0.9.0-alpha

v0.9.0-alpha

Released September 28, 2026

Added

Workforce labor analytics now report the average loaded compensation, the employer benefit cost total, the loaded-cost coverage share and basis, and a daily capacity utilization status.
The labor analytics gain the employer statutory cost coverage share, the disclosure of how much of the headcount the modelled statutory cost total covers, so that partial sum is always read beside its coverage.
An address that does not exist now shows a QortexOS page in your language with a way back to the home page, instead of a bare browser-style error.
On Department Setup, you can now search your departments by name. The search reaches every department, including those beyond the first 100 the page loads.
On the Contract Intake screen, every uploaded file can now be downloaded from the Uploaded files table; when a download is not possible the row says why, for example that the file has not passed its virus scan yet or is under a legal hold.
On the Budget Center, the Rolling Forecast card says when the forecast was refreshed if that happened this month, and the Patch Sets card, where available, shows how many patch sets are pending.
The Data Rules Engine can now find policies by name and filter them by enabled state and by schedule, beside the type.
In GDPR Compliance, the data policy now also shows the export record limit and the allowed export formats.
The Profit Center lets users who manage profit optimization adopt or reject each pending recommendation.
Revenue Forecast can now narrow its table to the days the forecasts were generated, with a from and a to date beside the revenue stream.
Exchange rates are now stored as dated, versioned sets that are never edited: a correction arrives as a new version, so any converted figure can still be explained by the rates it was computed from.
Contract timelines now show resolved findings and material health score changes detected by the analytics platform.
On the Connector Health screen, you can now search connectors by name or last error, and filter them by sync state and by whether they have errors.
The Rolling Forecast adds Refresh data, which recomputes the forecast figures, for users who can update budgets.

Changed

Four create requests of the API (signature envelopes, prompt execution feedback, tenant subscriptions and gold standard examples) no longer accept a tenant id: the tenant always comes from the signed-in session, and a tenant id that is still sent is ignored. Creating a signature envelope through the API now succeeds; it failed at the database before.
Security and system notifications can no longer be switched off for a whole workspace.
Adding an integration now stops at the workspace's integration limit; reconnecting an existing integration is never blocked by it.
Inviting users now stops at the workspace's user limit, counting pending invitations, and an invitation accepted after the limit is reached is refused with a message to ask the administrator.
The assumptions behind an acquisition recommendation are now written down in one place each: what counts as strong synergies, what counts as fair and as overpriced, what financing score counts as feasible, and what client retention the platform assumes survives an integration. No recommendation changes; the numbers simply stop being scattered.
In Finance, a refused action now says why when you can do something about it: a GL code whose allocations would pass 100%, a budget or a forecast run that changed since the page loaded, a file type or size the upload does not take, a file that has not passed the virus check, a mapped column missing from the file, a change that was refused, or too many attempts.
In Tenant Settings, a refused action now says why when you can do something about it: an address that already belongs to a member or already has an invitation, an invitation that changed since the page loaded, a deactivated member, the last administrator or access level, a department name already in use, a change that was refused, or too many attempts.
On Tenant Configuration, the Limits step now proposes 250 users and 25 integrations for a tenant whose limits are not saved yet.

Improved

The Extraction Workbench now asks for a field's correction history newest-decision first, so a field decided more than once keeps reading the reason from its latest decision.

Fixed

The manual refresh trigger for a materialized view marks the view for refresh again.
Benchmark intelligence reports now recognise your strengths: a top-quartile peer percentile is reported as a competitive advantage and your overall position is ranked on the right scale, instead of every metric reading as an improvement opportunity and the position always showing as lagging.
Signing in with an email address that has not been verified yet now offers a link to send a new verification email, a server error during sign-in no longer tells you to check your connection, and the two-factor steps no longer call every failure a wrong code or an expired code a server error.
Benchmark Enrollment now shows a tenant's revenue, employee, client and endpoint snapshot as soon as a cohort is assigned, with the currency the revenue was recorded in.
A sign-up that is refunded because it could not be completed is also unsubscribed, so it is not billed again.
After a rate limit, the sign-in and sign-up screens can now tell you how long to wait.
Revoking a sign-up invite from the staff screen now takes effect.
A refund of your sign-up payment also ends the subscription, so you are not charged again.
You stay signed in while you work: the session now ends after 30 minutes without activity (12 hours at most), not 30 minutes after sign-in, and a session that ended sends you to sign in again.
When your session ends, the portal now takes you to the sign-in page, tells you that you were signed out, and returns you to the page you were on once you sign in again. A temporary server problem no longer sends you to sign in.
Deactivating a user now ends their sessions under every way their id can be written, and tells you when the sessions could not be ended, so you can try again, where it could report success with a session still live.
A deactivated user is signed out everywhere at once, including a session opened in a second tab or a request that was under way when they were deactivated.
When a page fails to load, the Try again button now asks the server again, so it can recover from a temporary problem without reloading the browser tab.
Hiring requisition analytics now cover every requisition of the tenant, report a real average time to fill per skill category, and count only live requisitions as active.
On Subscription Management, a problem loading your default payment method no longer hides your plan, status and renewal date: only that one card reads Unavailable.
In Account Settings, a save that fails now explains what happened in your own language, instead of showing technical text such as "Bad Request".
On Department Setup, the Hierarchy depth card now counts the levels across all of your departments, not only the ones shown on the page.
On Department Setup, a department shown at the top level because its parent is inactive or not in the loaded list now says so under its name.
Saving a department under a name that is already in use, or a budget whose fiscal year or date windows are not valid, is now answered with a specific reason instead of a general server error.
On Contract Intake, a failed upload or a failed Record intake now explains what went wrong in your language, and the Recently recorded list says when it could not load instead of looking empty.
On Ingestion Flow, a pipeline run that fails to start or finish now explains itself in your language instead of showing the server's English error text.
Changes to a user's roles, to feature access, to the navigation menu and to settings, users and roles now reach everyone within minutes; before, the previous state could sometimes stay for up to an hour.
On the Extraction Workbench, the Correct form no longer offers a Save that the server always refuses, and a decision that cannot be saved now explains itself in your language instead of showing the server's English error text.
