docs/review-notes.md

# Microsoft Ads submission preparation

Prepared October 27, 2026. Original package identity
`app-6aa3b20dd9fc8191ad1ea4f46e9e2808`, version `1.0.0`, display name,
subtitle, and starter prompts are preserved. The original downloaded archive
is unchanged. The owner confirmed that the entry is unpublished/rejected and
its registered endpoint is https://microsoft-ads.mcp.adzviser.com/widgets/http.
The owner selected the Demo workspace for the review cases and authorized
replacing its expired Microsoft Ads connection. It now contains the owner's
Adzviser LLC - Expert account, copied from their existing bing ads new workspace.
The source connection and the other eight Demo connections were preserved.
The reviewer account must have access to Demo; fresh reviewer sign-in has not
been verified.

The archive contains matching portable and compatibility manifests, concrete
MCP configuration, source-specific skills, the supplied Adzviser icon, five
Data Mode positive cases, three unsupported-action negative cases, the supplied
demo link, all supported countries, website-only subscription information, and
public-facing release notes. No reviewer credentials or private app bindings
are included. No cloud plugin or submission was created or changed.

## Five positive cases

1. Overall Microsoft Ads Cost, Impressions, and Clicks, grouped by Currency code.
2. The same metrics by Campaign ID, Campaign name, and Currency code.
3. The same metrics by Device type and Currency code.
4. Comparison of August 2026 with July 2026, grouped by Currency code.
5. Daily Cost, Impressions, and Clicks, grouped by Currency code.

All cases use Demo and the fixed historical period August 1–31, 2026,
inclusive. The comparison also uses July 1–31, 2026. The owner confirmed the
campaigns are suspended, so rolling dates could eventually exclude their activity.
Keep these calendar months unchanged when the reviewer runs the cases. Only
the input dates are fixed; answers must be calculated from live returned data.
No observed numeric answers are embedded in the case expectations. The exact
natural prompts, tool names, and arguments are in the manifest review fields.
Preview UI is available only on explicit request and is not required by cases.

## Verification status: Five reporting requests passed

Verified on October 27, 2026 after the authorized connection replacement.
Workspace discovery identifies Adzviser LLC - Expert in Demo, and metric
discovery confirms all requested fields. All five actual Data Mode requests
returned structured report rows without an authorization error. Every request
used metrics Cost, Impressions, and Clicks and retained Currency code.

The verified range was August 1–31, 2026; the comparison range was July 1–31,
2026. All five requests were rerun after selecting these fixed periods.
An earlier recent-period probe returned no rows for September 26–October 25;
the historical cases do not depend on the campaigns resuming.

| Case | Status | Evidence |
| --- | --- | --- |
| Overall performance | Passed | One August account row: USD 481.70 Cost, 54,462 Impressions, 697 Clicks. |
| Campaign performance | Passed | Four distinct campaign IDs and names; all three metric totals exactly match the overview. |
| Device performance | Passed | Computer, Smartphone, and Tablet; all three metric totals exactly match the overview. |
| Period comparison | Passed | Both months returned rows. July has USD 863.50 Cost, 103,147 Impressions, and 1,604 Clicks. |
| Daily performance | Passed | 22 unique August daily rows; all three metric totals exactly match the overview. |

Daily rows span August 3–31. One returned daily row is explicitly zero;
nine dates in the requested month have no returned row. The case requires
disclosing absent dates rather than manufacturing zeros. No extra assumption
message was returned. Campaigns and devices retain their actual labels.

These observations are verification evidence, not hardcoded expected answers.
The fixed dates prevent the review period from moving past existing activity.
Continued historical availability and working authorization still depend on
the source service; verify reviewer access before submission. The earlier Demo and Bing Ads Demo/Togather
attempts failed authorization; replacing only Demo's Bing Ads connection resolved
that failure for the requests tested here.

The three negative conversations are Not run. They cover changing budgets,
pausing campaigns, and identifying individual ad clickers. Each expects an
explanation of the unsupported action without unnecessary tool calls or a
fabricated success. Fresh conversations through the saved submitted version
and fresh reviewer sign-in are also Not run.

## Server configuration and annotations

The public health endpoint responded successfully on October 27 with build
`6bf20c8d-20261002201522`. OAuth resource and authorization-server discovery
responded successfully for the confirmed Microsoft Ads endpoint. Anonymous
MCP initialization and tool listing returned the expected 401 OAuth challenge.
These checks do not prove the reviewer's login or upstream account access.

That deployed build predates the existing shared annotation correction in local
source. Its scoped reporting definitions set openWorldHint=true. The corrected
local profile sets openWorldHint=false for account-bound Microsoft Ads reporting
and workspace discovery. The correction needs deployment and a portal rescan;
the ZIP cannot change the live server's annotations. No deployment was performed
and no server implementation was changed during this preparation.

Local intended annotations, in readOnlyHint / destructiveHint / idempotentHint /
openWorldHint order, are:

- Metric catalog: true / false / true / false. It reads static field definitions.
- Retrieve and preview: false / false / false / false. Reporting reads authorized
  account data and records usage; preview compatibility paths can also persist
  a reporting preference. These tools do not edit advertising campaigns.
- Workspace listing: false / false / true / false. It resolves authorized
  workspaces and may clear a stale stored preference.
- Reporting-mode preference: false / false / true / false. It saves a presentation
  preference; repeating the same change is idempotent.

## Demo and listing

Demo: https://youtu.be/_lt7XO4DzAU. The recording is an unlisted, non-private
196-second Adzviser video. A signed-out isolated browser played sampled beginning,
middle, and final sections. The recording shows development-plugin setup, a
spend question, a no-data response for Matterport, and then a successful displayed
spend answer of USD 481.70 after switching to the bing ads new workspace.

The demo demonstrates a reporting response using the same underlying account
now connected to Demo, under its original bing ads new workspace name. The five
Demo requests were separately verified as recorded above; the recording itself
does not verify fresh reviewer sign-in.
No passwords or tokens were observed in the inspected frames; not every frame
was inspected. Playback initially stopped after about a minute; timestamped
restarts allowed the later sections and ending to play. An uninterrupted full
playback was not verified. The supplied link is preserved in the review metadata.

The website, contact form, privacy notice, and terms pages were retrieved without
sign-in and their contents inspected. They identify Adzviser and provide the
expected publisher information. The supplied square PNG is reused unchanged
at 256 by 256 pixels for the listing logo and composer icon.

## Remaining preparation and portal work

1. Deploy the existing shared annotation correction, then rescan in the portal.
2. Upload to the existing intended entry, preserving its identity and endpoint.
   If the platform rejects that update, inspect the exact error before creating
   another entry or changing the endpoint.
3. Enter reviewer credentials and exact sign-in instructions only in secure
   dashboard fields. Verify a fresh reviewer login reaches the prepared workspace
   without additional account setup. Rehearse the cases through the saved version
   and inspect imported review metadata before completing required attestations.

The five direct reporting requests passed. Archive validation does not verify
reviewer sign-in, the saved portal version, or approval. This package remains
a draft pending the checks above.