Today’s research: planning and daily review

1. Start your morning

1. Open the profile menu → Today's research. Check the date, then review assignments, deadlines and events in Today's direction. Use the date picker or arrows to view other daily records.
2. Choose Write or Edit beside Today's goal, enter your priorities and save. AI plan draft is optional; writing your own plan is fully supported.
3. Expand Select focus tasks to choose up to three. Use Act on a plan item to prepare a comment, message or other next step. The sections below explain applying drafts and executing actions.
4. Check new requests and follow-ups in Reports & actions. A red number counts new requests, not unfinished assignments or executions.
5. View all work opens My work. A written plan and GitHub task state are separate; check the original task state when work is finished.

2. Understand dates, counts and evidence

1. Selecting a date opens that day's personal notes. It does not reconstruct a complete historical snapshot of currently collected work. Check the date and collection time of each source.
2. Unfinished assignments, deadlines, events and evening checks measure different things. A 3/5 check-in means three checklist boxes are checked, not three completed tasks or submitted reports.
3. Expand schedule/collection status or observed participation to inspect evidence and limits. A scheduled event does not prove attendance; missing observations do not prove inactivity.
4. For missing work, check your GitHub ID, assignment, active grant/project participation and source collection. Inaccessible sources are omitted.
5. Recent records lets you review earlier private plans and outcomes. If the wrong date is selected, choose Today before entering new notes.

3. Write with Markdown, mentions and citations

1. Goals, results, blockers and next steps initially appear in reading mode. Choose Edit beside a field to change its original text.
2. Expand editing tools for headings, lists, bold text and links, then check the preview. Short sections with headings make both reading and action selection easier.
3. Type @ and select an exact researcher username. Type [[ to search and select an accessible document. Typing a similar name alone is not the same as selecting a resolved reference.
4. Save updates your private note. Mentions and citations in a private note do not send messages or grant source access.
5. Save and act first saves your changes, then opens section selection. If another tab has changed the record, resolve the conflict before saving again. Add files for outgoing messages in the action editor.

4. Generate morning and evening drafts separately

1. Use AI plan draft in the morning and AI review draft in the evening. They have different purposes and destination fields; review drafts prepare outcomes, blockers and next steps.
2. Use the arrow beside a generation button for instructions or scope that apply only to this request. Save reusable preferences in AI writing preferences.
3. Check preparation, ready or failure status. Review the prepared text and apply the sections you want. Generation alone neither replaces every saved field nor reports to your PI.
4. When evidence is sparse, distinguish observed activity, your own notes and matters still to check. Add relevant facts yourself rather than presenting unperformed work as an outcome.
5. Check source links and dates. If records or access changed during generation, prepare again with current information. Editing saved text is separate from approving an unapplied draft.

5. Personalize the role, instructions and scope

1. Open AI writing preferences and choose role-aware or another available perspective. PI writing emphasizes supervision, review, decisions and support; researchers' results are not claimed as your own work. Platform administration alone does not imply a PI role.
2. Use common instructions for length, tone and priorities, for example: For each project, list at most three decisions or questions for researchers. This does not change your real role or access.
3. Save separate morning and evening preferences for plans and reviews. Use the generation button's options for a one-time change.
4. Choose organization by researcher, grant or project. Multiple selections within one category use OR; selections across categories use AND.
5. Narrow filters can reduce evidence or exclude private notes without a matching scope. Check filters when evidence is sparse. Restore defaults resets writing preferences, not saved daily records.

6. Group items and inspect numbered evidence

1. Structured AI content offers Group view to change how items are displayed. Switching researcher, grant or project grouping does not run AI again.
2. Expand the short researcher/project context to see whose work is involved. An item may have more than one relevant affiliation.
3. Follow [1], [2] and other numbered references to inspect evidence. A link does not grant access; removed or newly restricted sources cannot be reused.
4. Group view is available for structured drafts. Manually written or extensively restructured text may not offer the same grouping, but saved content can still be used to prepare actions.
5. Changing display order is different from changing generation preferences. Save a preference if future drafts should use that organization.

7. Continue with Act and its count

1. Choose Act beside an AI item or a saved manual field. For manual text, select the heading, paragraph or list item to act on. Empty fields have no content to execute.
2. Candidates use accessible citations, links and exact mentions. Inspect and change the destination if the match is wrong or several destinations are possible.
3. The number beside Act counts confirmed publications or personal execution records. It excludes drafts, AI requests, page views, retries and edit history. Select it to see linked actions.
4. Later editing does not overwrite earlier execution content or history. Read any source-changed notice and check the text to use for a new action.
5. Read Actions and follow-up for comments, messages and report handoffs. Pressing Act alone does not publish anything.