On the Extraction Workbench, when contract names or decision reasons cannot be loaded, the queue now says so and offers a retry, instead of showing a dash or a bare status code as if there were nothing to show.
On Contract Intake, the Recently recorded list now shows only contracts recorded through that screen, not contracts created from a template or imported elsewhere.
The one-time organization setup page now opens its form for an organization that has no details on file yet, asking you to choose your currency, language and time zone, instead of showing a loading error.
On the Ingestion Flow, a person who may view pipeline runs but not start them no longer reads an instruction to pick a contract and run the pipeline; the page now says that they do not have permission to run it.
On Contract Intake, the Uploaded files table now has pages, a page-size choice, sorting by file, size and upload time, a search box and filters for virus scan and processing status, so files past the first 25 can be found and recorded; an open Record intake form is kept while you change pages.
On the Ingestion Flow, the Recent runs list now has pages and two filters, by run status and for the contract on screen only, so runs older than the newest 25 can be found and opened; the run you picked stays on screen while you change pages.
On the Ingestion Flow, the Changed since panel now says when its changes could not be loaded and offers a retry instead of reading as no changes, names each change by its type in your language instead of a code, and says when a version has more changes than the panel lists.
On Contract Intake, when some files of a selection do not upload, the upload area now keeps each of those files listed by name with the reason, instead of showing only the last file's result; and files added while an upload is running wait their turn instead of starting a second upload beside the first.
On the Ingestion Flow screen, a run's contract is now always shown by its title in the run summary, even for older contracts; where a title is not available in the recent runs list the screen says so and labels the value as a contract ID instead of showing a bare number, and the empty screen no longer says that no contract is recorded while the contracts are still loading or could not be loaded.
When one figure in a Data Platform KPI strip cannot be loaded, that card now says so and offers a retry, and the figures beside it stay on screen.
When a background refresh fails on Department Setup, the departments and counts already on screen stay there instead of being replaced by an error.
On the Contract Intake screen, the virus scan, OCR and processing statuses of uploaded files now update by themselves while a file is still being processed, so Record intake appears without reloading the page. A Refresh button re-reads the list at any time, and when an update fails the screen says that the statuses may be out of date.
On the Contract Intake screen, Word 97-2003 files (.doc) are no longer accepted for upload, because the document reader cannot read them; the screen says so and asks for the file to be saved as .docx instead of accepting an upload that could never be processed.
On seven console summary strips, figures already on screen now stay there when a background refresh fails, instead of being replaced by an error message and a retry button.
A feature-flag catalog row that targets a cohort an older app build does not know no longer switches off every flagged screen on that build; only that row stays off on that build (a tenant's own override still applies) until the build catches up.
Changing your plan or details during sign-up no longer ends with a message that an account with this email already exists.
Sidebar links now follow the current navigation catalog: retired gates no longer hide or expose items.
A paid sign-up no longer gets stuck when another company with the same name already has an account.
Changing the workspace time zone in Tenant Configuration no longer resets the saving administrator's own language and time zone, and a time zone the platform does not list is refused.
Account Settings labels your theme preference as the saved theme, since the theme in effect can differ on each device.
On At-Risk Signals, the account list now pages through every analyzed account, 100 at a time, in the whole ranking and within each risk band, instead of stopping at the first 100.
On At-Risk Signals, the evidence panel's driver chart now draws each driver's contribution as a signed bar from zero, in relative units, instead of a percentage waterfall whose driver steps drew flat.
On At-Risk Signals, with no account picked, the list you are looking at now selects its first account, under a risk-band filter or on a later page, and the evidence panel explains that account instead of the top account of the whole ranking.
On the Platform Maturity screen, the platform views needing refresh card now counts every view in the refresh registry, instead of stopping at the first 100, and warns when any of them is failing.
On the Platform Maturity screen, the data quality score now averages every quality report in its 30-day window, weighting each data layer by how many reports it had, and the card says what the score covers.
On the Budget Center, the Guardrails card no longer keeps its warning colour for variance events that no screen can mark reviewed yet.
In Budget Setup, a mapping or line that cannot be removed now says so; Remove used to leave the row in place with no message.
In Budget Setup, the fiscal year field now takes only the years a budget can be created for, 2020 to 2050, and a year that already has a budget now says where to find it.
In Budget Setup, the header now says that each step saves when you add or submit it, instead of saying that changes save automatically.
Budget Setup now opens for people who may only view budgets: they can walk its four steps, and the controls that change a budget appear only for people allowed to use them.
In Budget Setup, the account mapping step now shows whether each mapping is active or inactive.
In Budget Setup, the account pickers on the mapping and lines steps now offer every postable account in the chart of accounts; accounts past the first 200 were missing from them.
On the Cash Center, the portfolio line now says which day its figures are from and that they are in your company's currency.
On the Cash Center, the date an alert was raised and the date an insight was generated now read in your own time zone, so a late-evening alert no longer shows the next day.
On the Connector Console, the health strip now shows the success rate across all syncs, which agrees with the sync counts beside it, names the 30-day window each figure covers, covers every connector, and drops the uptime tile that repeated the success rate.
On the Data Platform Center, the attention band shows its Pipeline Console link only to readers who can open that screen; every reader still sees how many lakehouse layers are below their health threshold.
On the Data Platform overview, the sync success rate and the quality score now say they cover the last 30 days.
The Data Rules Engine's violations card now says it counts policies that have found violations at any time, and no longer shows that count as a warning.
When two reviewers decide the same extracted field at the same moment, the second is now told the field was already decided instead of seeing a generic save error.
On Feature Store, the link under the recently updated pipelines now opens the Orchestrator, where feature pipelines are listed, instead of the Performance Console, which lists none.
In GDPR Compliance, a retention record the anonymization run has processed now reads Anonymization: Completed instead of Required.
On the Hiring Center, the Open positions card no longer shows a warning just because positions are open; it is marked only when its figures could not be loaded.
The Industry Stress console lists the insights of the analysis its figures quote, without a repeat per analysis day.
On Model Experiments, an experiment awaiting a decision now offers Record decision: choose Promote treatment, Keep control or Extend test and give a rationale, and the rationale shows under the recommendation.
