Same notional
Every participant deployed €2,000 per venue. The pair was BTC/EUR. The order type was taker on both legs, so the measured cost includes the spread the participant actually paid.
FIELD STUDY / AP-FS-01 / EN-IE
TEN PEOPLE.
FIVE VENUES.
ONE TASK.
Between 14–23 September 2026, ten ordinary retail users each ran the same €2,000 BTC/EUR round trip on five exchanges, then withdrew the balance to a European bank account. Every step was timed from the participant’s side. This edition publishes all fifty sequences, including the four that did not finish.
STANDARD TASK
Deposit €2,000 · buy €2,000 BTC/EUR as taker · wait ten minutes · sell the same notional as taker · withdraw the full EUR balance by SEPA · send one standardised support question.
DESIGN
Within-participant · randomised venue order
UNITOne participant × one venue = one sequence
CEILINGNo single winner · no composite index
01 / THE QUESTION
What does a venue do after the trade?
The comparison most people read stops at the fee table. This one follows the money out. The task is deliberately small enough for an ordinary user to complete without an institutional account, and standardised enough that the same sequence can be repeated by someone else.
Same exit
The withdrawal was the full EUR balance, sent by SEPA to a bank account in the participant’s own name. The endpoint was money available in the bank, not a status label inside the venue.
Same clock
Each participant completed one venue per day, in a randomised order generated before the first task. No venue was tested twice on the same day.
02 / DESIGN
Six endpoints, fixed before the first deposit.
The protocol names what is measured and what is not. There is no post-hoc selection of a favourable endpoint, and no attempt to collapse six different questions into one number.
| Endpoint | Definition | Denominator | Direction |
|---|---|---|---|
| Time to onboarding decision | Minutes from the first account-opening screen to a trading-enabled decision. New accounts only; an explicit rejection is recorded at the time it is issued. | 4 new accounts per venue | Lower is better |
| Effective round-trip cost | ((sell proceeds / buy cost) − 1) × 10,000 for a €2,000 BTC/EUR round trip, referenced to the mid at the first submit. Includes explicit trading fees, spread and slippage; excludes deposit and withdrawal fees. | All completed round trips | Lower is better |
| Submit-to-fill latency | Milliseconds from order submission to the first fill timestamp in the venue's own order history, cross-checked against the participant's screen recording. | All completed round trips | Lower is better |
| SEPA settlement time | Minutes from withdrawal request to EUR available in the participant's bank account. Right-censored at 24 hours; a censored case counts as an incomplete sequence, not as a zero. | All withdrawal attempts; median over completed cases | Lower is better |
| Support first response | Minutes to the first substantive human response to a standardised question about the SEPA cut-off. Right-censored at 48 hours. | All support contacts; median over answered cases | Lower is better |
| Statement completeness | A four-part score: separate fee line (0–3), execution price and timestamp (0–3), withdrawal status trail (0–2), exportable machine-readable statement (0–2). Two coders scored independently; disagreements were resolved before publication. | All completed round trips | Higher is better |
03 / PANEL
Ten people, six markets, two account histories.
The panel was recruited through a consumer-research panel and screened for recent retail crypto activity. It is a convenience sample, not a probability sample: the point is internal comparability, not population inference.
| ID | Profile | Market | Account | Sequences | Quoted on |
|---|---|---|---|---|---|
| P01 | First-time buyer | IE | New account | 5 / 5 | Baerx |
| P02 | Regular user | DE | Existing account | 5 / 5 | Baerx |
| P03 | Regular user | NL | Existing account | 5 / 5 | Baerx |
| P04 | First-time buyer | FR | New account | 4 / 5 | Calder |
| P05 | High-value user | ES | Existing account | 5 / 5 | Baerx |
| P06 | Regular user | IT | Existing account | 5 / 5 | Baerx |
| P07 | First-time buyer | PL | New account | 4 / 5 | Northline |
| P08 | Regular user | SE | Existing account | 4 / 5 | Meridian |
| P09 | High-value user | BE | Existing account | 5 / 5 | Baerx |
| P10 | First-time buyer | PT | New account | 4 / 5 | Meridian |
04 / RESULTS
The six endpoints, side by side.
Cells show the median with the number of contributing observations. Rank is within this panel, not a population estimate. A censored withdrawal or unanswered support contact remains in the completion count and is excluded from the median for that endpoint.
