The useful outcome from a photography report is not a larger dashboard. It is a better next choice: which album launch deserves a second look, which photos belong in a shortlist, and which apparent changes should be left alone because the evidence cannot support them.
That sounds obvious until the report is built around calendar windows. An event gallery does not behave like a subscription product with steady daily traffic. An album can receive most of its attention soon after publication, then go quiet. Comparing its first few days with a later, empty period answers a question about timing, not about whether the launch landed.
The work moved the comparison to the album launch itself. A report can compare an album at day seven with earlier albums at day seven, then let the photo view support a different job: finding material to share, export, or revisit. Those are decisions. A total number on a dashboard is not.
Agents implemented the report and photo explorer in bounded steps, with shared comparison rules rather than calculations invented separately for each screen. Synthetic cases checked missing coverage and first-publication boundaries. Cold reviews of rendered screens checked a different question: could someone understand what to do next? Keeping those checks separate let the work change both the report’s rules and its presentation without treating one as proof of the other.
Start with the thing that was released
The first useful unit is the album, not the month.
Instead of asking whether the gallery had more activity than it did during an arbitrary prior period, the report asks how this album’s first week compares with other launches at the same point in their lives.
The report uses an album’s first publication, not the date it was later republished or edited. Otherwise, an old album could reappear at the top of a list and be treated as a new launch. Those are different facts and need different dates.
The change matters most for a newly published album. A calendar comparison can make new work look as if it grew from zero, even though there was nothing there to open during the earlier period. The report now labels that case as not yet published instead of a measured zero. There is no earlier activity to compare. Calling the difference growth would make the number say more than it knows.
This is a small rule with a broader use: compare work against the point when it became available, not against a time before it existed.
A photo ranking is a finding aid, not a critique
Album reports answer whether a launch deserves attention. The photo explorer answers a narrower question: which images should be considered for a share, an export, or a closer review?
It does that with a gallery view, filters, and three ways to sort the results: popular, rising, and recently active. An owner can add photos to a shortlist and export the current selection. That turns a report into a working set instead of a screen someone has to remember later.
The distinction is important. A rank records browser actions such as opens, download requests, favorites, or shares. It is not a rating of the photograph, and it is not a count of people. One browser can produce more than one action. A quiet image can still be the right photograph to publish. A frequently opened image might simply be part of a heavily viewed album.
The same restraint applies to the word rising. The calculation is available only when both compared windows have complete coverage. Equal windows compare recorded actions; unequal custom windows compare actions per calendar day. If part of either window is missing, the report says the comparison is unavailable. It does not substitute a clean-looking answer.
That is less dramatic than a chart that always moves. It is more useful when someone is deciding whether a change is worth acting on.
Three whole workflows, not three views of the same dashboard
Before the combined experience was built, three workflows were considered against the same representative states. A trend-led version began with daily change and was best for investigating a burst. An analyst-table version put comparison, export, and inspection first for known albums or longer shortlists. An album-led version began with the event itself.
The final shape kept the trend-led opening, the table’s inspection and export tools, and the album path. It rejected a permanent inspection rail that would crowd a phone screen and an album-only front door that would make cross-gallery questions take a detour.
That comparison is worth naming because it changed the outcome. The result is not a set of tabs with the same totals in different arrangements. It is one path for deciding whether a launch needs attention and another for choosing images from it.
The limits belong beside the number
The report uses a conservative traffic rule by default: audience and unclassified browser activity, rather than controlled or automated traffic. That is a filter, not an identity check.
It also carries its coverage state with the result. A partial collection window means the count is what was recorded, not a complete account of what happened. An unavailable report does not show a zero total or a zero ranking in its place. Those cases matter because an empty-looking dashboard can mean no activity, missing collection, or a failed report. Only the first is a conclusion about activity.
The September 29 production receipt verifies an earlier foundation: a 90-day reporting window, complete and partial coverage states, and a conservative traffic view. The newer launch-focused experience has October captures and local synthetic-data checks for the redesigned screens, filters, shortlist, export, and unavailable states. They do not verify the newest experience at the production edge.
Keeping those layers separate is not a disclaimer at the bottom of the page. It changes what the report is allowed to claim.
The next measurement is whether it changes a choice
The report can now produce a shortlist and a launch comparison. The remaining question is whether those artifacts lead to a better share decision or a better launch follow-up than the previous way of working.
That needs a real decision record, not more event counts. It would compare the report’s shortlist or launch signal with the choice that followed, while keeping the same limits in view: browser events are not people, attention is not quality, and incomplete coverage is not a trend.
For now, the useful step is simpler. Start with the album that went public. Compare it with launches of the same age. Then use the photo list as a place to look more carefully, not a machine to decide what is good.