In POR Import, a file still being checked for viruses no longer fails validation at once: the Upload step says the check is running, and the validation waits up to about a minute for it.
In POR Import, the account column's field now says it takes the chart of accounts system id, not the account code or your GL code.
In POR Import, the checks table now says what each check points at (a CSV line, an account, a GL mapping or the whole run) instead of heading every reference as a row.
In POR Import, the page head's exit now reads "Exit to Validation Workbench": it never saved the wizard's progress, and no longer says it does.
In the Payment Terms Optimizer, a comparison figure that did not change now reads as unchanged instead of being shown in green as an improvement.
On the Privacy Budget Console, New operation creates a differential-privacy operation with its epsilon budget and, optionally, its ad-hoc reserve, automated query cap and next reset date.
The Revenue Center says when an amount is shown in the tenant currency, and shows MRR in its own recorded currency once one is recorded.
In the Revenue Forecast, a KPI card with no reading now shows 'Not available' in the quieter absence style instead of the large figure style, so it can no longer be mistaken for a value.
In German, the Revenue Forecast's confirmation for cancelling a batch run now labels the button that keeps the run going as such, instead of a word that read like discarding the run.
Revenue Forecast's table can now be sorted by the point forecast, largest first on the first click.
Revenue Forecast's table now shows each forecast's point forecast, the figure its error is measured against, beside the P10, P50 and P90.
Revenue Forecast's detail card now shows the model accuracy, and each forecast's error and bias once its actual is recorded.
The Service Catalog's header now names the day of the catalog statistics as well as the day of the queue performance figures.
A benchmark category you leave unchecked in the enrollment consent step now stays out of newly computed benchmark peer numbers: your metrics of that category are neither contributed to nor ranked against the benchmark peer aggregates, and unchecking any category also takes your data out of the training set of the cross-MSP strategic benchmarking model.
On the Service Catalog, clients without a tiered offer are listed as No tier, after the named tiers.
Each workspace can now send up to 100 employee invitation emails an hour, resends included; past that, invitations and resends are refused until the hour is over.
On User Management, a member's roles now list the access level first with a +N control that opens the full list, the invitations list says when it shows only the newest invitations, and internal staff roles no longer show on a member's row.
On the Validation Workbench, saving a decision now keeps the confirmation on screen, announces it to screen readers and keeps the keyboard focus on it.
On the Validation Workbench, a decision refused because the check changed elsewhere now refreshes the check, and decision errors are shown in your language.
Account Settings now tells you when you need the update permission to change your preferences.
Account Settings now says when the time zone list could not be loaded, instead of reporting that no time zone matches your search.
Time zone pickers now show each zone's current UTC offset, daylight saving time included.
On At-Risk Signals, revenue at risk is stated in each account's own currency when the analysis records one, the total is never summed across currencies, and a note says when amounts are shown in your tenant currency.
On At-Risk Signals, the evidence panel's contributing factors now list only the factors the analysis scores; the price sensitivity and growth stage context rows no longer read as drivers.
On At-Risk Signals, the KPI tiles now say that their figures cover scored accounts, beside the count of unscored clients.
On At-Risk Signals, before a tenant's first churn analysis has run, every KPI tile now explains beneath its value that no accounts have been analyzed yet.
On the Budget Center, a health card with no value now says so in plain text instead of in the size and style of a figure.
On the Budget Center, the plan-of-record card keeps the fiscal year when a budget carries a status the page does not recognise.
On the Budget Center, the data-through line names the month the figures cover, for example "Data through January 2026", instead of that month's first day.
On Budget Setup, the Refresh KPIs button now appears only for people allowed to refresh the figures.
On Budget Setup, the chart of accounts mapping step no longer labels its counts with a currency.
On Budget Setup, the planned gross margin tile says how many months its figure covers when the plan does not yet fill the year.
On Budget Setup, the review step now says that Refresh KPIs recomputes every budget in your organization, not only the one on screen.
On Budget Setup, the planned gross margin tile writes its percentage the same way as the KPI table beneath it, in every language.
On Budget Setup, the COA coverage row of the budget card now says that it counts your whole organization's mappings.
On the Data Rules Engine, people who can view policies but not run or change them no longer see an empty Actions column.
On the Data Rules Engine, the line confirming a run, pause or resume now clears when you start the next action or change the filters, search, sort, page or page size.
On the Data Rules Engine in German, the page and browser tab title now read "Regel-Engine".
On Department Setup, a failed reactivation now shows under the department it failed on.
On Department Setup, a parent or a deleted department is always named, never shown as a long number.
On Department Setup, a department on the tenth level of the hierarchy is no longer offered as a parent, and the screen says why.
On the Industry Stress Console, Clear filters shows only while a severity filter is set.
On the Industry Stress Console, a KPI tile that shows Not available now says why when the reason is known: no churn analysis yet, or the tenant currency not set.
The Industry Stress Console now says under its table that external sector intelligence is not available yet.
On the Industry Stress Console, a KPI tile with no figure now reads as an absence, not as a number.
On the Industry Stress Console, an empty table no longer blames a severity filter that was never applied.
On the Industry Stress Console, the note that amounts are in the tenant currency now sits under the KPI tiles it qualifies.
On the Industry Stress Console, the Recorded date now reads in your profile's time zone.
On the Industry Stress Console, the Category and Severity columns no longer offer a sort that ordered by code, and the severity filter lists the severities from most to least severe.
On the K-Anonymity Console, the line saying a re-run was recorded now clears when you change the page, sort, band or search, or open a Re-run again. The Re-run confirmation is announced with its question, focus moves to its Cancel when it opens, and back to the row's Re-run button when it closes while that row is still on the page.
On Payment Terms Optimizer in German, the recommendation band now names the early-payment discount and the net benefit as the rest of the page does.
On Payment Terms Optimizer, sorting the recommendation table by annual revenue now starts with the largest client, as the other figure columns do.
On Payment Terms Optimizer, the comparison cards now show a plus sign on a money change that rises, as the day changes beside them do.
On Payment Terms Optimizer, each scenario on the net benefit line now shows whether it meets its own DSO target, and the Recommended badge goes to the highest net benefit that does.
