Evaluation Mode — specification

Status: built, dormant. src/licensing/features.ts gates every feature on licence state, and the rest of the visual honours it. It has no effect yet because LICENSED_SERVICE_PLANS is empty, so every session resolves to not-enforced and gets the full feature set. Filling that array in — once the Partner Center plan exists — turns Evaluation Mode on.

The decision

Ship a permanent, deliberately limited free tier rather than a timed trial. Paid licences unlock the rest.

A genuine 7–14 day unlimited trial would require tracking activation dates per user, which means a server, accounts, hosting, a privacy policy covering collected data, and outbound network access. Outbound network access alone disqualifies the visual from Power BI certification — the audit checks for it explicitly, and pbiviz package --certification-audit currently reports “no external requests found”. That is not worth trading away.

A free tier costs nothing architecturally: the licence state is already known client-side, and gating features on it needs no network, no storage and no identity.

This also sidesteps a one-way door recorded in the Phase 0 report: Microsoft states a free trial cannot be disabled later for the same published plan. Not enabling one avoids the commitment entirely.

A Microsoft-managed timed trial is not available for this offer type

Confirmed 26 August 2026 against the Microsoft Marketplace publisher FAQ, and it settles the question rather than reopening it.

The built-in free-period trial — customer charged $0.00 for 1 to 180 days, then rolling to paid automatically — applies to SaaS and VM offers only. For a Power BI visual, Microsoft’s only trial route is a listing-only offer that sends the customer to a web address of your choice to run a trial outside the marketplace, hosted and managed by the publisher.

That is the server-and-accounts path this document already rules out on certification grounds. So Evaluation Mode was not merely the better option for this offer type — it is the only in-product one.

What Microsoft supports

Confirmed against License models for Power BI AppSource visuals:

Some visuals have free trial versions, while others have a basic version available for free with extra functionality available for purchase.

And the matching notification:

If you have a free version of the visual, a banner appears with a link to upgrade your license. This banner will disappear after a while.

That banner is notifyFeatureBlocked(tooltip). The corner icon is notifyLicenseRequired(LicenseNotificationType.General). Both are host-drawn; the visual must not build its own purchase UI.

The split

Free — Evaluation Mode

Enough to check the arithmetic against a spreadsheet with real data, not enough to run a shift.

Three problems found in the first draft — all resolved

Problem 1: the split as written made evaluation useless

The stated goal is that someone can “verify the math against Excel using real data in ten minutes”. But “full tooltips and diagnostics” sits on the paid side, and the tooltip is where the calculation is visible.

Without it an evaluator sees a line with no numbers behind it. They cannot check offered load, cannot see required productive versus required scheduled, and therefore cannot reconcile anything. The free tier would demonstrate that the visual draws a chart, not that it computes correctly — which is the one thing it needs to prove.

Resolved: split the tooltip rather than withholding it.

Evaluation tooltip Paid tooltip adds
Interval Queue
Forecast contacts Scheduled agents
Average handle time Staffing gap
Offered load (Erlangs) Expected service level
Required productive agents Expected occupancy
Required scheduled agents Overload and data-quality warnings

The left column is the audit trail. It is also exactly the part a spreadsheet reproduces, so it is what makes the ten-minute check possible. Everything on the right depends on scheduled staffing, which is already a paid feature — so the split falls along a line that is easy to explain and needs no special-casing.

Problem 2: “the first 12 intervals” is often the wrong 12

Most interval datasets start at midnight, so a centre open 07:00–21:00 would show a flat line at zero.

Resolved: leave it. The first 12 in sort order, full stop. A report author who wants a different window filters to it — which they can already do, and which is one fewer rule to explain than “the first 12 with volume”. The truncation notice states the window, so nothing is hidden.

Problem 3: in Reading mode the reader may get no indicator at all

This is the one that matters most, and it comes out of the Phase 3 work rather than the docs.

LicenseNotificationType.General is only enforced in Edit mode in a supported environment — Microsoft’s own note says calling it in Reading mode or on a dashboard “doesn’t apply the icon and returns false”. notifyFeatureBlocked similarly no-ops in unsupported environments.

So a report consumer on an unlicensed tenant could see a chart showing 12 intervals of a 96-interval day with nothing indicating why. That is not a paywall, it is a wrong chart. Someone would staff a shift off it.

Resolved: render a neutral, factual truncation label whenever the data is cut, regardless of whether the host drew its notification:

Evaluation mode · showing 12 of 96 intervals

This is not licensing UX and does not compete with Microsoft’s notification. It carries no price, no call to action and no upgrade link — it explains why the chart shows less than the data contains. Withholding that explanation would be the actual violation of good faith, and the guidance Microsoft gives is against homemade purchase dialogs, not against a visual describing its own state.

The notifyLicenseRequired / notifyFeatureBlocked return values are already read and stored (LicenseManager.needsOwnMessage), so the visual already knows when the host declined to draw anything.

Resolved rules

A queue field without a licence refuses to calculate. Per-queue staffing sums higher than pooled, so silently ignoring the field would produce numbers matching neither the workbook nor the paid version. The visual shows a state message asking for the field to be removed or the licence upgraded, and calculates nothing. Refusing is honest; quietly changing the model is not.

A scheduled-agents field is ignored, not refused. Unlike a queue field it does not alter required staffing, so ignoring it changes nothing about the numbers being verified. The feature-blocked banner says why.

Twelve intervals, fixed, at every interval length.

Nothing blocks outright any more. unlicensed, unavailable and unsupported-environment all fall back to the free tier rather than a blank rectangle. For unavailable in particular that is the only defensible answer: a lookup that could not complete is not evidence that anyone is unlicensed. The old VisualIsBlocked overlay is deliberately no longer requested — it hides the visual entirely, which is the opposite of Evaluation Mode.

Paid outputs are not computed at all, not computed and hidden. The interval limit is applied before calculation, and a paid scheduled value is kept out of the y-axis range so it cannot leak through the axis scale.

What is tested

test/licensing/features.test.ts and test/licensing/evaluationMode.test.ts cover the interval limit, the suppression of paid outputs including the y-axis leak, the tooltip split in both directions, and the truncation notice — including an assertion that it contains no price, button or upgrade wording, since that is the line between describing state and building a purchase dialog.

What this needs from the owner

Nothing to start building. Two things before it can ship:

  1. The Partner Center offer must exist so LICENSED_SERVICE_PLANS in src/licensing/licenseTypes.ts can be filled with the plan’s Service ID. Until it is, enforcement stays off and everything renders as licensed.
  2. The private-plan purchase test, which is the only way to exercise the real licensed path end to end. Every state is already mocked in test/licensing/, so that test should be a confirmation rather than a discovery exercise.