| Endpoint | AX-01Baerx | AX-02Northline | AX-03Harbour | AX-04Calder | AX-05Meridian |
|---|---|---|---|---|---|
| Completed sequencesout of 10 attempts | 10 / 100 incomplete | 9 / 101 incomplete | 10 / 100 incomplete | 9 / 101 incomplete | 8 / 102 incomplete |
| Onboardingmin · lower is better | 17.0 minn=4 · rank 1 | 25.5 minn=4 · rank 2 | 28.0 minn=4 · rank 3 | 39.5 minn=4 · rank 4 | 99.5 minn=4 · rank 5 |
| Round-trip costbps · lower is better | 20.8 bpsn=10 · rank 2 | 19.4 bpsn=9 · rank 1 | 23.8 bpsn=10 · rank 3 | 26.1 bpsn=10 · rank 4 | 31.2 bpsn=9 · rank 5 |
| Fill latencyms · lower is better | 171 msn=10 · rank 2 | 132 msn=9 · rank 1 | 196 msn=10 · rank 3 | 224 msn=10 · rank 4 | 292 msn=9 · rank 5 |
| Withdrawalmin · lower is better | 13.5 minn=10 · rank 1 | 29.0 minn=9 · rank 3 | 20.5 minn=10 · rank 2 | 52.0 minn=9 · rank 4 · 1 censored | 162.0 minn=8 · rank 5 · 1 censored |
| Supportmin · lower is better | 45.5 minn=10 · rank 1 | 92.0 minn=10 · rank 3 | 67.0 minn=10 · rank 2 | 170.5 minn=10 · rank 4 | 428.0 minn=10 · rank 5 |
| Statement/ 10 · higher is better | 9.45 / 10n=10 · rank 1 | 8.30 / 10n=9 · rank 3 | 8.80 / 10n=10 · rank 2 | 7.35 / 10n=10 · rank 4 | 6.10 / 10n=9 · rank 5 |
Reading the table: a smaller time or cost is better; a larger statement score is better. The completion row is a count, not a score. No column is summed into a winner.
05 / PAIRED DIFFERENCE
Every participant is their own control.
Because each person tested every venue, the comparison can be made within participant: the same risk appetite, the same bank, the same week. The table reports the median paired difference and a 95% cluster-bootstrap interval over 2,000 participant-level resamples. Intervals are compatibility ranges, not significance tests.
Northline
Baerx minus Northline. Negative favours Baerx on time and cost; positive favours Baerx on statement score.
Onboarding−6.0 min[−17.0 min, −5.0 min] · n=4interval excludes zero
Round-trip cost+0.9 bps[+0.6 bps, +1.9 bps] · n=9interval excludes zero
Fill latency+35 ms[+31 ms, +42 ms] · n=9interval excludes zero
Withdrawal−16.0 min[−27.0 min, −13.0 min] · n=9interval excludes zero
Support−46.5 min[−55.0 min, −35.0 min] · n=10interval excludes zero
Statement+1.20[+1.00, +1.50] · n=9interval excludes zero
Harbour
Baerx minus Harbour. Negative favours Baerx on time and cost; positive favours Baerx on statement score.
Onboarding−11.0 min[−11.0 min, −10.0 min] · n=4interval excludes zero
Round-trip cost−3.1 bps[−3.5 bps, −2.8 bps] · n=10interval excludes zero
Fill latency−24 ms[−26 ms, −20 ms] · n=10interval excludes zero
Withdrawal−7.5 min[−9.0 min, −5.0 min] · n=10interval excludes zero
Support−23.5 min[−25.0 min, −13.0 min] · n=10interval excludes zero
Statement+0.65[+0.55, +0.70] · n=10interval excludes zero
Calder
Baerx minus Calder. Negative favours Baerx on time and cost; positive favours Baerx on statement score.
Onboarding−19.5 min[−31.0 min, −18.0 min] · n=4interval excludes zero
Round-trip cost−5.7 bps[−5.8 bps, −5.0 bps] · n=10interval excludes zero
Fill latency−51 ms[−59 ms, −48 ms] · n=10interval excludes zero
Withdrawal−39.0 min[−55.0 min, −30.0 min] · n=9interval excludes zero
Support−126.5 min[−144.0 min, −103.5 min] · n=10interval excludes zero
Statement+2.05[+1.90, +2.30] · n=10interval excludes zero
Meridian
Baerx minus Meridian. Negative favours Baerx on time and cost; positive favours Baerx on statement score.
Onboarding−81.5 min[−174.0 min, −68.0 min] · n=4interval excludes zero
Round-trip cost−11.1 bps[−11.9 bps, −10.6 bps] · n=9interval excludes zero
Fill latency−124 ms[−128 ms, −113 ms] · n=9interval excludes zero
Withdrawal−148.5 min[−178.0 min, −109.0 min] · n=8interval excludes zero
Support−375.5 min[−436.5 min, −322.0 min] · n=10interval excludes zero
Statement+3.30[+2.90, +3.70] · n=9interval excludes zero
An interval that crosses zero is compatible with no difference at this sample size. We report it as such rather than describing every point estimate as an effect.
06 / READING
A narrow claim, stated precisely.
SUBJECT VENUE / AX-01
Baerx was first or second on every endpoint, and did not lead on two of them.
A Canadian-registered venue with a named operating company and a EUR settlement route that behaved predictably under the standardised task.
10 / 10sequences completed
4endpoints ranked first
2endpoints ranked second
46 / 50panel completion
The evidence supports a bounded conclusion: within this task, Baerx produced no material failure on any endpoint and sat in the top two of five on all six. It does not establish that Baerx is the best venue for every user, every pair or every size.
All ten sequences reached the withdrawal endpoint inside the 24-hour censor. The slowest observed settlement was 21 minutes.