On Payment Terms Optimizer, the live DSO trend no longer draws the illustrative DSO target, and its scale now follows the trend alone.
On Revenue Forecast, Clear filters now appears only while a filter is set.
On Revenue Forecast, a link with an unknown revenue stream now shows the unfiltered empty message, matching the unfiltered list it shows.
Revenue Forecast's Forecast Runs list now says when it shows only the most recent runs.
Revenue Forecast's table columns no longer shift when the rows load.
On Revenue Forecast, the Business value card now says its amount is in the tenant currency.
On Revenue Forecast, the validated forecasts card now says it counts all time.
On Runway Planner, the stress-scenario definitions now read in your language, named by their scenario type.
On Runway Planner, the intelligence narrative now always belongs to the forecast whose facts it sits beside.
On Runway Planner in German, the hiring-pace unit now reads FTE, as the rest of the page does.
On Runway Planner, the stress test's net cash flow impact and maximum financing need now say they cover 91 days.
On Runway Planner, the comparison table's 18-month row now says it shows the cumulative growth cash flow.
On the Service Catalog, a figure that could not be loaded now says so, with a button to try again, instead of looking like missing data.
On the Service Catalog, the attention band links to the Service Lifecycle console only when you can open it.
On the Service Catalog, the Coverage & Tier Mix card lists your tiers by name, in the same order every day.
On the Service Catalog, the portfolio line and the Lifecycle card name one offer, mapping or queue in the singular.
On User Management, the page readout that a narrow team table footer shows in place of the page numbers now reads in your language.
On User Management, a KPI card whose figure could not be loaded now says so, and the strip offers a retry, instead of reading as not available.
On Team members, when an action on a member fails, the message now appears directly beneath that member's row.
On User Management, the confirmation line an action leaves behind (such as "Cleo Marsh reactivated.") now clears when you start your next action, instead of staying on screen until the page reloads.
On User Management, when the window is too narrow for the full layout, invitations now open in a side panel from the envelope button at the bottom right, and the Team members table leads the page.
On Team members, the Deactivate confirmation now opens directly beneath the member's row, and the Deactivate button tells screen readers which confirmation it controls; the same holds for revoking an invitation.
On User Management, a pending invitation's expiry date now reads in your time zone, so an invitation that expires in your evening no longer shows the next day's date.
On User Management, Active accounts now counts every active member, and Pending invitations and Roles in use now say which records they count once their list runs past 100.
On User Management, you can now sort the Team members table by name, email or last sign-in, search it by name or email, and filter it by status or role; the view you set stays in the page address, so a reload or a shared link opens it again.
Signing in with an account whose setup was never finished now explains that clearly and points to support, instead of showing a generic error.
On the Data Platform's pipeline, CDC, backfill, orchestration and data rules consoles, Backfill Setup and the connector screens, a failed action now names what it failed on in your language (the stream, backfill window, view, pipeline, policy, connector or provider), and on the connector screens a refusal from the connector service keeps its reason after that sentence; it used to print the server's English text alone.
In Finance, a failed action in Budget Setup, the POR import, the Validation Workbench and the forecast runs now names what it failed on in your language (the fiscal year, month, GL code, file or run); it used to print the server's English text.
On Renewals, Service Lifecycle, the Vendor Scorecard, the HRIS Sync conflict queue and Benchmark Enrollment, a failed action now names what it failed on in your language (the contract, service offer, vendor and its new status, conflicted field, cohort or contribution); it used to print the server's English text.
On the Federated, K-Anonymity and Privacy Budget consoles, Model Experiments, the Serving Console and Shadow Deployments, a failed action now names what it failed on in your language (the session and its round, the dataset, the operation, the experiment and its new status, the model version, the alert or the shadow run); it used to print the server's English text.
In Tenant Settings, a failed action now names what it failed on in your language (the member, invitation, role, department, retention schedule, billing portal session or settings group); it used to print the server's English text.
On Contract Intake, pressing Tab in the upload area now lands on Browse, instead of first stopping on a hidden file control with no label.
A data connector whose health has not been measured yet is reported as unmeasured rather than critical, and its availability is no longer reported against an uptime nobody measured.
A zero you send or store is now used as the number you meant. Where a zero would leave a run with nothing to do, the platform says so instead of quietly substituting its own figure.
Rates now apply at the rate you set them. A tax rate of 26.25 percent, an adoption probability of 5.5 percent or a growth target of 7.5 percent is carried through the money it multiplies instead of being cut back to a whole percentage point.
Computed money now rounds to the nearest cent, the same way the database does, instead of being cut downward. Forecast bands, loan schedules, price recommendations and risk figures no longer drift a cent low each time they are computed.
Averages, counts and verdicts across forecasting, quality, retention and workforce analytics are now taken over the rows the platform has actually measured, so one unmeasured row no longer drags a whole average down or decides a verdict, and the assumptions the platform does apply are named in one place each.
A payout preview now shows the same figures as the payout run it previews, to the same precision the platform records, so a leader who reproduces a payout from the preview reaches the number on the record.
Platform deployments are unblocked: two database changes that had been prepared out of order are re-issued in order, so the queued changes behind them apply again.
A model type's first evaluation is reported as having no baseline to compare against, instead of being compared against a score of zero that nobody measured.
Shadow-deployment analytics report a model type's validation confidence over the runs that were actually scored, so one unscored run no longer drags a whole model family's figure down.
In the Profitability Benchmark header, figures that could not be loaded now say so and offer the screen's retry, instead of reading as a tenant with no benchmark records.
Budget Setup and Benchmark Enrollment now wait for your permissions to load before asking the server for their data, so neither screen can read as empty while that answer is still on its way.
Optimization analytics report each figure over the periods and analyses that were actually measured, and a solve with no recorded time is no longer counted as one that finished instantly.
Model serving analytics and the model portfolio drift score are now averaged over the snapshots and models that actually carry each reading, so a snapshot the serving telemetry landed without a metric no longer halves the reported utilization, latency or error rate, and a model whose drift was never scored no longer reports the portfolio as drift-free. A portfolio with no evaluated model now reports no accuracy, no industry percentile and no maturity score, instead of being recorded as a mid-range performer scored at zero accuracy.
