Noviscan user documentation — index

Who this is for: whoever is looking for a way into the Noviscan documentation — prospective user, current user, or whoever runs it for others. Assumes: nothing.

Version française : ../README.md

This is not internal documentation. These pages address whoever uses Noviscan, never whoever develops it: the rest of docs/ speaks to agents and to developers, this directory speaks to the customer. The two audiences do not mix in a single tree, and that is the only reason this directory exists rather than docs/fr/ and docs/en/.

What "conforming" requires, and the three families: they live in DEV-Documentation/createurs/documenter-son-application.md, sections « Ce que le contrat appelle "Documentation" » and « Ce que "conforme" veut dire ». They are not copied here.

The writing order and what already exists: ../../USER_DOC_PLAN.md.

I write the Markdown; Sites web publishes it. The public location, the navigation and the PDF rendering belong to them — this directory is the source, not the surface.

Finding your way around the interface

The pages that follow name screens. Here is where they are, gathered in one place.

⚠️ This page is no longer the only one that says it, and that is deliberate since 2026-09-03. A third-party reading showed that a reader caught in the middle of a gesture does not go back up to the index: they need the location where they are acting. So the pages situate the screens they have you open. The two places must move together — this is the same assumed repetition as the retention periods, and it carries the same debt: the day the sidebar changes, this table is not enough.

The locations, once and for all:

The left-hand bar, always visible: Dashboard · Review · Alerts · People · Groups · Sources · Settings.

The Sources page is titled Sources and maintenance: it is the same screen.

The Settings tabs: General and Security for everyone; Users, License, Logs, Activity, Backup and Advanced for administrator accounts only; Legal notice for everyone. A tab you cannot see is not hidden by mistake: your account does not have the role.

At the top right, outside the left-hand bar: the Support button, also reserved for administrator accounts.

Your version — every page states which version it applies from; yours is shown under Settings → General, About section, the "Version:" line.

State of the three families

Defect pass of 2026-08-30 (Produit). All six pages were read by a reader who did not know the product, against five classes of defect (assumed knowledge, gap in the sequence, un-pasteable command, fr/en divergence, unverifiable claim); what came out was settled against the code, never against a memory. Two whole classes are closed: screens named without saying where they are, and acts with no success criterion. ⚠️ This is NOT the third-party review the "State" column asks for: a reader commissioned by the author is still the author. And no walkthrough had been redone on a machine — neither the installation on a clean VM, nor the phone page on a real device. ⚠️ The first half of that last sentence FELL on 2026-09-03: the Windows installation was played on a bench, and what it found is below, with its declared scope. The phone page has still never been retraced on a real device.

Platforms with no document — declaration

The three families above cover Windows only. For the other platforms, here is what the repository establishes — no date, no commitment:

Established by reading: packaging/ (holds windows/ only), .github/workflows/build-desktop-windows.yml (runs-on: windows-latest, no OS matrix), scripts/ (no build-linux, no build-macos), mobile/PISTE_PLAY.json and mobile/eas.json (submit.production empty). The corresponding pipelines are design, not the present: ../../RELEASE_PLAN.md.

✅ The three families now exist — that is point 1 of the master, and it alone.

✅ CONFORMITY REVIEW DELIVERED on 2026-08-17 by the Documentation scope, owner of the master: seven points conforming, one gap, plus two reservations. Point-by-point verdict, with its mode of proof: %LOCALAPPDATA%\Secufor\coordination\VERDICT-CONFORMITE-DOC-NOVISCAN-2026-08-17.md. Seven of the eight points do not attest to themselves — they are written to be checked by someone who does not know the product, which rules out the author of the pages, and that is why this finding comes from elsewhere.

• The gap (point 2) is FIXED by this commit: the "Who this is for / Assumes" block was missing from both indexes, although it heads all six content pages. The point says "every page", and the index is the first page read.
• Reservation 1 — point 4 conforms by VACUITY: there is no screenshot at all, so none can leak. A sincere and empty green, worth knowing before publishing.
• ✅ Reservation 2 — FIXED on 2026-08-30 (Produit). Point 6 conformed to the letter and was false the other way round: the installation page carried "NOT YET AVAILABLE" and spoke of the file in the FUTURE ("will be called", "will be published") while the channel exists. It now separates what is not open — the announcement — from what is: the channel. Decision D-058 is untouched: the page still gives no link. What exists is not described in the future tense; it is the same rule as point 6, taken the other way round.

⚠️ And the Windows installation carries a gap of ANOTHER kind, one that writing does not fill:

• ~~Nobody has ever followed it.~~ Played on 2026-09-03 — verdict, measurements and scope below. The bench freezes the order of arrival, and a missing step only shows when you follow it — the one that was missing was a CHECKBOX.
• Its "Getting the installer" section contains no step to follow: the Windows download is not ANNOUNCED (D-058). It therefore carries a "NOT YET AVAILABLE" mark rather than a link — an announced function is marked as such, or it does not appear.

