The correctness standard
RevByDesign builds software with correctness as its primary engineering objective. Correctness is not a permanent state, a certification, or a claim that a product is free of defects — it is the standard we assess our own work against, the evidence we gather, and the failures we are prepared to acknowledge.
01 — What this is
A standard, not a promise
Every studio says it cares about quality. The claim is unfalsifiable, which is why nobody believes it and why we have tried to write something you can actually check instead.
Below are six dimensions we hold our own work to, the things this standard explicitly does not claim, the places you can go to verify any of it, and the places it is currently not met. That last section is the one that matters. A standard with nothing admitted against it is marketing.
It applies to everything we build — Unity assets, applications, tools and integrations — and it does not replace the mechanisms that actually enforce it. Tests, schemas, release gates, changelogs and development records do that work; this page names what they are for.
02 — The six dimensions
What we check
-
Behavioural correctness
We check whether software behaves as documented, including at boundaries, under invalid inputs and when operations fail. We distinguish between behaviour that has been inspected, compiled, automatically tested, integration-tested and observed in use.
A passing test is evidence for the behaviour it exercises, not proof that the entire product is correct.
-
Architectural correctness
We aim for deliberate responsibilities, explicit dependencies and defensible boundaries, and we document what a component owns, what it does not own, and where responsibility transfers to you.
Architecture is not assumed correct because it is modular, sophisticated or extensively documented.
-
Documentation correctness
Documentation must describe the implementation that actually exists. READMEs, XML comments, examples, tutorials, API references, product pages and release notes are all subject to it.
Known limitations must not be hidden behind favourable wording. Unsupported behaviour must not be presented as supported.
-
Integration correctness
We aim to make published interfaces usable without requiring you to discover hidden assumptions by trial and error — integration requirements, component placement, dependencies, refusal behaviour, diagnostics, and how much adapting your side has to do.
We do not assume that a documented API is necessarily a discoverable or usable one.
-
Historical correctness
We keep a truthful record of material changes, defects, limitations, mistakes and reversed decisions — including issues found internally and fixed before anyone else met them.
A defect does not disappear from the record because it has been fixed.
-
Ethical correctness
We describe products accurately and give you enough information to make your own decision. Where a defect may exist in your code for the same reason it existed in ours, we say so and say how to check.
We do not inflate compatibility claims, represent incomplete verification as completed testing, or present aspirations as implemented features.
03 — Limits
What this standard does not claim
Kept at the same weight as the standard itself, because a standard without stated limits is a slogan.
- That RevByDesign software is defect-free.
- That every architectural decision is optimal.
- That every integration requirement has already been discovered.
- That passing tests establish universal correctness.
- That every internal change requires a public incident report.
- That a documented limitation is automatically an acceptable limitation.
- That adopting this standard certifies any product as compliant.
04 — Evidence
Where to check it
None of this is worth reading unless you can test it, so here is where to look. Where the evidence is internal, this page says so rather than implying otherwise.
Every product page states what it isn’t
Not as a habit — it is a required field in the content schema, and a product page that omits it fails the build. The check: go through the index and find one without it.
Everything carries its real state
Available, in development, prototype, unreleased or archived, on the index and on its own page. Nothing is hidden for being early, and there is no “experiments” shelf doing the work of a polite word for abandoned.
How RevFramework is tested, in numbers
The testing page ↗ states the CI counts rather than claiming “well tested”, explains the two gates that fail a build which quietly tested less than it should, and notes that results are parsed directly because Unity will report success for a run that executed nothing. Each of the nine systems then has its own testing page with a section headed what is deliberately not tested.
What RevFramework refuses to do
Authority ↗ opens by stating that the framework does not implement networking, replication, prediction, rollback or reconciliation, and where responsibility passes to you instead.
Defects, in the changelogs
Each product’s changelog carries the defect accounts, with the user-visible symptom, the cause, and what changed — RevFramework ↗, RevLearning ↗ and Snap Studio Pro ↗. Where a defect might be in your code for the same reason, they say so — RevFramework’s on a rollback that could lose an item: “if your resolver maps aliases only, your rollbacks can lose items the same way.”
Some of our strongest evidence is internal and you cannot read it: a commissioned adversarial review of Snap Studio Pro that concluded the product was not ready and named a defect shipped in three releases; gate records for Skint that pass on evidence rather than on the work looking done; and a development log for ORDA with a verification section per milestone. They are named here because pretending the published record is the whole record would breach dimension 3 on this very page.
05 — Shortfalls
Where it isn’t met yet
An unresolved issue is not a reason to abandon a standard. It is a reason to record where the standard has not been met.