Model portfolio analytics now read only the figures an analysis actually recorded, so a portfolio whose ROI, automation rate or maturity was never measured is no longer reported as zero, is no longer placed at the bottom of its industry, and no longer receives advice to fix a number the platform never took. The analytics also read the series in the right direction: the industry benchmark position now comes from the most recent analysis that measured one instead of the oldest, and the automation and incident-resolution trends no longer report an improvement as a decline.
The average performance improvement reported for model retraining now counts only the retrainings whose improvement was actually measured, so a model family with an unscored retraining is no longer reported as having improved less than it did.
Document processing reports now read only the figures the platform actually recorded: a document whose pages were never scored is no longer listed as low confidence, and a job whose processing time was never recorded no longer pulls the reported average speed down.
Document processing metrics now report the jobs that failed. The failure count read zero and the success rate read one hundred percent no matter how many documents had failed, because the report only ever looked at the jobs that succeeded.
Budget KPI analytics now average only the months that recorded each figure, so a month with no margin, revenue or EBITDA to report no longer drags those averages down or invents a margin trend.
The model bias analytics no longer read a detection whose demographic parity was never scored as a score of zero: each model type now averages only its scored detections, and a type or a trend point with no score reports none instead of the most biased value.
A failed refresh on the Platform Maturity, K-Anonymity Console, Model Experiments and Shadow Deployments screens now keeps the entries already listed instead of replacing them with an error message.
A failed refresh on the Rolling Forecast, Covenants, Serving Console and Subscription Management screens now keeps the entries already listed instead of replacing them with an error message.
The payout explanation now prints the same figures the payout run records, so a leader who reproduces the calculation from the steps reaches the number on the record.
An acquisition price split into a down payment, a loan and a seller note now adds up to the price exactly. The three shares used to total a cent more than the price they divide, so a buyer was told to fund a cent that did not exist.
An average of money amounts now rounds the half cent the way the rest of the platform does, instead of always dropping it, so a reported average no longer sits a cent low.
On the Platform Maturity screen, the list of recent intelligence snapshots now stays on screen when a background refresh fails, instead of being replaced by an error message and a retry button.
Retention averages - SHAP explanation quality, feature importance by category and model C-index - are now taken over the records that actually carry a score, so a record an explainer or an evaluation never scored no longer counts as a zero and drags the average down.
QBR analytics averages - prediction accuracy, meeting effectiveness, client engagement, pathfinder scores and business value - are now taken over the reviews that actually carry a score, so a review nobody scored on a measure no longer counts as a zero and drags that measure down for the whole set.
Workload analytics now average over the records that actually carry a figure. A model nobody has evaluated, a department nobody has measured and a forecast comparison nobody has scored no longer enter those averages as a zero, so reported error, utilization and pattern strength describe what was measured.
A model, feature or privacy operation that has never been measured is no longer judged as though it scored a perfect zero. A model nobody has evaluated no longer counts as meeting its accuracy target, a feature with no drift measurement is flagged for attention, and a privacy operation with no recorded budget use is treated as having no budget left.
Client and QBR analytics now report how many records their figures were computed over, beside how many records matched, so a figure never describes more records than were actually read.
Amounts inside analysis text, warnings and scenario names now appear in the tenant's own currency and the reader's own locale, with the number of decimal places the currency itself uses, instead of a dollar sign and two decimal places written into each sentence.
When a compensation guardrail is breached, the payout explanation now shows the measured value and the threshold to the same six decimal places the platform records, so a narrow breach no longer appears as two identical numbers with no visible cause.
The RMM telemetry correlation analytics now report, for each skill category, the automation opportunities recorded against that category's correlations, where every category used to report zero.
On the Validation Workbench, a budget chosen in the budget picker now follows the budget list, so a budget locked after you chose it shows as locked once the list refreshes (for example after a refused decision).
In Budget Setup, a budget that was locked while you worked on its lines now shows as locked as soon as a change to its lines is refused, instead of still offering the draft's editing.
In Budget Setup, adding lines to a budget that was locked meanwhile now says so in your language, as removing a line already did.
At sign-up, the payment step's card form and a declined card's message now follow the portal's language setting instead of the browser's.
On the Renewal Console, marking a contract Renewed now announces it by name once the renewal is saved; the announcement never showed before.
Dates, numbers and the payment form now follow the language of the page's text. A browser set to a language the portal does not offer yet sees them in English, like the text, and one that lists German after such a language now gets German.
Tenant Configuration can now be set up on an account that has no settings yet: the first save creates them, where the page used to say they were not provisioned.
The GDPR Compliance erasure queue and the Drafting Studio comments zone no longer show sample records: each now says that none are recorded yet.
On the K-Anonymity Console, Re-run now judges a validation run in the run's own privacy tier, with the generalization evidence the run recorded, instead of the standard tier. A failed Re-run's message is now tied to the run it belongs to and clears when you change the page, sort, band or search. An empty table no longer says filters are applied when the page address names a k-value band the console does not recognize. While a band or search filters the table, a line under the figures above it says they count every validation run.
On Payment Terms Optimizer, the intelligence narrative in the reasoning panel now always belongs to the forecast whose facts it sits beside.
When two reviewers decide the same budget validation check at the same moment, the second is now told the check was decided elsewhere instead of seeing a generic save error, and a decision can no longer slip through while its budget is being locked.
The service profitability view's 30-day average cost per ticket is now a true trailing 30-day average. The 30-day revenue and margin averages stay empty until daily contract-value history is available, instead of repeating the same-day figure.
On Contract Intake, the file size in the uploaded-files table now reads in the reader's own number format, so a German reader sees "2,0 MB" where an English reader sees "2.0 MB".
On Contract Intake, a file with no content is now refused at the upload area with its own sentence instead of a failed upload, and Record intake now names the field it still needs (the contract title or the counterparty) instead of doing nothing when you submit.
On Ingestion Flow, the Fields tile now shows how many fields the document version actually has (it used to show only what one page held) and says what the review figure covers when there are more fields than one page; a single field now reads "1 needs review"; and re-running a contract now refreshes the version labelling, so the run band names the new version and its source file instead of showing a raw id.
The Ingestion Flow now shows an unfamiliar run status as its plain code instead of failing to render the run.
