Changelog

Question Library • versions carry a label, not just a number

A question's version number counted every revision the Library holds, including versions recovered from historical surveys, so the highest number was often an imported 2021 questionnaire rather than the version actually in use. Versions now carry a semantic label per programme: 1.0 is the current baseline and the versions that came before it are numbered below it. One version per question can be marked as that programme's default, and no question can have two. Version cards are titled and ordered by the label and show a Default or Imported badge, with the internal number still readable underneath. When you add a question to a survey and leave the version blank you now get the designated default, and where no default has been designated yet you get the version the most recent completed survey actually used rather than whichever number happens to be highest. A version can also be pinned by its label. The 2026 IT-Trender questionnaire has been designated the IT baseline.

Companies • a link to an absorbed company finds its successor

When one company absorbs another, the absorbed company's page used to become a dead link for anyone holding the old address. Such a link now redirects to the company that absorbed it, carrying whichever view you had open, using the record the merge itself wrote at the time. If that record is ambiguous or names a company that no longer exists, you get the dead link rather than a redirect, because being sent to the wrong company page looks like an answer where a dead end is obviously a dead end.

Analytics • breakdowns that hold together across years

A demographic breakdown took its list of categories from one version of the question but its answers from every version, so a survey wave that had renamed an answer option matched no category and disappeared from the breakdown entirely, which reads exactly like a demographic that was never asked. Renamed options are now declared as equivalent and resolved when a figure is read, which places more than a thousand respondents across four waves that had silently stopped being counted. Nothing recorded about what a respondent actually saw is rewritten. The same correction reaches the service attributes, where three renamed keys had been counted as separate series and made a fourteen attribute battery render as seventeen lines on the chart. A respondent the breakdown genuinely cannot place is now reported on every row of the metric rather than on a row that is never displayed.

Companies • 55 companies added to the register

The company register has been reconciled against the Konsulentguiden IT consulting and management consulting listings: 55 companies added, eight alternative spellings recorded as aliases so an imported name resolves to the right company, and one display name corrected. Every candidate was checked against the Norwegian company register by organisation number rather than by name, which is what makes an absence provable rather than merely unfound. Two candidates whose legal names matched exactly were rejected, because the organisation numbers now carrying those names belong to unrelated entities: a name freed by an acquisition can be taken years later by somebody else. Adding a company to the register does not place it in a survey. Roster membership is separate and nothing here creates one.

Companies • every change to a company is recorded

Twelve places in the platform could change a company and only one of them left a record, so the company audit trail read as a history of company changes while being a history of a single workflow. Every application write now records what it changed, in the same transaction as the change itself, so a company row can no longer be written without its audit entry. A change to a company's organisation number is recorded under its own action name with the old and new values in dedicated fields, because that number is what survey rosters, historical imports and analytics all resolve a company through, and it could previously be repointed from the edit form leaving no trace at all. The one deliberate silence is the weekly register sync stamping that it checked and found nothing changed, which would have buried the trail under entries saying nothing happened.

Methodology • three corrections, at revision 1.6

Three claims the methodology white paper makes about itself have been corrected, with no change to any calculation, threshold or behaviour. The principle previously called reproducible selection is now called reproducible comparison basis, which is what the paper can actually guarantee: a benchmark is reconstructable from its stated basis, rather than generated from stated criteria. The section on precision now names how much the answers vary alongside how many people answered, instead of attributing precision almost entirely to the count. And the section on verification now says plainly that automated checks demonstrate the implementation follows the adopted methodology and do not, on their own, validate the methodological choice. Updated in all four languages together.

Integrations • an issued partner key can be read back

A partner key used to exist for exactly one render: it was shown once and stored only as a fingerprint, so a key that was lost before it reached the partner had to be replaced. An issued key can now be recovered from the Integrations page. This reverses what an earlier entry on this page said about keys being unrecoverable, and it is stated here rather than left to be discovered. The key is stored encrypted and bound to the row it belongs to, so a stored value moved elsewhere refuses to open, and the fingerprint remains the only thing that authenticates a request, so there is no second way to present a key. The copy button on the Key ID has been removed at the same time. That identifier is not a key, and offering it for copying had twice sent it to a partner as though it were one.

Companies • a shorter list, and a visible wait on register sync

The company list no longer repeats each company's website under its name. It doubled the height of every row to restate something the list is neither sorted nor scanned by, and the detail page already carries it as one labelled field among the company's other facts. Searching by domain still finds the company. Both register sync surfaces now show the platform's own loading mark while they work: the per-company sync shows it from the moment the dialog opens rather than after a pause, and the bulk sweep, which makes one live register lookup for every Norwegian company, now shows that it is running instead of only greying out its button.

