Section 05
Where the evidence is weak
Nine things this work does not establish, and three figures in general circulation that do not survive checking. Both belong here — before either interpretation — because they change how much weight the conclusions can carry.
Why this section exists, and why it is here
Most analysis puts its caveats at the end, in smaller type, after the reader has already formed a view. That ordering is a choice about who the document is for. This one is placed before both interpretations so a reader can test either argument against the missing evidence before accepting it.
A gap here means a question we could not answer, recorded as unanswered rather than filled with a plausible estimate. Each one is scored by how much of the analysis leans on it, so that a reader can see which unknowns are load-bearing and which are merely untidy.
What is not established
Each of these was left open deliberately. In every case the alternative was to produce a number by judgement and present it alongside measured ones, which is how research quietly becomes opinion.
Scope of this count
The chart counts dependencies inside the original research claim graph. It does not rank the questions introduced by comparing the payments-company and platform-company interpretations. Those decision gaps are listed separately below because zero original findings can depend on a question the packet never asked.
| The question | Why it is open | Findings resting on it |
|---|---|---|
| End-to-end settlement time distribution on a live corridor | we decided not to measure it | 5 |
| Banking-access fragility of ramps and issuers | work is commissioned or under way | 1 |
| VoP non-match rate (EEA) | nobody publishes it | 1 |
| Per-PSP SEPA value limits | nobody publishes it | 2 |
| Real (not list) pricing from counterparties | only a counterparty conversation closes it | 3 |
| Exit comparables for a sub-scale orchestrator | nobody publishes it | — |
| Can weekend prefunding be structured through a TARGET account type excluded from the holding cap? | needs a legal opinion, not more research | 1 |
| No dispersion evidence above €10,000, which is where B2B lives | only a counterparty conversation closes it | 4 |
| The frontier firm's marginal cost per corridor, which sets the ceiling on what we can ever charge | nobody publishes it | 9 |
Three of these matter more than the rest, and they share a property: no amount of further desk research will close any of them.
Nobody will tell us their real prices. Every counterparty price in this work is a published list price or a secondary estimate. Real terms are negotiated and confidential. That caps the confidence of any cost model built from public inputs — including ours.
We never measured how long a payment actually takes. The end-to-end distribution on a live corridor — not the average, the tail — was deliberately not measured. That was a decision rather than an oversight, and it is an awkward one, because the payments-company interpretation makes a guarantee about timing central once it can be priced. That interpretation therefore rests on a risk distribution this work has not observed.
The evidence above business ticket sizes is thin and recent. Nearly all measured price dispersion was gathered at retail sizes. A later exercise did measure European pairs at €100,000 and €1m and found dispersion widens with size — but against a single cheap provider rather than a competitive field, which is a different thing entirely.
Questions that decide between the two interpretations
These are not caveats to one model. They determine which model Atlas should be building, so they should be read alongside the dependency-ranked gaps rather than hidden inside either conclusion.
| Decision gap | Evidence required | Which way it cuts |
|---|---|---|
| Who has a budget for Atlas? | 12–15 buyer conversations with payment files, fee statements, a named budget owner and commercial terms | Workflow and reconciliation budget supports the platform path; demand for a delivered-payment price supports the payments path |
| What is production-live in the banking platform? | An independent capability audit separating live, coded, gated, simulated and planned functions | The platform interpretation depends on an existing asset whose production boundary has not been evidenced |
| What operating role is available before authorisation? | Counsel on technical-provider, outsourcing, agency, funds-possession and principal-FX boundaries, including DORA obligations | Determines whether role-first software can launch cleanly and what the partnered payments path can retain |
| How much conversion spread survives the full stack? | Executable RFQs held over 90 days and reconciled against liquidity, tax, compliance, prefunding, failures and operations | A strong retained contribution margin supports principal payments; a thin one removes the reason to carry the balance sheet |
| Does the platform create recurring economics? | Paid design partners, implementation cost, gross retention, expansion, support load and evidence of switching friction | Without these, the platform path is plausible architecture rather than an evidenced business |
Figures that circulate in this market and should not be used
Three numbers appear repeatedly in industry material and do not survive checking.
A widely-quoted cost-and-speed baseline for business cross-border payments turns out to be consumer remittance pricing, misattributed — roughly four times the measured business figure. Anyone quoting it in a room with a payments professional will be corrected.
Every vendor claim of approval-rate uplift found during this work was unaudited, without exception. The category has been making the same claim for fifteen years without producing an independent measurement of it.
And arguments built on emerging-market currency-control spreads are three years stale. The gaps those arguments depend on have largely closed — in the largest case from tens of per cent to low single digits.