In the Extraction Workbench, a decided field now names its outcome in words, so the queue reads "was Corrected · reason OCR misread" instead of the wire codes.
In the Extraction Workbench, a confirmed field now reads "Confirmed" on its own instead of carrying a correction reason that says nothing.
In the Extraction Workbench, every confidence figure now reads in the reader's own number format, so a German reader sees "0,72" where an English reader sees "0.72".
In the Extraction Workbench, the evidence caption no longer says an overlay is shown when none is drawn, and it leaves out the page line entirely when there is no page to name.
In the Extraction Workbench, the decision history of a field now shows the year beside the day, month and time, so entries from different years can be told apart.
The Extraction Workbench subtitle now names the review threshold the screen actually uses, in the reader's own number format.
On the Cash screen, the recent net cash flow card now calls itself by the same name throughout, instead of switching to referring to positions.
In Connector Setup, when connecting a provider fails on the way back from the provider, the message now names the provider where it is known and says what went wrong: an expired link, an approval the provider did not confirm, or a provider that could not complete the connection.
In Connector Setup, when a provider returns no organizations to choose from after you approve access, the screen now says so and lets you reconnect the provider or finish later from the connector console, instead of showing an empty list.
In Connector Setup, when a provider needs no identifiers, the step rail now marks the Identifiers step as skipped instead of showing it as done.
In Connector Setup, the page subtitle no longer says the first sync runs when you activate a connector; activation saves the connector and tests its connection.
On the Contract Center, the certification tier in the portfolio line now reads as Bronze, Silver or Gold in your language instead of a raw code.
On the Contract Center, the portfolio line no longer shows a stray "across 42 clients" when the contract count cannot be loaded.
On the Contract Center, the portfolio line now shows your current certification tier even when a newer award has been revoked or superseded.
On the Contract Center, the contract events timeline now orders and dates each event by when it happened, not by when it was recorded.
On the Contract Center, each entry in the contract events timeline now shows its event type in your language, such as Signed or Renewed, instead of an English sentence.
On the Contract Center, the portfolio line now shows the portfolio health score out of 100.
On the Contract Center, the portfolio line now reads "1 contract across 1 client" in the singular, in English and German.
On the Contract Center, the renewal band shows its Review renewals link only to people who can open the Renewal Console.
On the Contract Center, the snapshot caption in the page header now shows the date and time the portfolio snapshot was last refreshed, in your time zone, once that time is recorded; until then it shows the snapshot's date.
On Contract Pulse, the Campaigns card and Needs attention now count and list only active remediation campaigns, and the counts cover every active campaign rather than the first fifty loaded.
On Contract Pulse, remediation campaign due and target dates now follow your time zone, and the overdue chip counts calendar days there, so it agrees with the date beside it.
On Contract Pulse, the portfolio health disclosure now states that revenue exposure is not shown, and why.
On Contract Pulse, the portfolio health score now shows how it moved since the previous stored snapshot, and that snapshot's date, instead of always reading unchanged.
On Contract Pulse, the portfolio health score shows when its snapshot was last refreshed, in your time zone, once that time is recorded; until then it shows the snapshot's date.
On Contract Pulse, the portfolio health figures and the remediation and cyber readiness cards now show when their statistics were last refreshed, in your time zone.
On Contract Pulse, the posture shares and the remediation completion rate read as whole percentages, such as 63% rather than 63.0%.
On the Covenant Console, covenant alerts now show a recommended action in your language, and a minimum cash alert also shows its projected shortfall below the required minimum.
On the Covenant Console, the covenant alerts panel on narrow screens no longer repeats its Covenant alerts heading.
On the Covenant Console, the Covenant and Risk columns no longer offer a sort that ordered rows alphabetically by code; the table keeps its highest breach probability first order.
On the Covenant Console, the alert rail now lists covenant breach alerts only and is titled Covenant alerts; cash-out and large outflow alerts no longer appear there.
On the Covenant Console, a covenant type filter now sits beside the risk level filter, so the table can show one covenant type at a time.
On the Covenant Console, the first KPI card is now labeled Cumulative net cash flow, and its info explains that the figure is a running total with no opening balance, not a cash balance.
On the Covenant Console, the Recorded date now follows your time zone setting instead of your browser's, the same zone as the page's refresh time.
On the Covenant Console, a link with a filter value the console does not recognize no longer reports that no covenant risks match the current filters, since no filter was applied.
The Demand Forecast recommended-actions panel keeps its title and card while it loads, when it is empty, and when it fails to load.
The Demand Forecast Console's weeks-above-target count now reflects calendar weeks, and says at least when the run is longer than the history it reads.
The Demand Forecast screen no longer offers Clear filters, or says no forecasts match the filters, when no skill filter is applied.
The Demand Forecast Console's accuracy tile now says a figure could not be loaded, and offers a retry, instead of showing it as not available when one of its reads fails.
The Demand Forecast table no longer repeats a "Baseline" label in every row.
The Demand Forecast Console's Status tile now shows the retraining recommendation, and the accuracy tile names the model version it measures.
The Demand Forecast table now lists the current forecast only, so its deepest-deficit ranking shows today's staffing gaps rather than past forecasts.
The Demand Forecast table's trailing error column no longer shows a dash when its data failed to load; it says so and offers a retry.
The Demand Forecast screen now flags a skill's trailing forecast error of 15 percent or more whether the model ran light or heavy.
In the Drafting Studio on a narrow screen or pane, the document outline with the guardrail policy and the review panel now open from buttons at the bottom right, instead of being hidden with no way to reach them.
In the Drafting Studio, the guardrail policy zone now finds the statement-of-work policy profile however many profiles the tenant has, instead of reporting none when it sorted past the first fifty.
In the Drafting Studio, each guardrail policy rule now shows its description and the limits it sets: the maximum drift, the minimum risk rating, the required locale pattern and whether it links a clause library entry.
In the Drafting Studio, the note under a long list of policy rules now counts the rules loaded and says that only the active ones are listed.
In the Drafting Studio, a policy profile whose first page of rules holds no active rule now says more rules exist past that page, instead of stating the profile has no active rules.
In the Drafting Studio, the policy rules list is now in alphabetical order of the clause type names shown in your language.
