Get a current advice list
After file preparation and exchange or API setup, the briefing sits directly before the advice list. It covers the game situation, changes, conclusions and priorities; short labels and lists are formatted for reading. Each new reply replaces the current list; previous notes are preserved for AI.
How to use it
- Before a new analysis, save notes and decisions on earlier advice. Sync includes them in the assignment.
- Upload the new reply. Once validated, only that analysis's current recommendations remain; no manual cleanup is needed.
- Done completes your action, but not necessarily observation: a later analysis can ask you to wait for data, review the outcome, or notice early harm. These recommendations appear only when an analysis runs; MARPLA does not promise a live alert.
- If the file is rejected, obtain a corrected reply in the same chat. Your current list remains until a successful import.
Advice and practical context
The current list represents one analysis
A newly validated reply replaces the working set, so save decisions and notes that the next analysis needs before importing it.
Example of changing priorities
Example: when an old recommendation is complete, record its result before a new run; the AI receives that history and can avoid reopening closed work without cause.
A new list should reflect changed evidence or priorities rather than erase unfinished work without explanation. Before replacement, confirm important notes are saved and prior statuses match reality. After a corrected file is accepted, review recommendations that disappeared or appeared. History helps AI account for earlier decisions, but the user remains responsible for checking the current list before acting.
A failed import replaces nothing
A rejected file leaves the active list intact. Correct the reply in the same context and distinguish format repair from requesting fresher data.
Technical detailsCalculations, permissions and behavior
AI analytics and advice tables
After file preparation and exchange with AI or API setup, the AI briefing sits directly before the advice list. It covers the current situation, important changes, conclusions and priorities; short labels and lists read as ordinary text. It shows the analysis date and period.
A successful file or API reply updates the briefing; pending and failed runs do not replace the last completed one. To receive advice, connect the selected game to Roblox first: open AI connection → Roblox report access → Set up Roblox access and allow analytics reads. This is a Roblox key, not an AI key.
The assignment and table stay hidden until completed-day reports arrive. The access panel is only in AI connection. In Advice, the refresh button shows health: yellow means check the connection or permissions, red means an error, and green means waiting for data with an update check available.
A healthy connection uses the normal button. The yellow or red button opens AI connection, where you can see the reason and retry. If only some data is available, open Reports that still need a connection there for report names, exact missing permissions and links to this game’s settings.
Reading reports does not require game-edit permissions. If the key is verified but data has not arrived, wait for the reports and check the connection again. Previously saved advice and notes are not deleted.
The AI advice button on a chart opens current recommendations affecting that metric, with older recommendations in a collapsed read-only archive. Archived advice no longer requires action or highlights charts. A highlight means there is advice without your reaction.
Read here clears the highlight in this chart only. To do / Done / Not doing in Your notes saves the decision for the whole advice. Done and Not doing clear its highlight in all linked charts; other pending advice remains highlighted.
Notes save when you leave the field; the next analysis includes your explanations and edit history. The wide Advice column contains the task, its location and your steps directly below it. Clickable thumbnails appear inside the sentences discussing each creative and beside its metrics.
Select a thumbnail to enlarge it. All related images remain available under All images in this advice. Evidence and verification plan expands across the full width with reasons, risks and result checks.
Technical details open separately. Download a new AI assignment for the new task format; old advice and saved marks are not rewritten. Wait for processing to finish when uploading a file.
Open AI analytics under My projects in the sidebar, immediately after Advertising. A large game picker with its icon sits below the heading; search on the right finds advice and notes across every table page. File or API mode and settings are in AI connection.
What data will AI see? on the right of Whole-game advice immediately opens a separate assignment-contents window; details load inside it. In Through chat, no API, follow the three blocks linked by wide green arrows. In the first, select Collect data for AI: the server prepares a file from the latest saved game data even if you leave the page.
When Data collected, you can download the file appears, select the separate Download file button. The time below it shows when the file was created, not when Roblox sources were refreshed. One file includes all available game sections, products and groups, prices and history, advertising, creatives and sets, audience, retention, economy, events and community.
Attach marpla-assignment.txt to your AI chat: the full assignment is already inside, with no extra prompt to write. The AI should return a downloadable marpla-answer.json. Download it and drop it onto the third block in MARPLA, or select Upload AI reply.
The block highlights while dragging; reading, upload and processing start automatically. Use one JSON or TXT file of up to 700 KB. The service validates it, routes advice to its sections and opens All advice.
Large source data is stored once with the assignment; Supporting data includes a compact preview and a full download. If the table fails to refresh after saving, refresh the page without regenerating the reply. Retry the same file after a save error.
A failed new preparation does not cancel the available previous assignment. Section buttons filter existing recommendations without limiting the next assignment. Already loaded advice switches without waiting again.
