The FX transaction-cost intelligence platform

How far off mid are you?

Banks price every FX trade off the interbank mid-market rate. The distance between that rate and what you actually paid is the fee nobody shows you. Midarx measures it on every trade, in basis points — and defends the number with an evidentiary chain.

the feeyour priceinterbank mid-market
Your executed priceInterbank mid-market rateThe distance = the fee, in basis points
Stable rule IDs
Deterministic exception & calculation rules, each with a fixed ID
Basis points
Markup measured per trade against the mid-market rate
0%
Of figures carry an evidentiary chain to source
SHA-256
Hash-chained analyses — tamper-evident records
Every batch
Timezone reconciliation on every upload

The fee nobody shows you

Your bank never invoices the FX markup. It’s in the rate.

There is no line item. The cost is folded into the exchange rate you were quoted — a few basis points here, a wider spread on the small or off-hours tickets, a drift that compounds across a year of flow. Midarx reconstructs the mid-market rate for each trade, measures the distance you paid above it, and turns a number you never saw into one you can defend in a negotiation.

Estimate your gap

Drag your annual FX volume. Watch the gap.

Larger programs command tighter pricing. This is our public fair-price curve — the fair markup band for a program your size, and the range an unaudited program of that size typically overpays. It’s an illustration from the curve, not a measurement of your trades.

$28M / year
Fair markup for this volume
0.0bps
Fair band 6.216.6 bps · straight off the public curve
Typical annual overcharge, unaudited
$0$0
6.220.8 bps above fair, on $28M/yr of flow

Illustrative only. Modeled from Midarx’s public fair-price curve — not a quote, and not a measurement of your actual trades. Your real number comes from your own trade history.

How it works

Six steps from raw trade files to a defensible number.

The same pipeline the analysis tool runs — intake to export, every figure traceable to its source.

1

Intake

Upload trade history as CSV, Excel, or pasted bank-confirmation text — the files as you receive them. Columns are resolved and every row is normalized to a canonical currency pair.

CSV · Excel · confirm text
2

Exceptions

Duplicates, implausible rates, spot-vs-forward mismatches, and outsized tickets are flagged against stable rule IDs — surfaced for human review, never silently corrected.

Flag, never fix
3

Benchmark

Price against the bundled reference, or bring your own licensed market-data exports — your licensed data never leaves your environment. Sources are cross-reconciled and disagreements flagged.

Bring your own data
4

Agreement compliance

Load your negotiated pricing agreement. Midarx tests every trade against target spreads and variance bands, and assembles a refund-claim schedule where the bank charged above the agreed rate.

Target spreads · refunds
5

Reporting

Produce an audit-grade PDF, a monthly exception-led draft, and trend views — markup by provider, currency, ticket size, and month-over-month movement.

Audit-grade PDF · trends
6

Assistant & export

Ask questions in plain language and export the normalized dataset. Answers cite the rule, the record, and the source — or say when they can’t.

Cite or decline

Who it’s for

One engine, two ways to use it.

For advisory firms

White-label the platform. Scale your book, not your analyst hours.

FX advisors, treasury consultants, and bank-fee audit practices run Midarx as their own platform. Your methodology stays yours — it runs only in your tenancy, never in the shared engine — while the deterministic pipeline does the reconciliation work that used to eat analyst days per client.

  • White-label tenancy under your brand
  • Your methodology runs isolated in your tenancy
  • Audit-grade PDFs and refund-claim schedules per client
  • Grow the number of clients without growing headcount
Request a platform walkthrough

For mid-market corporates

Moving $2M–$200M a year through bank FX? Know what you actually paid.

Upload a sample of your recent trades and see your markup against the mid-market rate — by currency, by ticket size, over time. It’s a preliminary review: a fast, concrete read on what your FX is really costing, positioned as early access to the platform.

  • Upload a sample, see your markup in basis points
  • Spot where spreads widen — small tickets, off-hours, specific pairs
  • Walk into your next bank conversation with the number
  • Early access — start with a preliminary review
Request a preliminary review

DefensibilityPatent pending

Built to be believed — and to hold up.

A number is only useful in a negotiation if it survives scrutiny. Midarx is engineered so every figure can be independently re-derived, and every inference is on the record.

Deterministic computation

Every markup, exception, and compliance verdict comes from a rule with a stable ID — the same inputs always produce the same output. No black box.

Disclosed inferences

Where the engine infers — a timezone, a carried-forward reference date — it says so, in the report and in the record, with its rationale attached.

Tenant isolation

Advisory methodology and client data run only in your tenancy. Nothing crosses into a shared engine or another tenant.

No model training on your data

Client data is never used to train shared or public models. It is analyzed, reported, and left in your environment.

Hash-chained records

Each analysis is committed to a SHA-256 hash chain. Any later change to a record breaks the chain — tampering is evident, not silent.

Bring-your-own benchmark

Use your own licensed intraday exports. They are cross-reconciled against the reference and never leave your environment.

Get started

Request access

Tell us a little about your firm or your FX flow. We’ll follow up with next steps — a platform walkthrough for advisory firms, a preliminary review for corporates.

This form composes a prefilled email in your own mail client to info@celesium.ai. Nothing is stored on this site — no account, no backend, no data at rest.

See it on real numbers first.

The sample analysis renders a full report on synthetic trade data — KPIs, exceptions with rule IDs, the fair-price curve position, and the hash-chain head. No login.

Open the sample report →