On five more summary strips, across Benchmark Center, Platform Maturity, Hiring Center, the License Utilization lens and Compensation Center, figures already on screen now stay there when a background refresh fails, instead of being replaced by a failure message and a retry button.
On two more console summary strips, the Quarantine Workbench header and the Strategy Center health strip, figures already on screen now stay there when a background refresh fails, instead of being replaced by an error message and a retry button.
On the Model Serving console, the note saying telemetry could not be loaded no longer appears while the deployment tiles are still showing that telemetry; it now appears only when no telemetry was loaded at all.
On the Orchestrator, a feature pipeline with one generated feature now reads "1 feature" rather than "1 features".
The Orchestrator now says that its materialized-view refresh queue and counts are platform-wide, shared by every organization, and that its feature pipelines are the organization's own.
On the Orchestrator, the Priority column and input now say that 1 refreshes first and that priority runs from 1 to 1000, and the Configure strip refuses a priority above 1000.
On the Orchestrator, Refresh now also reloads the feature pipelines, so one click refreshes the whole page.
On the Orchestrator, the refresh queue now shows every reader each view's minimum refresh interval, maximum staleness and refresh strategy.
On the Orchestrator, a fifth KPI card counts the views past their staleness limit, and those views carry a Stale badge in the refresh queue.
On the Orchestrator, the queue's State column no longer offers a sort that ordered the views by one of its badges only.
On the Orchestrator, the Configure and Run generation buttons now tell assistive technology which panel they open.
On the Orchestrator, a state filter in the page address that the page does not know no longer shows Clear filters or says that no view matches the filters.
In the Payment Terms Optimizer, when no client meets the current constraints for an early-pay discount, the recommendation now says the optimizer proposes none, instead of presenting a 1% discount that no client takes as a successful proposal.
On the Pipeline Console, the Actions column now shows only to users who can pause and resume streams, instead of standing empty for everyone else.
On the Pipeline Console, the Pause and Resume confirmation now stays open showing Working... until the change completes, then closes and returns focus to the stream's row while it is still on the page.
On the Pipeline Console, the filters, Clear filters and Refresh now stay in view above the stream list on wide screens, instead of scrolling away with it.
On the Pipeline Console, the stream table now filters by layer, streaming state, tripped circuit breaker and detected backpressure beside the status filter, and Clear filters clears all five.
On the Pipeline Console, a KPI card with no figure to show now reads Not available as plain text, never styled as a number beside the real counts.
On the Pipeline Console, the Layer column and the Layer filter now show each layer's name beside its code, and throughput reads in records per second.
On the Pipeline Console, the page readout that a narrow table footer shows in place of the page numbers now reads in your language.
On the Pipeline Console, Refresh now also updates the connector names in the recent sync runs list.
On the Pipeline Console, a stream's pause or resume error now clears when you change the filters, sort, page or page size, or refresh.
On the Pipeline Console, the second figure in the KPI strip now counts the streams with streaming enabled and says so, so pausing or resuming a stream moves it.
On the Pipeline Console, the line confirming a pause or resume now names the streaming change, and it clears when you filter, sort, change the page or page size, refresh, or start another action.
On the Pipeline Console, each run in the sync activity rail now starts with its date in your time zone, and a line under the rail says it lists the 12 most recent sync runs across all connectors.
On the Pipeline Console, a sync run whose status the page does not recognize now reads "Sync status" with its code, instead of "Sync in flight".
In the Extraction Workbench, a decided field's history now says whether you or another reviewer made the decision.
On the Connector Health screen, the sync confirmation now offers only the data types you can leave out, and names the ones every sync always includes.
On the Connector Health screen, Force sync now shows only on connectors that can sync on demand, so it no longer appears on a paused connector or on QuickBooks Desktop, whose data arrives through the Web Connector.
On the Connector Health screen, a failed sync, connection test, pause, resume or delete now says why: the connector no longer exists, the request was refused, another sync or test is running, today's sync limit is reached, the provider did not complete the sync, or the connector can no longer sync as it stands.
On the Connector Health screen, a connector's details now offer only the recovery its own connection method supports: Reconnect provider for a connector connected through the provider's sign-in, and Replace credentials for one connected with a key or token.
Where QuickBooks Desktop is not available, Connector Setup now shows it as unavailable instead of leaving it out.
In the Extraction Workbench, a field the extraction pipeline flags as critical now carries a "Critical" mark on its queue row and in its detail view.
On the Extraction Workbench, every queue count now counts the whole queue instead of one page, Needs review lists the fields with no measured confidence first, oldest first, instead of leaving them off the page, Decided lists the most recent decisions first, and a count that is still loading shows no number instead of a zero.
On the K-Anonymity Console, validation runs show their overall privacy score, can be searched by name or dataset, and sort from their column headers.
On Model Experiments, an experiment still in design or draft can now be started.
On Model Experiments, a link can no longer sort the table by a date the table does not show.
On Model Experiments, Evaluate is offered once an experiment has collected its minimum sample, and each row shows its samples against that minimum.
On the Quality Console, the quality-gate summary now counts the L5 gate, so a report that fails only L5 no longer reads as passing every gate.
On the Quality Console, a report with invalid records now links to its issues in the Quarantine Workbench, for users who can open the workbench.
On the Quarantine Workbench, you can now sort the issue queue newest first or oldest first, as well as most affected first.
On the Quarantine Workbench, the issue queue now pages through every quality issue, 100 at a time, instead of stopping at the first 100.
On the Quarantine Workbench, you can now search the issue queue by name. The search covers every issue, not only the page in view.
Revenue Center's ARR now shows the date of the snapshot it comes from.
On the Serving Console, each deployment tile now shows its request count for the last hour, its average latency and its CPU and memory utilization beside the p95 latency, error rate and replicas.
On the Serving Console, the Scaling events card now counts the scaling events of the last 24 hours, and its label says so.
On the Serving Console, each alert's Metric column now shows the observed value and its threshold, not only the metric's name.
On the Serving Console, the alert state filter now offers the same states as the State column: Open, Acknowledged and Resolved.
On the Serving Console, each deployment tile now shows how old its telemetry snapshot is, so stale telemetry no longer reads as current.
