What changed, and whether it breaks anything

The last five releases

Taken from the same data the application ships, so the version you are running and the version described here cannot disagree.

3.7.1 — 2026-09-14

• Two sentences a buyer could read the wrong way, said unambiguously.
• No breaking changes.
• Fix. Pricing said the interface, the reports and this site are written in English and Spanish. "The reports" reads as the documents somebody exports, and those commit to English — 3.7.0 said so in its own entry. The page now distinguishes the two: the interface and the report views speak both languages, and an exported document uses English as its document language. A buyer deciding on a Spanish rollout should not have to guess which was meant.
• Fix. The consultants' page still promised an archive that verifies itself "when the package ships". It shipped: the evidence package is in the product, in Preview, and downloads from the reports screen. The page says that instead of waiting for it.
• Fix. Both retired wordings are now forbidden on every served page, and the replacement for the first is required on the two pricing pages. This is the fourth piece of copy found by a reader that had outlived the thing it was waiting for — a schedule left Planned two releases after it shipped, two home-page claims explaining under an Available chip that the screen did not exist, and this one. The sentence was true when it was written and nothing re-reads it, which is why the ban is by pattern rather than by correction.

3.7.0 — 2026-09-14

• What a report says it cannot say now speaks the reader's language, numbers included.
• No breaking changes.
• Fix. «Lo que este informe no puede decir» was a Spanish heading over English paragraphs, on every report. Those paragraphs are the part a careful reader reads first — what the analysis could not see, what rests on somebody's judgement, what a clean row does not prove — and they were written where the report is built, so only an English speaker could read them. Twenty-three of them, plus the lines that carry counts, are now named by a key and said in the reader's language; an exported document still commits to English.
• Usability. The lines that count agree in number, and answer rather than counting. "0 fields are neither confirmed nor ruled out" is arithmetic where a reader wants the answer, so with nothing unsettled the record says every field is either confirmed or ruled out. The same for one activity against several, one earlier analysis run against two, one logging call against many.
• Fix. The quality gate was measuring the whole repository as new code. Both projects compare against the previous version, and the scanner had never been given a version to record — so every analysis was labelled "not provided", none differed from the one before, and the baseline fell back to the project's first analysis: twelve and a half thousand lines reported as new on a commit that touched thirty. The scanner now records the version the product ships, read from the same constant /health announces.

3.6.2 — 2026-09-14

• The home page stops explaining, under an Available chip, that the screen does not exist yet.
• No breaking changes.
• Fix. Two claims on the home page carried an Available chip over a body written while they were waiting for one — "nothing yet puts it on a screen" and "this page will keep saying planned until it is not", in both languages, published with a changelog entry saying the contradiction had been fixed. Moving the chip is half the correction and the half a test was watching; the sentence under it was the other half. Both bodies now describe what the product does, and the check no longer takes the chip as the whole claim: an Available row whose own words say the screen is missing fails the build.
• Usability. Pricing says plainly what a workspace is today. "A consultancy running assessments for several clients" invited a picture the product does not offer yet: one deployment holds one organisation's projects, so a consultancy keeps clients apart by keeping deployments apart, under one commercial relationship. Isolating several clients inside a single deployment is specified and Planned, the consultants' page already said so, and the two pages now say it in the same words.

3.6.1 — 2026-09-14

• Every button in the product is a button, and the two sections about the assessment print on a page of their own.
• No breaking changes.
• Fix. Two words of the product's button vocabulary had no declaration anywhere every screen could see. secondary was written out nine times inside components and a tenth time nested inside the report surface, and link four times character for character; a screen that refined one instead of restating it got a bare browser control with a colour applied. That is what happened to "Walk the changes", "See what would be sent" and the "Update report" beside the notice saying which analysis a report is about. Both words now live where primary and ghost already did, and the rebuild button inside that notice is sized for the notice rather than for a landing page.
• Usability. What changed between the last two assessments, and the evidence package, print on a page of their own at the back of the report. They are about the assessment rather than about the document: printed in the middle they read as part of the report, and after a page break they are what they are — the end matter somebody receiving the paper uses to check it. The controls that assemble them are not printed, because a button nobody can press is not information.

3.6.0 — 2026-09-14

• What changed between two assessments can be walked, subject by subject, with what decided each one.
• No breaking changes.
• Feature. The strip above a report has said four fields, two findings raised, one resolved for several releases, and a reader who wanted to know which four had nowhere to go. The walk is now on the reports screen: every subject that moved, what it no longer says and what it says now, and the rules that reached the new conclusion. Only what moved is drawn — a real comparison is mostly agreement, and the number that agreed is said rather than listed.
• Usability. A subject touched by seven rules carries two hundred characters of conclusions on each side, differing in one clause. The walk compares the two and draws only the difference, counting a conclusion reached more than once rather than folding it away — three endpoints exposing one field reach the same conclusion three times, and treating the sides as sets would report a fourth as no change at all.
• Usability. See exactly what changed between two assessments moves from Preview to Available on the home page. It was Preview because the comparison was served and reachable over the API while nothing put it on a screen — a capability reachable over the API is not an accessible experience, and the chip follows the experience.

Older releases are in the application, under About.

Every release names its breaking changes explicitly, including when there are none. "No breaking changes" is a piece of information; a missing section is ambiguous between nothing to report and nobody having checked.

Changes are typed rather than lumped together, because a reader upgrading has different questions about each: what the product could not do before, what is the same capability made clearer, what strengthened a guarantee or closed something with security consequences, and what was simply wrong and is now right.

How to read an entry

Feature

Something the product could not do before.

Usability

The same capability, easier or clearer to use.

Security

A guarantee strengthened, or a defect with security consequences closed.

Fix

Behaviour that was wrong and is now right.

Where to find it

Open the application and choose About, or click the version number in the footer of any screen — that takes you straight to the release you were asking about rather than to the top of a list you then have to search.

The changelog ships as data inside the application rather than as a document fetched at runtime, so a release cannot go out with a changelog that failed to deploy.