Localization • metric names in the language you are reading

Metric names on the dashboard and on the scores cards rendered in one fixed administrative wording, so a Norwegian page showed English titles. They now resolve to a display name for the active language on every admin surface, including the chart legend, with the service and outcome metrics translated alongside the brand ones. A metric with no translation still appears under the name it already had rather than disappearing from the page. The partner API deliberately keeps resolving its own titles, because those are part of what a partner has integrated against and are not a display choice. The Norwegian word for a score that the platform owner ruled out has also been removed from the places it still survived, including the instruction block that gets sent to a partner.

Dashboard • the leading companies, and ten entries instead of five

The admin dashboard now shows, under the activity tiles, one ranked card per brand metric with the companies currently at the head of that ranking. Each row links through to the company's own Scores tab, where the same figure appears with its precision beside it. A metric that is switched off or has never been calculated contributes no card, and a dashboard with no cards renders no section rather than an empty one. The existing top lists have been widened from five entries to ten.

Companies • Historikk withdrawn, Score covers it

The Historikk tab has been removed from the company page. Its cards, its per-wave tables and its trend graph answered questions the Score tab now answers from the same figures, so keeping both meant two places to look and two chances for them to disagree. The year selection, the per-year figures and the trend expansion all remain on Score. A past year no longer renders a rank column at all, rather than a column of dashes: a past period has no ranked set to place a company in, so a dash under a header the current view fills read as missing data instead of as something not offered. The current score keeps its rank.

Integrations • the outbound destination cards are gone

Integrations is now the partner key page and nothing else. The cards for adding and for listing outbound destinations have both been removed. They fronted a capability with nothing behind it: no part of the platform ever sent anything to a destination, and no destination had ever been saved, so a form with a working save button could only suggest a feature that did not exist. Removing the list alongside the form, because a list that can never gain a row is worse than no list at all. An error message that told you to check a URL has gone with the page that had one.

Integrations • limits a partner can plan around, and a sheet you can send them

The partner lookup API now states and enforces its own limits. A key is throttled at 60 requests a minute and at 500 distinct organisation numbers over a rolling day, and a request past either limit is refused with a response telling the caller when to try again. A card below the keys holds one plain text block a staff member copies straight into an email to a partner: the endpoint, how to authenticate, a working example, a real abridged response, what an absent score means, the publication floor, both limits and every error code. It is generated from the limits the platform actually enforces rather than written out beside them, so the two cannot drift apart. The block also tells a partner to store the last response per company and refresh once a day after 03:00 UTC, because the figures come from one nightly calculation and polling more often returns the same answer while spending both limits to get it. Keys and identifiers now have copy buttons instead of having to be selected by hand.

Settings • one floor per shape, and one card fewer

The minimum number of answers a subgroup figure must rest on is now administered as two numbers rather than one, because a breakdown with exactly two groups and one with three or more are not the same disclosure. Both are read by the calculation rather than only recorded, so the settings page and the figures agree. They share one form, one explanation and one button, but each is stored, audited and explained under its own key, so a change to one never appears in the other's history. Separately, the floor for external publication has been withdrawn as a setting of its own: what a partner receives is now governed by the same floor as results per company, which leaves one number to maintain where two were saying the same thing. The guidance text has been corrected where it still described the old arrangement.

Companies • each score broken down by sector and company size

Every row on the scores card now expands to show the same figure per respondent sector, private or public, and per company size in three bands. The publication floor is applied to each band on its own: a band resting on fewer answers than the minimum is named and marked as below the floor, while the bands above it show their score and their count. Which minimum applies depends on the shape of the breakdown, since a variable with two groups and one with three or more are governed by their own settings. Respondents who answered in a wording the current breakdown does not recognise are reported as such rather than quietly dropped, so a partial population is never presented as a whole one, and an answer of do not know or prefer not to say is treated as a non response rather than as a fourth group.

Companies • a score chart above the cards

The Scores tab now opens with a chart of how a company's scores have moved. Each metric you select gets one line across every completed survey year and then the current rolling score, with its confidence interval drawn on each point and a hover card carrying the same row you would read in the table below. You choose what to plot with a checkbox per card and per metric. A period with no published figure breaks the line rather than being drawn through, because a straight line across a gap asserts a trend over a period whose figure was never published. The axis carries its values on every gridline and never spans less than ten points, so a small movement cannot be made to look like a large one. The chart shows the same figures as the cards beneath it and is bound by the same publication rules, so it can never show something a card withholds, or show it at finer precision.