8. Review outcomes and continue tomorrow

1. Before leaving, record results, blockers/support needed and tomorrow's steps. Separate directly verified outcomes, other people's reports and unresolved questions.
2. In an action, open Results and next check to inspect replies and later activity and save your result. A new evening draft uses accessible actions, observations and personal notes within its evidence window.
3. The five checks record your own review of work registration, evidence, status, reporting needs and next steps. Checking them neither completes GitHub work nor sends a report.
4. Manage unresolved actions with a next-check date. Marking your follow-up finished removes it from future follow-up candidates; it is not research completion or PI approval.
5. For tomorrow's plan, review remaining follow-ups and newly collected evidence. Yesterday's saved review is not automatically rewritten when a new comment arrives.

9. Reports, requests and sharing with your PI

1. Use Reports & actions for received reports, incoming requests, sent requests and follow-ups. Request a report appears according to permission; ordinary researchers can answer received requests.
2. Personal goals and results are private by default. To share with a PI, choose the grant and the exact text, then review the actual recipient and content.
3. Shared check-ins appear in LUNA's received reports. A platform notification is created; email and push depend on recipient settings and delivery configuration. Shared does not guarantee email receipt.
4. If no participating grant with another PI is available, check the grant's responsible person and participation. The app does not infer a PI from an arbitrary researcher.
5. Submit a response to a report request from that request's response screen. Saving a personal note or sharing a separate PI check-in does not submit that response.

10. Reminders, scheduled drafts and AI usage

1. Set morning/evening times, workdays, time zone and delivery channels in Notification settings. Reminders and scheduled drafts require opt-in; opening the page does not enable automatic delivery.
2. Automatic morning/evening preparation creates drafts. It is different from applying all drafts or sending external comments on your behalf.
3. Check today's AI usage. The consultant limit is five per day for ordinary users and fifty for platform administrators; common personal AI limits and permission still apply.
4. Reading, manual saving, grouping and basic destination matching do not use AI allowance. Optional AI recommendations, wording and advice count according to request and execution policy. Failed attempts that started may count.
5. For pending, failed or limit messages, inspect usage and settings. Missing evidence and changed permissions are different problems; read the message and check pending work before repeating generation.

11. Use the same record through ChatGPT

1. Check the ChatGPT connection's additional consent for daily-note saves and research actions. Existing read access does not automatically allow writes.
2. For example: Read my manually written goal today and suggest a question for the cited issue. Show a draft; do not send it. Check the date, section, destination and evidence before choosing.
3. Save ChatGPT's wording as a LUNA action draft, then review the actual actor, destination and body before approval. The web and ChatGPT share records, revisions and execution history.
4. Text composed by ChatGPT is different from requesting LUNA AI generation. Reads and ordinary draft saves do not run extra LUNA AI; explicit LUNA AI requests use its allowance.
5. If tools are missing, check connection metadata and consent and start a new conversation. Use LUNA's web action screen when the host does not expose a required tool. Disconnecting does not undo existing publications.

12. Troubleshoot content, controls and translation

1. If Act is missing, first save nonempty content. Manual-field controls and structured AI-item controls appear in different places. Refresh the browser to load the current UI.
2. If Group view is missing, check whether the text is manual or retains structured AI items. Do not delete notes or tasks to recover controls.
3. English mode uses translated reading content with an original view. Browser support, language-model availability and cached translations affect readiness. Editors preserve the original text.
4. Quota errors, sparse evidence, source changes and access changes are different. Follow the specific message, add facts or reselect the destination. For uncertain delivery, check the original and action history first.
5. If needed, contact support with the page address, time, button and error text. Do not include passwords, tokens or private research content.

AI draft time limits and recovery

1. Daily plan/review drafts use the administrator call timeout.
2. Shared daily allowance, the daily research cap, per-request call budget and timeout are separate. Details and recovery shows applied settings and linked attempts.
3. A received but rejected response is encrypted for up to 24 hours and can be revalidated without AI. Personal notes are not automatically changed.
4. Retry proposes one new call; replaying the same approved review does not duplicate execution. Old failed responses that were never retained cannot be recovered.

Draft recovery submission and progress

1. Retry review and saved-response revalidation first record a receipt, then check evidence and access in the background.
2. Received, evidence checks, execution queue, AI generation and response validation reflect actual states. Accepted requests survive closing or reloading the page.
3. Confirm the review before accepting a new AI attempt. Saved-response revalidation does not call AI; resending the same request is deduplicated.
4. If disconnected, reload to find the accepted request. Only status reads retry automatically, never failed AI generation. ChatGPT uses the same access, review and call-budget rules.

Research daily limits and the overall allowance