our method
How we build this site
Poker content is full of numbers that nobody can check. This page describes the checks that run before any page here is published, and the things we have refused to publish because they could not be checked. It is an account, not a pitch.
A wrong number is worse than no number
Around nine hundred thousand people a month search for one thing: what beats what. If the hand ranking page gets that wrong, it teaches them the wrong thing and not one of them can see it. The same applies to a pot odds figure, to an amount we charge, and to the price of somebody else's software. So the rule on this site is not that we try hard. The rule is that the build refuses to publish a page whose claims it cannot verify.
Everything below is in the repository this site is built from. The commands are ordinary ones, they run on every deploy, and each of them has failed at least once on something real. What they caught, and what had to be corrected afterwards, is listed in the changelog. That is the only reason to trust any of it.
what runs before publishing
Eight checks.
controleer-prijzen.mjs
Every price matches one file
Our own amounts live in a single data file. The check reads that file, then walks every page looking for a plan name coupled to an amount, and fails the build when the two disagree. It also reads structured data, where the currency sits behind the number instead of in front of it. That form is invisible to a human reader and it is exactly where an outdated annual amount survived for a while: the visible page had been corrected, the machine readable block had not.
test-handevaluatie.mjs
The hand evaluator is tested, not trusted
Twenty three cases run before anything is built: the wheel, an ace that may not sit in the middle of a straight, a flush that has to be picked out of six cards, the kicker, and a split pot. If one of them comes out wrong, no page is produced at all.
poker-hands.astro
Every hand shown is re-evaluated
The ten example hands on the hand ranking page are not believed. Each one is fed back through the evaluator, which has to agree with the label printed next to it, and each hand has to genuinely beat the hand below it, measured pair by pair rather than assumed from the order they were typed in. The royal flush additionally has to end on the ace, otherwise any straight flush would slip through. The six practice questions check their own answers the same way. The combination counts have to add up to 2,598,960, which is every five card hand in a 52 card deck.
woordenlijst-nl.ts
The glossary recomputes its own numbers
Every hand type claimed in an example is confirmed by the evaluator, and every percentage in the running text is calculated again while the site is built. The flush draw on the flop is counted out as 378 of 1,081 remaining two card combinations, which is 34.97 percent. Flopping a set with a pocket pair is 2,304 of 19,600 flops, or 11.76 percent. The tournament equity figures come out of the ICM model rather than a table someone typed. A number that used to be right is not the same thing as a number that is right.
controleer-links.mjs
Every internal link has to arrive somewhere
The check walks the built site and follows every internal link to a page, a file, or a redirect rule. This exists because a domain move once left behind a few hundred links inside object literals and template strings, which is precisely where find and replace does not reach. A dead link now blocks a deploy instead of sitting on the site for a month.
controleer-huisstijl.mjs
Four house rules you cannot see
No dashes in visible text, in either language. Exactly one main landmark per page, because two of them turn "skip to the main content" into a menu, and you only notice that if you are listening rather than looking. A usable title and a usable description on every page. And a language attribute that matches the path, so an English page cannot announce itself as Dutch.
controleer-bereikbaarheid.mjs
Pages that only the footer links to are reported
This check deliberately ignores the navigation and the footer. Those appear on every page, so counting them would make everything green and measure nothing. What is left is the question of whether the content points at itself. The first run, on 30 August 2026, found 292 pages that could not be reached from the homepage through the content, and 19 with no incoming link at all, including several section hubs.
BaseLayout.astro
Translation pairs are verified during the build
The list linking each English page to its Dutch counterpart is written by hand, so it can age the moment somebody renames a page. The build therefore checks both sides exist and throws if one does not. A translation tag pointing at a missing page tells a search engine the translation exists, and then the page that does exist disappears from the results of the country it was written for. Loudly broken is cheaper than quietly wrong.
All of these run from one command, in this order: prices, the hand evaluator, the type check, the build itself, the link check and the house style check. The reachability check is run by hand, because a page that is only reachable from the footer is a weak signal rather than an error, and blocking a deploy on it would be the wrong trade.
what we do not do
Six things we have thrown away.
Rules are cheap to write and only mean something once they have cost you a page you liked. Each of these was on this site, and each was removed rather than repaired with a better sounding version of the same claim.