Purchase recognition is fixed, not confirmed
The test-suite download checked your Asset Store invoice against a list of product titles by exact match, and one entry was only the leading part of the full listing title. If Unity returned the full title, a Complete Bundle customer could have been told their purchase was not a RevFramework product. Since 29 October the check matches on the leading part instead, so either form is accepted and that refusal can no longer happen.
What has not happened is anybody watching it work. Nobody has yet observed what Unity actually returns for a Complete Bundle purchase: the diagnostic that would show it only runs once a real invoice gets through, and until 29 October nothing could get through at all — see the correction below. So this is fixed by removing the dependency rather than confirmed by observation, and those are different things. If a purchase of ours is ever refused, tell us and you will be sent the download directly.
Verification is documented unevenly
RevFramework’s testing documentation is the standard the others are not yet held to. For the remaining products the suites exist and run, but nothing published describes what they cover or where they stop. That is a documentation gap rather than a testing gap, and it is ours to close.
Three of the six dimensions rest mostly on practice
Documentation, history and ethics are carried by mechanisms that fail loudly — required fields, build gates, release checks. Behaviour, architecture and integration are carried further by judgement, and judgement erodes more quietly than a failing build does.
06 — Corrections
Claims we have withdrawn
This record starts today, and a short list is not a claim that little has been wrong. Corrections made before now live in the product changelogs, where they were written at the time; site-level corrections will be recorded here, dated, from this point.
30 October 2026 — the test suite shipped two releases behind
For eight weeks, from 4 September to 30 October, the test-suite download served a package built for version 1.2.0 while the framework itself shipped 1.3.0, 1.3.1 and 1.3.2. The page offering it said the download is always the current release.
The instruction printed beside that claim was update the framework first if you are behind — asking the customer to be current while the download was the thing that was behind. The same page explains that the suites are pinned to the release they shipped with, and that running them against a different version reports version drift rather than real bugs. So anyone who hit this had the explanation in front of them and no reason to apply it: the page had already ruled out the one cause that was actually in play.
Not a cosmetic mismatch. Version 1.3.0 carried a migration, so the likely experience was compile errors, or a screen of failures in code that was working correctly.
We cannot say who this affected — nothing records downloads, so there is no number to give you. Corrected on 30 October by publishing the 1.3.2 package and confirming the stored copy is byte-for-byte identical to what was built. The underlying cause was that republishing the suite was not part of the release routine, only of the documentation describing it.
29 October 2026 — the test-suite download could not be reached
For at least nine days, and probably longer, the invoice form on the RevFramework documentation could not reach the service that verifies it. The documentation had moved to its own subdomain; the verification service was still only accepting requests from the address the documentation used to live at, which had not been changed since July. Browsers blocked every request before it arrived.
The message shown was “Could not reach the verification service. Please try again shortly.” — which reads as temporary. Anyone who hit it would reasonably have waited, retried, and given up, with no particular reason to report something that claimed to be fixing itself. A misleading error is worse than a blunt one, and this is the second kind of defect in the same fault.
We cannot tell you who this affected. A request blocked by the browser never reaches the service, so nothing recorded it; the failures were invisible to us by construction. An estimate here would be a guess dressed as a figure.
Fixed and verified on 29 October by submitting the form and watching a real answer come back. If you tried to download the test suite and could not, it works now. If it still does not, say so and you will be sent the package directly.
LoggerPro 2.0.0 — release notes described a build you could not get
The 2.0.0 notes on the Asset Store described a package that was not downloadable from that page. Everything they claimed was true of the 2.0.0 build — exported, verified against a project created from scratch, fifteen checks of fifteen passed — but it was never the file the store served. The claims were accurate about the software and wrong about what you could get. Recorded in the listing’s ↗ own release history under the heading correcting what we published, rather than edited away.
Old links passed none of the credit they were supposed to
Every old RevGaming URL ends in a slash, and the redirect rules were written without one. You still arrived — a meta refresh carried you a beat later, so nothing looked broken from a browser — but the redirect passed none of the ranking signal it existed to pass, and had been doing that since it was written. Found by following a real old URL to a final status code instead of testing the redirect in isolation. Fixed, and the build now fails if a redirect stops matching the URL anybody would actually request.
07 — Accountability
Revising this
This standard applies to our own work, not merely to the claims we make about that work. It is open to revision where evidence shows its requirements are insufficient, impractical or wrongly expressed, and material changes get dated and explained rather than quietly replaced.
If you find something on this site or in a product that contradicts any of the above, say so. Being shown a mistake we did not know about is the standard working, not an embarrassment to manage — the response is to establish what is actually true, correct the implementation or the claim, and record it.
The objective is not to demonstrate that we were always right. It is to make our decisions examinable, our claims supportable and our corrections visible.