sale; debits = credits; Bank = cash position; Customers, Suppliers, Agents, VAT accounts tie; Gateway and Suspense nil; every receipt allocated to its invoice; no invoice over-received | All 53 hold |
| Self-test (522 checks) | Every money rule, guard, permission and document end to end (sales of every type, refunds, statements, VAT, splits, deals, corporate) | 522 / 522 |
| Security test (74 checks) | Sessions, two-step sign-in for money actions, signed document links, payments can't be faked | 74 / 74 |
| Ledger stress test (5 331 checks) | Random sales of every type and channel through the real engine, every combination of VAT, agents, card fees | 5 331 / 5 331 |
| Books stress test (1 314 checks) | Random refunds, chargebacks, cancellations, reinstatements, statements, supplier and owner payments over months | 1 314 / 1 314 |
| Cross-screen comparison (Sep, Oct, Nov) | Dashboard, Finance and Reports show the same sales and the same income | Identical to the cent (after the fix below) |
| Reconciliation Aug–Nov | 177 card payments all matched to bookings; 194 bookings; no unbalanced booking; no payment without a booking | Clean |
| Dashboard money checks | Margins set, gateway set, no loss-making packages, no orphan payments, no cancelled bookings holding money | No money issues (2 security reminders: two-step sign-in, Android settings link) |

## 3. Findings and what was done
1. **Dashboard "sales" used a different source from Finance** (FIXED 16 Nov). It counted every booking not cancelled, paid or not, by the date made, in UTC months. Finance and Reports count PAID sales from the ledger in South African months. The figures happened to match on current data but would drift. Now the dashboard reads paid sales from the ledger, SA time — one source.
2. **128 sales earned no margin; 2 lost money (R96.45 and R316.24 on flights of R10 753 and R18 332).** All in September–October, BEFORE any margin was set (margins first set 8 Nov). Books are immutable, so they stay as recorded; they are test data. Since then: margins exist for every sector, and a payment-time guard (`live loss check`) refuses any live sale that would lose money.
3. **No gateway cost on 190 sales.** All were made before the gateway rate was entered (0% until 8 Nov 19:06). The ledger books the gateway at the rate in force when the sale is made, once. Zero sales since then are missing their gateway line; the engine is proven by the stress tests.
4. **A fixed R250 flight fee lost money on big fares** (safe only up to an R8 264 fare at 2.9% + R2). The advisor flagged it; on 15 Nov the owner moved flights to 10% and added a 1% customer card fee. The guard would have refused those sales anyway — costing bookings, not money.
5. **Kestrel Mining (corporate) with a looking-after agent:** on a R5 000 flight the agent earns R130.43 and the platform keeps R73.93 (1.36%) — POOR for us. Action for the owner: renegotiate the service fee or the agent's share.
6. **Saving Fees & margins also wrote the monthly-fee split as a fixed 50%** (it was blank = "follow the live rule"). Same result today; it will no longer follow if the live split changes. Owner to decide.
7. **Handbooks quoted old figures** (R250 flights). Marked historic; TravelClip uses live data and the advisor for today's numbers.
8. **Reports show two "income" figures per month** — owners' shares of SALES only, and ALL income including monthly fees (the ex-VAT shares of R2 000 a month: R500 Bushveld + R1 500 Kestrel = R1 739.13 ex VAT). Both are labelled; "all income" equals the dashboard and Finance.
9. **The money position (Aug–Nov, test data):** bank R1 573 751.64 against R1 536 520.61 owed to suppliers (test suppliers never paid), R9 303.76 to agents, R12 210.86 to the other owner, VAT owed R4 552.56; "ours" R11 163.85. In real trading suppliers are paid as trips happen — watch "owed to suppliers" weekly.

## 4. How we investigated (the method, step by step — reuse it)
1. Ran the automated proofs first (books proof, self-test, stress tests). They test the ENGINE.
2. Asked "does every sale earn a margin?" — grouped every paid sale by channel and type; counted zero-margin, loss and no-gateway sales.
3. Dated the exceptions (by month and against the date settings were first entered) — this separated test-period data from live risk.
4. Checked whether each exception could still happen today (is there a guard? was it after the setting existed?).
5. Compared every money screen for the same months through each screen's own API, on a copy of the database.
6. Read the code behind each difference (dashboard query vs ledger) and fixed the source, not the symptom.
7. Re-ran every test after each fix; recorded findings for the owners and for TravelClip.

## 5. What we learnt (lessons for the future)
- **One source per figure**: a screen that computes money its own way will eventually disagree. Money figures come from the ledger.
- **Test data is not evidence about the business**: date every exception against when the rules existed before drawing conclusions.
- **Immutable books mean old mistakes stay visible** — that is correct; fix the rule going forward and explain the history.
- **Fixed fees are dangerous on variable ticket sizes**: test small, usual and BIG tickets; prefer % or a customer card fee for big fares.
- **Agents' shares on corporate clients can exceed our own take** — always look at who earns what on a sample sale before agreeing terms.
- **The guards stop losses but can cost sales** — set margins so the guard rarely has to refuse.
- **Keep the handbooks honest**: when a setting changes, examples go stale; live figures must win.