On Shadow Deployments, each validation log entry names the shadow run it belongs to, and a failed check is marked by its own severity.
Platform Maturity's refresh card now says how many platform views are failing to refresh.
On the Connector Health screen, the sync confirmation now shows that its list of data types is loading, instead of an empty box.
On the Connector Health screen, the line under the table now says rows sort by Health by default, instead of claiming that failing connectors sort first.
On the Connector Health screen, the Provider filter now lists only the providers you can connect.
On the Connector Health screen, the event log no longer calls a connector unknown while names are loading, when they could not be loaded, or when you have more than 100 connectors: it waits for the names, or shows the connector id and says why.
On the Connector Health screen, the sync confirmation and a connector's data scope now name its data types in your language.
On the Connector Health screen, a connector with no uptime figure now reads Not available instead of an empty cell.
On Connector Setup, Connector Configuration and Connector Health, each data type's name and description now read in your language.
After the first sync, Connector Setup now says how long it ran.
Finishing a Sage Accounting or FreshBooks connection now asks which business to read, not which organization.
Connector Setup and Connector Configuration now show region and platform choices by name instead of as codes.
Data Platform Center now says when some of its platform facts could not be loaded, with a retry, instead of leaving them out.
The Data Platform Center's sync-run timeline and the Connector Console's event log now say a run's connector was deleted instead of calling it unknown.
In the Extraction Workbench, a pending field's status and the screen's subtitle now read in the reader's language, so a German reader sees "AUSSTEHEND" instead of "PENDING".
Feature Store's statistics caption now says when the statistics were computed, not just which day they cover.
On the Performance Console, experiment start dates now read in your time zone instead of UTC.
On the Performance Console, Evaluate and Abort on an A/B experiment now show to anyone allowed to configure models, the permission those actions need.
On the Performance Console, each retraining run now shows the date of its latest step: when it started, when its training finished, or when it was deployed.
Revenue Center now shows the date of the model monitoring its forecast health comes from.
The sign-in page now says why an account that signs in with a text-message code is refused, and what to do instead.
Sign-up now stops a payment for a billing address in a country we do not serve yet, before any charge, and says so.
On eighteen summary strips, retrying a figure that failed to load no longer blanks the whole strip: the other figures stay on screen, and the failed card keeps its message until the retry answers.
On the Serving Console, the alert count card now reads Unresolved alerts: it counts every alert not yet resolved, acknowledged ones included.
On the Performance Console, a link can no longer sort the experiments by a date the table does not show.
When a request is blocked, the screen now says it could not be processed and shows a Contact support link with the request ID to quote, instead of asking you to reload the page or contact your administrator.
On Tenant Configuration, the Security step now says the IP allowlist is saved but not yet enforced in this release.
Sign-up now stops a payment for a billing address in a country served from a different region than your company, before any charge, and says so.
The Changed-since panel on Ingestion Flow now lists the changes of the selected version, not the latest changes across all contracts.
A filter on revenue-at-risk now means the same amount everywhere. Asking for risks worth at least a given amount no longer quietly includes ones worth up to 99 cents less.
An opened Rolling Forecast variance event now shows its controllability confidence as a percent.
The Rolling Forecast table no longer says no months match the current filters when no filter is applied.
The Rolling Forecast ledger table now shows the gross margin amount beside revenue and cost of goods.
The Rolling Forecast ledger table now shows gross margin as a percent of revenue.
The Rolling Forecast KPI strip now shows the month's gross margin and its revenue change against the prior month.
The Rolling Forecast variance events show when each was detected on one line.
The Rolling Forecast now dates a variance event's review in your configured time zone.
Opening a variance event on the Rolling Forecast now shows the drivers behind it.
On narrower screens, the Runway Planner now shows on the page when its live forecast, cash position or forecast narrative fails to load, each with its own retry, and says when no baseline forecast has landed yet.
On the Runway Planner, each stress test result now lists its recommended actions in your language, and a run with no cash position on file says once that its risk levels are relative to a zero cash basis.
On Account Settings and Tenant Configuration, each section's Retry button now announces which section it reloads, so screen-reader users can tell several failed sections apart.
A failed refresh on the Rolling Forecast, Service Lifecycle and Demand Forecast consoles now keeps the rows already on screen instead of replacing them with an error message.
On the Workload Center, when the capacity figures cannot be loaded, the page header now says "Analytics unavailable" with a retry, instead of leaving them blank.
On the Workload Center, the insights timeline now dates each insight by the analysis it comes from and lists the newest analysis first.
On the Workload Center, the portfolio line's counts and the Demand Forecast card's unit now read in the singular for one, such as "1 endpoint monitored" and "1 forecast".
On the Workload Center, a recent forecast with neither surplus nor deficit now reads "balanced", and hours show up to two decimal places instead of rounding to whole hours.
On the Workload Center, the recent demand forecasts now list the latest forecast only, deepest deficit first, each dated by the period it forecasts.
On the Workload Center, the recent demand forecasts' dates no longer read a day ahead for viewers west of UTC.
On the Workload Center, the portfolio line now states the 30-day utilization, and the Demand Forecast card is captioned with the day's forecast error (MAPE), naming an accuracy breach in words.
Filtering capacity optimisations by margin impact now means one amount, in one unit, and the figure that comes back keeps every cent of it.

Security

Rate limits on password reset and verification-email resend now actually apply, and a route can limit per account as well as per address.
A temporary lock after too many failed attempts now lasts its full length instead of sometimes ending within seconds.
Sign-in limits and temporary address blocks now always apply to the address a request actually comes from.
Hiring metrics and applicant tracking sync health now apply tenant row security to every record they read.
Contract overview and churn analytics views now apply tenant row security to every record they read.
A file downloaded from the file store now saves under its original name instead of opening in the browser tab.
Workspace administrators can now shorten the idle timeout below 30 minutes; saving it signs everyone in the workspace out, so the new timeout applies at once.
Removing an approval step from a workflow, or a tag from a contract, now requires the contract delete permission rather than the create permission.
Switching to another account now opens that account's home page and reloads your other open tabs into it, so no screen keeps showing the account you left.

Limitation

Connector Setup lists BambooHR as not yet available until its new connection method ships.