Charter

This is what FlowFinds Solutions Kft. has bound itself to. It is written as commitments rather than values because a value cannot be breached and a commitment can. Each one below is specific enough that a reader could catch us failing it, and several of them already cost us something — that is how we know they are commitments rather than descriptions of what we felt like doing anyway.

A charter of this kind is only worth the mechanism behind it. Where a commitment is enforced by a script, a gate or a published artefact, that is stated. Where it is enforced only by our own discipline, that is stated too, because the difference matters to anyone deciding whether to believe us.

Part one — What we publish

1. We publish the runs the agent failed, not only the ones it passed

Every recorded run of our benchmark is published, in full, whatever it says. A failed run is not held back for a later release, quietly re-run until it passes, or summarised into a sentence that omits it. The research section reads the recorded runs at build time, so a page cannot show a different result from the artefact it came from.

The commitment has teeth because of its converse: we do not publish a number that has no recorded run behind it. Every run we have is listed with its date and the model version that produced it.

2. We publish the method, so the result can be checked rather than believed

Scenario construction, the scoring rubric, the run protocol and the hashing of inputs are published in full before anyone is asked to take a result seriously. A benchmark whose method is private is a marketing asset with a decimal point in it. The methodology is here, and the procedure for running it against us is published for people who would rather check than read.

3. No page on this site types a benchmark figure by hand

Scores are read from artefacts at build time. Every other part of the site links to the research section instead of restating a number, which means a stale figure cannot survive in a paragraph somebody forgot to update. This is enforced by rule rather than by intention: a hand-typed score is treated as a defect.

4. A regression is a release blocker, not a footnote

When the numbers move the wrong way, that is grounds to stop shipping the thing that moved them. The benchmark is a constraint on the product rather than a promotion of it. A capability we cannot measure does not ship as a promise.

5. We publish the limits on the agent’s authority

What the agent may spend, what it may sign for, what it may publish under an operator’s name, and the points at which a person is required are all written down in public. An undocumented limit is not a safety property; it is a habit. The safety approach states them, and product security states how they are enforced.

Part two — What we claim

6. We claim no certification we do not hold

We hold no security certification and we display no badge suggesting otherwise. No audit report, no attestation, no compliance logo, no “enterprise-grade” standing in for one. When we obtain a certification it will be named, dated and verifiable; until then the security page describes controls we actually operate and says plainly what is absent.

7. We claim no customer, partner, endorsement or logo we have not been given

A company name appears on this site when that company has agreed to it. An integration is described as an integration; it is not upgraded into a partnership because the word reads better. Partners states the nature of each relationship rather than implying one.

8. We do not invent a number to make a page feel substantial

No fabricated user counts, revenue figures, growth rates, testimonials, team members or funding. Where we have no figure, the page says less and stays true. This is the single rule that most often makes our pages shorter than a competitor’s, and we accept that trade deliberately.

9. Every claim the product makes to an operator carries its evidence

A recommended product carries the observed listings behind it. A margin carries the basis it was derived from. A storefront section order carries the reason it was ordered that way. A campaign carries the pass criteria set before the spend, not after it. An operator is entitled to disagree with the machine, and disagreement requires the working.

Part three — What we fund and what we charge

10. The first launch is funded from company money

The domain and the initial advertising spend behind an operator’s first launch are paid for by us. No card is taken for it, it is not a trial that converts, and it does not require inviting anybody. We carry the cost of the first attempt because we are the party claiming the first attempt is worth making. What we fund is set out on pricing.

11. Nothing is metered in a way we will not show you

Usage is charged against a published per-action rate card, each feature has its own allowance rather than a single pooled bucket, and when an allowance is exhausted the operator is told. We do not quietly serve free work, and we do not quietly serve worse work — a silent downgrade is a lie told by a system rather than a person, and it is still a lie.

12. Advertising money is never consumed by a subscription

A plan pays for the FlowFinds machine. Advertising draws on a separate, visible balance, and an amount is shown as funded only when a balance actually holds it.

13. We take no payment for a recommendation

No supplier, no vendor and no advertiser can pay to have a product recommended, ranked higher, or presented as a better opportunity than the evidence supports. A recommendation that can be bought is worth nothing to the operator receiving it, which would make the entire find engine worthless. This is a commercial commitment with a real cost, and it is the one we would be offered money to break first.

Part four — What we will not do to grow

The commitments above describe how we work. These describe the growth we have decided not to have. They are here because every one of them is a technique that works, that we are capable of, and that we have chosen against.

14. We will not sell the outcome as a certainty

We will not market FlowFinds with income promises, earnings screenshots, lifestyle imagery, countdown pressure, or a story in which the software makes the risk go away. Selling online can fail with our agent as it can without it. The honest sale is that the work is done properly and shown; the dishonest sale converts better, and it is not available to us.

15. We will not make a referral programme into a recruitment scheme

Referral rewards qualify on an observed event — a person joined and reached their dashboard — and pay a fixed amount that funds a launch. There is no tier that pays on another person’s recruits, no earnings from anyone’s downstream, and no structure in which introducing people becomes the product. The rules are published.

16. We will not make leaving difficult

No dark pattern in the cancellation path, no retention offer that hides the cancel button, no support queue in place of a button, and no hostage-taking of an operator’s store, domain or data on the way out. What we hold, for how long, and how it is exported or deleted is set out in data handling and retention.

17. We will not sell or trade the data an operator gives us

Customer data is not sold, brokered, or shared for another party’s marketing, and it is not used to train models — an obligation we also impose in writing on every party that processes data for us. Every such party is listed, with at least thirty days’ notice before that list changes.

18. We will not compete with the operators we serve

A product claimed by an operator is exclusive to that operator on this platform, and we do not take a claim for ourselves out of the same catalogue in order to run against them. The exclusivity mechanism is worthless if the house is allowed to play.

19. We will not ship a capability we cannot measure and then describe it as measured

Where something is unmeasured, it is labelled unmeasured. The gap between what the agent can do and what the benchmark covers is a known gap, tracked as one, rather than a silence.

Part five — Holding us to it

20. A breach is a defect, and it is fixed as one

If we fail one of the commitments above, the response is the same as for any other defect: it is written down, its cause is found, and where the failure is mechanically checkable, a check is added as part of the fix so it cannot recur unnoticed. A defect found once is a bug; the same defect found twice is a missing check. How we build describes the machinery this relies on.

21. This document changes in public, or not at all

A commitment here is not quietly softened. If one of these is wrong, or we can no longer keep it, it is changed in the open and the change is dated. A charter that can be edited silently is a preference, not a commitment.

22. Tell us when we have broken one

If you can show us a page, a product behaviour or a message that breaches this charter, write to [email protected]. A security issue goes to responsible disclosure, which states what we commit to in return. Everything else is answered by a person.

The reasoning behind these commitments is in the mission; the machinery that enforces them is in how we build; the output they are meant to protect is the research.