Section 08

The evidence

Twenty-three research documents, 218 findings, 1121 data points and 538 sources. The complete set is downloadable — including everything that contradicts the reading in section 07.

The analysis pack

All 26 research documents in full, the synthesis, the open-gaps register, and the extracted claim graph as machine-readable JSON. Anyone wanting to work with this, check it, or dispute part of it should start here rather than with the pages above.

Download the analysis pack (.zip)

How solid is this?

The honest way to judge a piece of analysis is not to read its conclusions but to count what they rest on. Every number in this work carries a label recording how it is known, and those labels are counted below rather than asserted.

The distinction that matters most is between a figure somebody published and a figure somebody estimated. Both look identical in a slide. Only one survives being checked.

Basis of the underlying data	Count	Share
Sourced to a named party	804	71.7%
— of which primary documents	177	15.8%
Measured directly	23	2.1%
Inferred, with the derivation shown	138	12.3%
Analyst, vendor or survey	36	3.2%
Estimated — judgement, no source	23	2.1%
No basis, or not published	97	8.7%

2% of the data behind these conclusions is judgement with no source behind it. That is the number to hold this work to, and it is the reason the labelling exists at all — an unlabelled body of research cannot tell you this about itself.

Two caveats on the count. It is generated automatically from the research documents, so a figure mislabelled inside a document is counted as labelled here. And a label is a claim about provenance, not about correctness: a number can be perfectly sourced to a party who was wrong.

How to read a research document

Every one follows the same structure. The question states what the document was commissioned to answer and what decision it informs. The bottom line is a short list of conclusions, each written to be argued with rather than agreed with. The key numbers table carries every figure with its basis, its confidence and the date it was true. Then the findings themselves, the variance and tail risk, what is still open, and finally an explicit statement of what would change the author's mind.

That last section is the most useful part of any of them. If you want to know whether a conclusion here is fragile, read what its author said would overturn it.

The documents

A Thesis

Document	Findings	Data points	External sources	Confidence
Is the revised position defensible, and for how long?	10	24	—	medium-high qualified
Who actually buys this, and why would they pay?	7	34	24	medium
Is the thesis strictly cross-border? Where the scope boundary actually sits	8	27	—	medium-high qualified

B Competition

Document	Findings	Data points	External sources	Confidence
Direct consumer/SMB-facing comparables — Augustus, Wise, Revolut	10	68	62	high
Cross-border infrastructure incumbents — own vs broker, economics, and whether "orchestration" is already their product	7	32	48	medium
Stablecoin-native payment infrastructure — competitors and counterparties	7	52	40	medium-high
Does the rail layer absorb orchestration from below?	8	66	31	medium-high
The card-orchestration precedent — how the last orchestration wave ended	7	34	40	high qualified

C Rail economics

Document	Findings	Data points	External sources	Confidence
Fiat correspondent banking / SWIFT leg — the incumbent baseline	8	88	26	medium-high
Instant & domestic scheme rails — operating economics and behaviour	7	37	35	medium-high
The FX leg — how spreads are formed, what they cost, and whether the FX book is the business	7	40	24	medium
Stablecoin leg — real all-in economics of on-ramp, chain hop, off-ramp	8	0	13	medium-high
Push-to-card payout rails — where they genuinely win, and whether we route over them	8	83	21	medium-high
EM local collection & payout rails — cost, timing, prefunding, licensing	9	65	16	medium
KYT, Travel Rule, sanctions screening and compliance holds — what they cost in money and in latency, per leg	10	78	23	high qualified

D Regulatory

Document	Findings	Data points	External sources	Confidence
EU licence perimeter — what authorisation does an orchestration layer actually need?	9	48	27	high qualified
MiCA in practice — what a stablecoin-touching orchestration layer actually has to hold	12	56	16	high qualified
Digital euro & the e-money question — how do we remain relevant in the EU?	6	42	25	medium-high
Corridor licensing — what it takes to be principal on the conversion in the corridors that matter	10	50	22	high qualified

E Build-vs-partner

Document	Findings	Data points	External sources	Confidence
Counterparty tier map — who sells unbundled single-leg wholesale, and at what tier	7	29	13	medium
Vertical integration scoring — per rail and per leg, own / partner / broker	10	47	—	medium
The value-add test — where does this business beat a sophisticated buyer going direct?	9	21	—	medium-high

F Corridors

Document	Findings	Data points	External sources	Confidence
Corridor economics — does the routing decision move all-in cost by >30 bps, repeatably?	6	39	2	medium-high qualified
South Africa (ZAR) and the CEE / Nordic cross-currency pocket — do either beat Mexico?	8	52	10	medium
The Gulf and China — two corridors the programme never analysed	10	9	13	high qualified

G Gap closure

Document	Findings	Data points	External sources	Confidence
Validating the comparison panel against published provider coverage	10	0	7	high qualified

A dash in the source column marks an analysis document — one that reasons over the other documents rather than gathering external sources of its own. “Qualified” means the author gave different confidence to different parts of the conclusion; the full wording is in the document.