User guide

Compare Home creatives and sets

Content → Home lists saved set compositions. The Created column shows the Roblox set creation date without a time. The opened set header shows Created and Completed dates and times in UTC.

Replacing its images does not change the set creation date. Member rows show their set's creation date. Observation or sync times are never presented as creation or completion.

Unknown dates use a dash; active sets have no completion date. Sync saves Roblox corrections to completion dates. Earlier images known only from statistics do not receive invented set dates.

Open a set name: its name and dates appear in the top heading, followed by totals, members and separate creative chart lines. Press Escape or the area outside the drawer to return from a creative to its set, then to the list. The chosen period is retained.

Returning to the main list also restores its scroll position, page, search, status filter and sorting. Use period and metric controls, filters, search and sorting to compare results. Each summary figure shows the days contributing to its value.

Impressions and qualified plays are totaled; QPTR is qualified plays / impressions, and session length is weighted by qualified plays. Incomplete totals have an asterisk. A reused image's daily result cannot be divided between intraday runs.

Ambiguous days and late reports after deactivation do not extend a set. 7D QPTR and winning segments are not added together. Signal compares QPTR on shared dates with sufficient data; it is a guide, not proof of a winner. Milestones show saved changes with Before → After thumbnails.

Viewing or syncing history does not change active Roblox images. Set performance over time appears below the general table: every known set has a persistent colour and thumbnails. It includes all sets in the chosen date range, independently of table search, status and pagination; choose lines in the legend.

Controls include impressions, qualified plays, average session and QPTR, date presets, day/week/month steps, straight/smooth lines and bars, milestones, tooltips, AI advice and refresh. The guide icon sits next to Set performance over time or Performance over time, and AI advice appears below it on the left. The selected metric title is not repeated; advice follows that metric and the open set or creative.

Set exact dates above the table; the chart does not repeat those fields. Larger steps first total source counts, then calculate QPTR and weighted session length. A weekly or monthly point uses its first available data day.

Line and smooth modes always show a single overall timeline: one total per date, including earlier creative evidence. Counts from both sets on a replacement day are added before recalculating QPTR and weighted session length. Colour identifies the largest impression contributor in that day or larger bucket; it does not indicate an exact switch time.

Tooltips show the overall total and set contributions with their image thumbnails. Rows without a value on the selected date are hidden; zero values remain visible. A short first day no longer restarts a separate line near zero.

Missing dates have pink dashed connections without fabricated observations. Bars show separate set series. Inside a set, the chart still contains only its own creative lines.

The set list, the overall creative list and each set's creative table have independent sorting. Closing a set restores the list's own order; reopening the set preserves its member sorting. Average session, QPTR, 7D QPTR and Signal use colours relative to the best available value in that column: at least 95% is green, at least 80% amber, below that red.

Sorting and pagination do not change the comparison base. Missing, single and identical values remain neutral. Colour helps read the table; it does not indicate statistical significance, and different sets may cover different dates.

Signal still requires shared dates and a minimum display sample, which does not prove a winner. AI reviews creatives by exposure and effect size, without a mandatory seven-day wait. New assignments include up to 90 completed days of saved Home history: each composition, its members, shared dates, impressions and qualified plays, and previous occurrences of proven copies of the same image.

Advice must name the baseline conversion, worthwhile improvement, actual sample and next per-variant exposure checkpoint. At a 1% baseline, approximate impressions per variant are 42,693 for 1.2%, 7,750 for 1.5%, and 2,319 for 2%. This plans independent observations, equal allocation, one preplanned comparison, 5% error and 80% power; it does not prove significance for the current set.

There is no universal click or play count. Home provides qualified plays, not per-thumbnail clicks. Roblox personalization changes audience mix and exposure shares, so ranks and QPTR intervals are descriptive.

AI must consider winning segments and conflicting prior runs, without pooling different or overlapping tests. When a large gap and sample support action, it can propose the next test now: a strong control plus new candidates to confirm and improve. It may shortlist one, two or three observed leaders while retaining close and segment-specific alternatives.

A single control candidate does not mean leaving only one active image. If evidence is weak, advice states the missing exposure; an ETA requires an observed traffic rate. Mean session without dispersion does not establish audience quality.

Short creative review horizons in new assignments do not change price safeguards or older downloaded assignments.