No number without a source
An article about the Poker IQ score described a scale that had already been corrected in the app months earlier, and said it was calibrated against roughly 1,500 players in a closed beta. Those players never existed. The article was rewritten around the figures that are actually in the code, and the invented cohort was removed rather than replaced with a smaller one.
No competitor price without a date
The comparison hub once listed two subscription tiers of a competitor, by name and by amount, that the vendor does not sell under those names. A five year cost table was built on top of one of those invented amounts. The table was not recalculated, it was deleted, because converting it into another currency would only have made it age with the exchange rate as well. What replaced it is what the vendor published on the day we looked, in the vendor currency, with that date on the page.
No invented research
A page about small stakes cash claimed a finding drawn from more than 1,500 hands at a stake where not a single hand had been played. There is no in house study behind anything on this site. Where a claim rests on arithmetic, the arithmetic is in the repository and runs during the build. Where it rests on somebody else, they are named.
Where we do not know, it says so
Two tiers of a competitor went up in price without the vendor publishing the new amounts, so those cells are empty. Where a poker room has no verified export route documented here, there is no route rather than a plausible one. An empty cell is more honest than a filled one you cannot check.
No promise about behaviour that only code can keep
The word unlimited was used for a coaching feature that has a fixed limit written into it. Priority support was promised on a page while nobody had committed to a rota. Both were removed, and the same reasoning is why the help page on this site names no response time.
Nothing that only looks like measurement
Every page loaded an analytics script for an account that did not exist, so each visit sent a request to a third party and no measurement came back. The script now loads only when the account is configured. Nothing was gained by the appearance of measuring, and something was lost.
Where the checking stops
A green build proves a narrow set of things. It cannot tell you that an explanation is clear, that the example was the right one, or that a page deserved to be written. It cannot catch a claim about the future, and it cannot keep a competitor price fresh: the solver comparison carries the date it was read and will quietly age from there. Where the numbers do come out of a model, you can see one end to end in the Poker IQ explanation, limits included.
Some things cannot be verified at all, and then we say so instead of dressing it up. Whether poker rewards skill over a long enough run has a real answer with real evidence, and the page on skill and chance gives the evidence and its limits rather than a slogan. Whether you personally will win money is not something anyone can promise, which is why responsible play is a page here and not a footnote.
The last honest bit: this site is written by one person with an assistant, not by a content team. That is why there are checks in the first place. Volume without verification is the one thing a small operation genuinely cannot afford, and it turns out to be the thing readers cannot afford either. If you want to see what the checks are protecting, the glossary and the calculators are where most of the recomputed numbers end up.
questions
Fair questions.
Who writes this site? +
It is written by the person building Pokiq, working with an AI assistant, and every claim is checked against the code, a vendor page, or a calculation that runs during the build. There is no content team and no freelance pool, which is why the volume is modest and the checks are strict.
What happens when one of these checks fails? +
The build stops and nothing is published. That is the whole point: the checks are wired into the same command the hosting platform runs, so a dead link, a second main landmark or a price that disagrees with the source file blocks a deploy rather than living on the site for a month.
Does a passing build mean the page is good? +
No. These checks are mechanical. They can prove that a hand is a flush, that a link resolves and that an amount matches its source. They cannot tell whether an explanation is clear, whether the example was well chosen, or whether the page was worth writing. That part is judgement, and judgement is not automatable.
How current are the competitor prices? +
Each comparison page states the date its figures were read from the vendor. Software pricing moves and these pages do not update themselves, so anything more than a few months old should be treated as an indication and confirmed with the vendor before you pay.
Do you use affiliate links? +
No. No poker room and no software vendor pays for a mention here, and there are no tracking links on the comparison pages. That is worth stating plainly, because the usual business model of a poker site is exactly the opposite.
Can I point out something that is wrong? +
Yes, and it is the most useful thing you can send us. The contact address is on the about page. A correction with a source attached gets acted on faster than one without, for the same reason everything above exists.
next step
See the same standard applied to a report
The example report shows what an analysis of a real session says, how each finding is supported, and where it states that the sample is too small to conclude anything yet.
Example report →