Skip to content
Skip to main content
Novel Systems home
Corrections

What this site used to say

If you saved a page from this site and a number on it no longer matches, this is where to look. 7 entries, newest first. Each one quotes the old wording so you can match it against your copy, states what the page says now, and links the decision that argued the change.

Last entry August 19, 2026

Why this page exists

On 6 August 2026 someone compared an archived copy of this site against the live one and reported five differences. He was right about four of them, and the fifth turned out to be a live reading behaving as designed. His conclusion was that a site whose numbers move between visits is not a record of anything.

That conclusion was fair, and it was fair because we had given him no way to reach a different one. A changed number with a dated entry behind it reads as a company that corrected itself. The same change with no entry reads as a company making things up. The difference is not in the number.

The reasoning behind each change was already published in the decision log, but that log has grown past ninety entries and answers a different question: why the platform is built the way it is. A record only usable by someone who already knows which entry they need is an archive, not a record. This is the short list.

  1. C-007August 19, 2026Corrected

    How many industry verticals the platform covers

    What it said
    The opening paragraph of /solutions read "The platform is the same in all five." The grid immediately below it rendered eight, and the counter tile above it printed eight, because both are derived from the catalogue. Only the sentence was typed by hand.
    What it says now
    The same sentence, interpolating the catalogue: the platform is the same in all 8. It cannot disagree with the grid beneath it again.

    Why

    The number of verticals grew and the prose did not follow. This is the same failure the openGraph description on that page had on 5 August 2026 — a hand-typed count on a page whose every other figure is derived — and it survived because a reader who noticed would have had to count the cards to catch it. It was found while auditing the page's headings for search, not by a check, which is the honest way to record it: no rule caught this one.

  2. C-001August 6, 2026Corrected

    Named customer case studies and attributed testimonials

    What it said
    Five case-study cards naming commercial contractors as customers, each with a testimonial attributed by name and job title to a member of their staff. The names were replaced once before they were removed, so an archived copy may show either set.
    What it says now
    No customer is named anywhere on this site, and no testimonial is published. The cards are deployment scenarios: they describe how the platform handles a workflow, and every figure on them is produced by the live engines and reproducible in the sandbox. The demo workspaces are labelled Sample Workspace in the tenant name itself, and the work orders inside them carry facility types rather than organisations.

    Why

    None of those companies were customers and none of them were asked. The testimonials were written in-house and moved from placeholder to approved in a single edit that every guard then in place passed, because the guards were checking whether the wording looked finished rather than whether anybody had consented. Replacing one set of invented names with another set was the wrong fix and is why the archived copies disagree with each other. The removal went further on the same day: the sample workspaces were describing work for eighteen real, identifiable Ontario organisations — a transit agency, hospital networks, universities, a municipality, and several large commercial landlords — as though they were clients, complete with dollar values and gross margins. Publishing a quotation now requires a consent artifact committed to the repository; a build with a named third party and no such file is a red build.

  3. C-002August 6, 2026Corrected

    The founding narrative and the length of the track record

    What it said
    A growth timeline running from 2019 to 2026, a company founding year of 2019, and page metadata that variously implied the software was spun out in 2025. Read together they described roughly seven years of history.
    What it says now
    The engine was written inside Novel Blinds Inc. in 2023 and ran that business for 3 years before Novel Systems was spun out as a separate commercial platform in 2026.

    Why

    The timeline stretched a three-year history across seven. Correcting it was right, and the current statement is the accurate one — but the only evidence available to anybody reading the page twice was that the years had moved, and a correction and a fabrication look identical from there. The narrative now has a single owning constant that every surface interpolates, and a build rule fails the build on a founding or spin-out year typed as a literal anywhere else, so it cannot drift again without somebody deciding to change it.

  4. C-003August 6, 2026Disclosure added

    The availability percentage

    What it said
    A percentage under the word 'measured', with no indication of when it was read or that reading it again would answer differently. Visitors comparing visits saw 99.95%, then 99.90%, then 99.99%, and reasonably read one figure drifting.
    What it says now
    The same live reading, with an Eastern-time stamp at the point it is rendered and a sentence stating the mechanism: the worst of the monitored checks rather than the average, re-polled roughly every sixty seconds. Where no measurement window is long enough to publish, the tile says so and prints the contractual floor, labelled as fixed by contract.

    Why

    Nothing regenerated those numbers, and this is the one item on this page where the reviewer's mechanism was wrong. Availability is polled from the external monitor at request time, and the headline is deliberately the minimum across the checks rather than the mean, because an average lets a healthy component paper over a broken one. A minimum is supposed to move. Two of the three figures he compared were not readings at all — they are the contractual SLA targets the page falls back to before a window is publishable. Three numbers from three sources, and the page gave him no way to tell them apart. Freezing the number was rejected: the previous state was a hardcoded constant captioned as a measurement that had never been taken, and its stability was exactly what made it credible and exactly what made it false.

  5. C-004August 5, 2026Corrected

    The margin floor

    What it said
    A single headline figure of 50.0%, described as the margin floor per line and enforced by the engine — stated as one fact about the whole platform.
    What it says now
    Two floors, each named with the price book it governs: 50% on manufactured lines and 28% on commercial bid work. Both are floors rather than targets — the engine refuses below them rather than silently clamping — and any surface quoting one has to say which.

    Why

    Both numbers were always correct; the headline named one of them as if it governed everything. The calculator three sections below the tile rendered the lower floor for five of the six trades, so a reader who opened a second tab disproved the headline without leaving the page. Commercial bid work is not priced like manufactured goods, and a single unqualified floor figure is wrong on this platform whichever of the two you pick. A build rule now rejects a floor percentage written as a literal outside the two modules that own them.

  6. C-005August 5, 2026Corrected

    Quote turnaround

    What it said
    '4 min — Average quote turnaround', captioned as down from a 6.2-hour manual baseline. The tile was retired, and the same 4 min returned three weeks later under the label 'Quote-to-work-order handoff'.
    What it says now
    No platform-average turnaround figure is published. The site cites measured operation latencies from the benchmark table, which time a specific call and say what the comparison models, or it says nothing.

    Why

    Neither number was produced by anything. No module owned them, no measurement generated them, and the platform has never timed a quote end to end because the clock starts with a phone call. The figure came back after being removed because the pass that removed it searched for the old label rather than the old number, which is why the replacement is a build rule that matches on the shape of the claim rather than a one-time edit.

  7. C-006August 6, 2026Checked, unchanged

    Route kilometres saved per vehicle

    What it said
    Reported by the reviewer as having shifted. It had not: 41 km per vehicle per working day has been the published value in every revision of the module that holds it.
    What it says now
    Unchanged at 41 km per vehicle per working day, now published with the scope it was measured in — dense-corridor work where stops cluster and resequencing has something to bite on — and with an audit note travelling in the exported memo saying the model constants are revised and dated.

    Why

    Checked against every commit that touches the constant, because the reviewer raised it beside four items that were real and it would have been convenient to concede this one too. What changed around it on the same day was the scope statement and the audit note, which is likely what made it read differently. It is recorded here as checked and unchanged rather than left out, because a corrections page that silently drops the item it cleared is a page whose omissions cost more than its admissions.

Novel Systems adds an entry here whenever a claim that was published changes, including when the change is a withdrawal and including when we checked a reported change and found none. If you have a saved copy of a page that contradicts the live one and it is not listed here, that is a defect in this record and we would like to know: tell us what you found.

The full decision log