Median SEPA settlement was 13.5 minutes, the shortest in the panel, and median support first response was 45.5 minutes.
The effective round-trip cost was 20.8 bps: second to Northline, 3.1 bps below Harbour and 5.7 bps below Calder on paired differences. The tested fee line was zero, so the measured cost is spread and slippage.
The post-trade statement separated spread, fee and settlement and exported as CSV without reformatting. Median completeness was 9.45 out of 10.
OPEN THE BAERX FILE →WHY NO COMPOSITE SCORE
The trade-off is the result.
Northline was cheaper and filled faster. Baerx onboarded, settled and documented better. A single score would force one preference to dominate the other and hide the choice a reader actually has to make. The report keeps the six numbers separate so the preference remains yours.
07 / PARTICIPANT VOICES
What the ten said afterwards.
Each quote was collected in a fifteen-minute debrief after the participant’s final task, translated where necessary, and checked by the participant before publication. Quotes are edited for length only, never for sentiment. Negative observations about the subject venue are included.
I needed the fee line before I confirmed the second leg. Baerx showed the spread and the fee separately; one of the other venues folded both into one number.
First-time buyer · IENew account · quoted on Baerx
observation ↗The transfer was in my account before I finished the debrief. The confirmation screen stated the payment cut-off, and the money moved inside it.
Regular user · DEExisting account · quoted on Baerx
observation ↗The second leg filled a couple of ticks worse than the price on the ticket. Nothing dramatic at two thousand euro, and the confirmation showed both numbers, so I could see where the difference came from.
Regular user · NLExisting account · quoted on Baerx
observation ↗Calder asked for the same address document twice. I uploaded it again, waited, and stopped checking the status page.
First-time buyer · FRNew account · quoted on Calder
observation ↗The two-thousand-euro round trip cost me 17.8 basis points. What I valued was the itemised statement; I could reconcile every cent without building a spreadsheet.
High-value user · ESExisting account · quoted on Baerx
observation ↗Support answered in seventy-three minutes with a real sentence, not a ticket number. Harbour replied in ninety and sent a template.
Regular user · ITExisting account · quoted on Baerx
observation ↗Northline rejected my residence proof and never told me why. Baerx approved the same document in twenty-two minutes.
First-time buyer · PLNew account · quoted on Northline
observation ↗Meridian's withdrawal was still processing the next day. The money arrived on the second day, but the status page never changed.
Regular user · SEExisting account · quoted on Meridian
observation ↗I ran the same two-thousand-euro pair on all five. Only Baerx and Harbour produced a statement I could hand to my accountant without reformatting.
High-value user · BEExisting account · quoted on Baerx
observation ↗The onboarding queue at Meridian passed three hours, and then it asked for another document. I stopped there.
First-time buyer · PTNew account · quoted on Meridian
observation ↗08 / BOUNDARIES
What this edition cannot tell you.
Sample size
Ten participants and fifty sequences are enough to describe this panel, not to estimate a population. The intervals are wide because the sample is small; we show them rather than hide them.
One pair, one size
The task used BTC/EUR at €2,000. It says nothing about altcoin liquidity, larger clips, market orders in a stressed book, or derivatives.
One fortnight
Collection ran from 14 to 23 September 2026. Operational performance can change after a release, an incident or a change of banking partner.
Participant-side timing
The clock includes the participant’s browser, bank and network. It measures the experience, not the venue’s internal processing time.
No solvency audit
The study tested withdrawal behaviour, not reserves, custody, segregation or internal controls. A successful withdrawal is not a solvency opinion.
No on-chain identifiers
The endpoint is fiat settlement, so the study does not publish transaction hashes. Withdrawal timing is evidenced by bank credit, not by a block explorer.
09 / CONTROLS
Independence, consent and the right to be wrong.
C1
No result-linked payment. Participants were reimbursed for trading and withdrawal costs and paid a fixed sitting fee. The fee did not depend on which venue performed well.
C2
No manuscript review. No venue, including the subject venue, saw the draft or the results before publication. The subject venue supplied no data beyond the retail account used in the task.
C3
Pseudonymous publication. Participants are identified by study ID only. Identity documents, bank details, account numbers and wallet addresses are not published.
C4
Failures stay in. The four incomplete sequences are reported with their reasons. A censored withdrawal is not recoded as a success, and a missing statement is not imputed.
C5
Corrections are versioned. If a participant or venue identifies an error in a record, the correction is dated and linked from the affected page. The original figure remains in the change log.
REPRODUCTION
The full protocol, field list and statistical treatment are public. A third party can rerun the same task and compare the observation register row by row. The benchmark library behind each endpoint is published at /benchmarks.
10 / CLOSE
The useful claim is the checkable one.
A venue can be cheap and still be difficult to leave. It can be fast and still be hard to audit. This edition measured six things that happen to an ordinary user’s money and published every observation, including the ones that weaken the headline. The result for Baerx is strong, but it is strong in a specific, bounded way: complete exits, short settlement, responsive support and a statement that reconciles. That is the claim a reader can inspect.