Subgroup publication floor for variables with three or more groups

The same floor for a demographic variable broken into three or more groups, such as company size, is now a second and independent setting. Both values were seeded from the single number that governed every subgroup before this change, so nothing moved when it shipped. They can now be moved apart, and which one applies follows from how many groups the variable reports rather than from anything about the company. The value is set on Settings, is recorded in the change history with who changed it and why, and is applied by the calculation immediately.

Technical focus: app_settings.p5_min_subgroup_n_3plus, migration 103, governing decision 7v

Subgroup publication floor for variables with two groups

The minimum number of responses a subgroup must carry before its result is shown is now set separately for a demographic variable with exactly two groups, such as sector. A two-group breakdown leaks more per published band than a wider one: publish one band and withhold the other, and the withheld band follows by subtraction from the total on the same screen. That case is now administered on its own rather than sharing one number with every other shape. The value is set on Settings, is recorded in the change history with who changed it and why, and is applied by the calculation immediately.

Technical focus: app_settings.p5_min_subgroup_n_2, migration 103, governing decision 7v

Integrations • look a company up by organisation number

An approved partner can now ask SRVY about one company by its Norwegian organisation number and get back that company's display name, registered legal name, organisation number, website and its calculated scores, grouped the same way the Scores tab groups them. Access is per partner: you issue a key under Integrations, the key is shown once and stored only as a fingerprint, and you can revoke it at any time without losing the record of what it was. Every response carries the publication rule it was produced under, so the recipient can see the minimum number of answers a figure had to rest on. A figure resting on fewer answers than that is left out rather than sent, and a figure that was left out looks exactly like one that has no data behind it, so nothing about the smaller companies can be read from what is missing. Ranks are never included. Until a minimum is set under Settings, no scores are published at all.

Localization • every admin page in your language, and no lost place

The last English-only admin page has been translated, so Settings and its threshold cards now read in the same language as the rest of the portal. Switching language keeps you exactly where you were, including which tab you had open, instead of dropping you back on the first view of the page. Dates are written the way the active language writes them rather than in one fixed format, and the Norwegian wording for a score was corrected throughout.

Methodology • export the white paper as a PDF

The methodology document can now be exported as a proper PDF. SRVY renders it in the background rather than in your browser, so the result is the full document with its figures and tables intact, and the page shows a progress state while it works and starts the download on its own when it is ready. Recent exports are listed underneath with their state, so a slow one can be picked up later and a failed one retried. Saving the document straight from the browser print dialog is still offered as the quick alternative.

Companies • a Scores tab on every company

A company now has its own Scores tab beside Details and Users. It groups the calculated results into three cards, Brand and position, Service satisfaction and Outcomes, and each row shows the score, its 95 per cent confidence interval, how many answers it rests on and where the company ranks among those measured. The figures come from the daily calculation, and the tab says when that last ran. Reading it requires the right access, and a company with nothing calculated yet says so plainly instead of showing an empty table.

Legal • what the benchmark is, and who owns which part of it

The public legal pages now describe the benchmark the way it actually works. They say who conducts the benchmark surveys, separate ownership of the methodology from ownership of the questionnaires and editorial content, and state that the resulting dataset is jointly managed through the cooperation behind SRVY while customers get access under subscription rather than ownership. The data processing agreement is scoped so it cannot be read as covering the benchmark surveys, a benchmark respondent has a named address for a data rights request, and the Question Library is described as a system separate from the content stored in it. All four languages were updated together.

Settings • the disclosure thresholds, visible and audited

The rules that decide when a figure is too thin to publish are now administrable from Settings instead of living in code. Each threshold has its own card stating the minimum in plain language, a short guide reading back what the current value means for the figures it governs, and a field you can change. A value below the ratified floor is refused in the page's own words rather than quietly accepted, and every change is recorded with who made it, when, and the reason they gave, so the history sits under the setting it belongs to. A read-only sample quality guide sits beside the cards it documents.

Sharing • a proper preview card for every SRVY link

A link to any SRVY address now unfurls properly when it is pasted into Slack, Teams, LinkedIn or a message, with a title, a description and an image instead of a bare URL. Each surface carries its own card, so a link to the public site, the admin portal, Insight and the beta preview are told apart at a glance. Preview fetchers are answered with real directives rather than being bounced into a login page, which is what stopped the cards rendering before.