GOXEVA cost & mechanics desk
English

Editorial policy

How a piece gets chosen, sourced, drafted and checked before it goes up. Including the part most sites in this category leave out, which is what role a language model plays.

What this site covers, and what it refuses

The remit is narrow on purpose: the cost and mechanics of moving money through a crypto exchange. Fees, transfers, margin arithmetic, why a process stalled. Things that have a checkable answer.

What is deliberately out of scope: price predictions, which asset to buy, which project is undervalued, anything resembling a signal or a portfolio suggestion. Not because those are uninteresting but because this desk has no basis for them, and a site that answers "what does a withdrawal cost" and "what will BTC do next quarter" with the same confidence is telling you something about its standards.

Sourcing, ranked

Three tiers, used in this order, and a piece says which one a given claim came from:

  1. Primary — the exchange's own published pages. Fee schedules, help articles, contract specifications, terms. Authoritative for anything the exchange controls, and volatile, so every figure taken from one carries the date it was read. Several pages carry a dated screenshot of the source page so you can see what it said rather than take our word for it.
  2. Protocol-level facts. Block times, how a memo field works, what a confirmation is. Stable, and independently checkable against a block explorer or the protocol's own documentation.
  3. Working through the products. Opening accounts, moving funds, noticing where a flow stalls. This is where the sense of which step actually trips people up comes from.

The third tier is the least verifiable, so it shapes what gets explained and never supplies a hard number. You will not find an account screenshot, a claimed fill price or a transaction hash on this site, because those are trivial to fabricate and impossible for you to check.

Where a language model is used, and where it is not

Stating this plainly rather than leaving it to be inferred, because a lot of sites in this category are model-written and quiet about it.

Used for: drafting and revising articles, comparing language editions, finding and reading public sources, and creating or running checks. AI assistance is part of both writing and verification; it is not evidence that a claim is correct.

Evidence first: platform rules and figures need a dated public source. Model output alone is not a source. The corrections log records changes to published claims. A source check may use browser or automated tools; it should not be read as a claim that every figure was transcribed by a person.

Not acceptable: inventing experience, evidence, statistics, cases or quotes. A worked example must be labelled as an example, and a screenshot must show the public page it describes.

AI and automated checks can miss context or repeat an error. Source links, calculation steps and explicit limitations let readers check the result. Publication decisions remain the responsibility of the site owner.

Fact-checking before publication

Checks depend on what is being changed. An editorial update does not mean every figure or external link across the site has been checked again:

  • New or changed platform claims are checked against a relevant official source when accessible. An older dated figure remains a historical snapshot unless it is explicitly rechecked; the account screen and current official rules govern an actual transaction.
  • Arithmetic in worked examples is recomputed independently of the text that describes it. Tables are recalculated row by row, not eyeballed.
  • Calculator formulas are checked against known reference cases before the page ships, and the page states what the tool does not model.
  • Sources supporting changed claims are opened where accessible. A failed or unavailable source is a verification limit, and a date beside a citation does not guarantee that the destination still says the same thing today.
  • A deterministic script checks the whole site: internal links resolve, structured data parses, FAQ markup matches the visible text, images exist and carry dimensions, and the affiliate link count and attributes are what they should be.

None of that catches a bad judgement call. It catches the mechanical failures, which is most of what goes wrong.

The three language editions, and what that does not mean

Each article and calculator has English, Kiswahili and French editions. English is the source edition. The other editions are maintained as separate pages and use AI-assisted translation and rewriting, with checks of numbers, steps and risk explanations against the English version. Examples may be adapted for local payment methods or currencies, including CFA francs in the French edition. This process does not amount to review by a native-speaker financial editor.

What that does not mean is that either the Kiswahili or the French edition has been signed off by a native-speaker financial editor. Neither has. Nobody with that background has read either one end to end, and until somebody has, expect wording that is accurate but occasionally stiff. Where two editions disagree on a fact, treat it as an error worth reporting rather than as a deliberate regional difference.

All three editions are held to the same sourcing and fact-checking rules set out above, and a correction logged against one is applied to the others in the same pass.

The commercial arrangement and what it is not allowed to touch

This site carries one Binance referral link. That is disclosed on the disclosure page, at the point where the link appears, and in the footer of every page.

The rule applied to every page: would this still be worth publishing if the referral link did not exist? If not, it does not go up. In practice that means no page exists solely to route people to a sign-up, the navigation carries no call-to-action, and the piece that does carry the link spends a section explaining that the discount is worth less than people expect.

No page is reviewed, approved or paid for by any exchange. There is no sponsored content, no paid placement, no guest posting, no link insertion. If that ever changes it will be labelled on the page itself, not buried here.

Corrections and updates

Errors get logged with a date and a description on the corrections page rather than silently edited out. The log distinguishes two cases that are easy to blur: this site was wrong, and the exchange changed something. Collapsing them into one is a way of never quite admitting the first.

Dated figures and screenshots describe what was available when checked or captured. A wording correction does not refresh their dates. Changes to facts or tool behaviour are recorded separately so readers can see the scope of an update.

Not logged: typography, spelling, and phrasing tightened without changing meaning. Logging those would bury the entries that matter.

Authorship

Everything here is published under the pen name Aurel Kestner. The about page says why, and says plainly what that costs you as a reader: you cannot verify a professional history, because none is claimed. There are no invented credentials on this site, no follower counts, and no linked social accounts.

What is offered instead is that the work is checkable. Figures are dated and attributed. Formulas are derived rather than asserted. Calculators show their arithmetic. That is the only kind of authority a pseudonymous desk can honestly put forward, and it is the standard this page exists to be held to.