The list updates after notes are saved or a new reply is uploaded. The table has three columns: Advice, Before / after and Your notes. The decision switch saves immediately; notes save when you leave the column or select Save note.
Instead of a result column, select Add result, write your observation and select Save and complete: the text is saved, the editor closes and the advice becomes Done. For Not doing, Save and close preserves that decision. Existing observations open with Your result.
Adjust an actual completion time under Change time and save the note. A proposed goal is not a promised result. MARPLA checks supported settings against fresh saved observations automatically and shows when to reanalyze.
When ready, download an updated assignment: saved decisions, notes and verification results are included automatically. There is no When column: MARPLA keeps the review schedule and shows Before → now when a comparison is ready. Above the checklist, Best time to change game settings is collapsed with the local Roblox day-boundary time visible.
Expand it to learn why mid-day changes mix old and new settings' results. New assignments require simple actions, clear explanations and concise numbers; future results are labeled as goals, not promises. You can download the assignment again and return later to upload the reply.
If your AI cannot create files, use its complete JSON response through Or paste the reply. In Through API, the owner connects their own key and chooses allowed data. Done completes your action, but not necessarily observation: a later analysis can recommend waiting for data, reviewing the outcome or noticing early harm.
These recommendations are generated only when analysis runs; MARPLA does not promise a live alert.
- Background file preparation
Select Collect data for AI. MARPLA queues preparation on the server, so you can leave the page and its state restores automatically when you return. When Data collected, you can download the file appears, select Download file.
You can download the previous prepared file during a rebuild or rebuild it again. The time below the button is when the file was created, not when Roblox refreshed its source reports: the assignment uses the latest data already saved in MARPLA.
- Clear tasks and data quality
Done has a soft green highlight; Not doing has a soft red highlight. New advice must explain your action, where to do it and how to check completion. If MARPLA needs development work, the AI must distinguish it from your actions: describe a data limitation or clearly explain what to send to support.
Internal field names belong in technical details. New assignments combine equivalent hourly timestamps and use the latest fetched revision; unresolved conflicts are not converted into numbers. Advertising dates without exact hours do not establish matching windows or complete data.
To obtain advice under the new rules, sync and download a new assignment. Existing advice is not automatically rewritten.
- Current advice only
After a successful new reply import, the list shows only that analysis's recommendations, including advice on charts. Earlier advice, decisions, results and notes remain in history for the next assignment. AI must include unfinished tasks again when still relevant and omit resolved issues.
A valid empty recommendations list removes earlier advice from the analyzed sections. Invalid files and reuploads of the same reply erase nothing. A single-section analysis leaves other sections' advice intact.
Moving advice into history does not mark it Done.
- Chat transfer
An assignment contains data only for the selected game and table. You choose which chat receives it. Replies are validated against the saved assignment; uploading again creates no duplicates and does not erase your notes.
A new assignment replaces the previous one for that table; replies are accepted for 7 days. Invalid replies do not change the table. API keys are never included in assignments.
- Data for pricing advice
Assignments include the entire saved catalog, current prices and availability, groups, daily purchases and sales revenue for each product over 28 completed days, and observed prices and changes. Counts of included products, prices and changes appear below the assignment. Products without sales are included.
Unmatched sales items are sent separately. Retention and milestones are included with the corresponding permissions; results of that analysis are visible only to users with those permissions. If an improved-assignment message appears, download the updated file and send it to your chat again.
The old reply has not analyzed the added data.
- Price history accuracy
Missing data is null; zero is used only when recorded by the source. The latest daily revision replaces, rather than adds to, earlier revisions. Price history includes observations from the period start through assignment preparation and the last earlier observation.
Detection time is not the exact effective time; exclude transition days from comparisons. Current prices are never projected backward and purchases are never inferred from revenue. Groups reflect current membership.
If a full package exceeds capacity, the service reports it instead of presenting a truncated package as complete.
- Connection and billing
The API key is encrypted and never returned to the browser. Connection verification is a short billable request without game statistics. Up to 10 checks and 20 analyses per day; set a monetary budget with the provider.
You start paid analysis. Background setting verification does not call AI.
- How results are checked
Verifiable advice identifies an exact product, field and expected value. An old catalog cannot confirm a new change. Checks run on save and through the background collector, normally every six hours.
Observation time is not the exact effective time. Gaps, small samples, other price changes and milestones prevent a clean comparison. Ready for analysis does not mean proven success.
Other advice requires manual verification using its plan.
- Data and security
What data will AI see shows aggregates for 28 completed days. Recent advice, notes and checks are included automatically. The server validates response structure, references to this game's facts and permissions; an expired response cannot replace newer results.
Roblox settings are not changed automatically. Disconnecting AI removes the key but preserves advice.