⚠️ These two points used to be read together — "the page cannot be put to the test as long as the file is not downloadable; the trial is not waiting for a slot, it is waiting for a channel". The second half has FALLEN (measured 2026-08-16: the desktop-0.1.1-… release is published on Secufor/noviscan, a public repository, and its three assets are served without authentication; the checksum of the installer as served was recomputed and matches; three versions published as of 2026-08-30, enumerated at the source — their numbers are not copied here, they are read from the releases of the public repository). The channel exists. What remains is not to ANNOUNCE it, which is a decision (D-058) and not an impossibility.

Consequence for this corpus: the blank-VM trial is no longer blocked by the channel — it is waiting for a bench to be powered up, which is an altogether different obstacle. A wait justified by a false fact goes on forever, because the fact never changes.

✅ THE BENCH TRIAL HAS BEEN PLAYED, on 2026-09-03 — and the page was wrong on the heaviest point of its "Install" section. It announced three checkboxes: the installer offers four, and the fourth one is "Allow Noviscan to access the Internet (updates, online licence)". Unticked by default like the other three — so a reader following the page to the letter installed an offline product without ever reading that they had just decided so. The checkbox does exist in the version described: verified in packaging/windows/noviscan.iss at the commit of the published release (30d789a), not only on main. Fixed the same day, fr and en, with the EXACT wording of the wizard rather than paraphrases.

What the trial measured, on the published release of 2026-08-23 (30d789a) downloaded from the public repository, installed on the Windows 11 bench (build 26200) after a clean sweep:

⚠️ Declared scope, and it is not slight:

• The bench is not a BLANK machine, it is a machine wiped back. The VM is encrypted and its snapshot cannot be restored without a password: the clean sweep (uninstall, deletion of %LOCALAPPDATA%\Noviscan and of the leftovers) approaches the new state, it does not establish it.
• Direct consequence: "the installer lays down WebView2 if it is missing" is not measured — the bench already carries it, and install-report.json writes so itself (webview2: deja-present).
• The download was not played from the bench: the VM has no DNS resolution. The installer was copied there from the workstation, after being pulled from the public repository. The "Getting the installer" section therefore remains untested, which is consistent: it gives no link (D-058).
• The wizard was not driven on screen (that trial came the next day, below): the four checkboxes are measured by their effect in a silent install, which applies the same defaults. What the reader SEES remained to be observed; what the installer DOES when nothing is ticked is established.
• ✅ The first launch WAS played — and it FAILED, which is how a blocking defect was found. What is ESTABLISHED: the data-location window does open before the wizard, as the page announces, and its text is the one described (read on screen, dialog answered with BM_CLICK). What FAILED: after "Yes", no window opened — the process died while trying to download its recognition models on a machine with no network. Fixed the same day in the product; the defect did not belong to the page, but only the page followed to the letter could meet it.
• ✅ The fix is PROVEN ON THE ARTEFACT, on 2026-09-04 — not on the sources: the defect was measured on the delivered binary, its correction had to be measured there too. Signed build 96a8941 installed on the same offline bench, same conditions: location dialog at t+12 s, "Yes", window open at t+18 s (1180×820, process alive, API ready) where the published release returned nothing. Both guards can be read in the application log: the refusal to go out at startup, then the engine's — two sites, two traces.
• ✅ And the wizard DOES appear: its first step is on screen, in French, and the checkbox carries exactly the wording the page announces — "J'ai lu et j'accepte le contrat de licence utilisateur.", unticked, with its "Télécharger le texte" and "Copier le texte" buttons. The window does not overflow the screen.
• ✅ The first three screens of the wizard are PROVEN, driven on screen and not merely looked at (2026-09-04). The page is exact on everything that could be measured:
  • 1. Terms of use — the "Continuer" button returns IsEnabled = False before the tick and True after: the page's most fragile claim, the one it gives as the only sign, is the one a screenshot could not settle — an image shows a button, it does not say whether it answers. The click does bring up Emplacements.
  • 2. Locations — the Données field carries "non modifiable ici" on screen, the other two are editable, as the page announces.
  • 3. How to use this device — the three routes exist with the exact wording of the page's table, and the exclusivity warning is displayed in the application, not only in the documentation.
• ⛔ The walkthrough STOPS there, and that is a boundary, not an oversight: the next step creates an account and asks for a password. Creating an account and entering a password is not mine to do. So the following remain untested: the owner account (§4), product improvement (§5, the box unticked by default), Finish, and the two pairing routes. A person can play them in ten minutes on this bench — a declared gap is a deliverable, a blockage is not.
• ⚠️ And one thing this bench can never establish: that the text stays in French "whatever the language of your Windows". That Windows is in French. A check whose input already equals the expected output measures nothing.

The rule that decides what goes into each page

We document the PUBLISHED version, never main. Every page carries "Applies from version", and that number is the one of the last release published to the public repository — not the one of the working repository. The master forbids describing in the present tense what does not exist yet, and what a user runs is the published installer, never a merged branch.

Accepted consequence: these pages describe less than what the repository knows how to do, and they are updated at every release, not at every merge.