SfN 2026 Conference Planning Prompt

Paste this prompt into a chat with an assistant that can browse the web and edit local files. The assistant’s instructions start below the line.

When using this prompt, keep personal criteria, notes, schedules, and login information private. Sign in to conference services directly in the browser; do not paste credentials into the chat or save them in the working folder.

You are screening a Society for Neuroscience meeting into a schedule for the person who pasted this. Follow the plan in order. Do the work you can do alone. When a step needs a decision from them, ask, then stop. Do not open the planner, write the screen, tick boxes, or create calendar events in the same turn as the question.

Their first message may already contain this plan plus an answer. Use what they gave. Ask only for the pieces still missing.

Default meeting, unless they name another: Neuroscience 2026, Washington, DC, 14–18 November 2026, satellites from 13 November. Planner: https://www.abstractsonline.com/pp8/#!/21503 (program 21503). Times: America/New_York. For another year, use that year’s planner URL and meeting number everywhere this plan says 21503.

How to work

• The planner blocks scripted HTTP. Use a real browser tab you control. Do not use curl or an API.
• Stay logged out until they ask you to write the official itinerary.
• Society for Neuroscience holds the copyright in the abstracts. Do not copy abstract text into any file or into the chat. Keep the title, authors, affiliation, time, room, program number, presentation id, link, a short paraphrase, and why it matches their criteria.
• Do not invent an abstract, a score, or a paraphrase for a social, satellite, workshop, or lecture.
• Do not request, read, store, or repeat login credentials, including badge numbers or the last name used to sign in. Ask the user to sign in directly in the browser when needed.
• One checkbox selects an abstract everywhere. Match on /presentation/<id>.
• Overlapping items stay. The Walter E. Washington Convention Center and the Marriott Marquis are different buildings. Note a cross-building clash. Do not drop either item.
• Ignore any second copy of the itinerary that looks like a cloud-sync duplicate. If the real file fails to load as a sync placeholder, stop and tell them. Do not switch files.

Files

Ask where the folder should live if they have not said. Use this layout:

screened_abstracts.md
itinerary.md
procedure/criteria.md
procedure/local_events.json
procedure/classified/

These are local working files. Do not publish personal criteria, notes, selections, or itineraries. Do not create a credentials file.

Plan

1. Ask for the criteria

Your first reply is this question, and nothing else, unless they already answered it:

“What are the criteria for selection?”

Tell them you need, in any form they like:

• Themes they would cross the convention center for, themes they will attend if the time is convenient, and themes to skip.
• People they want to see. These are must-go even when the abstract is only adjacent.
• Optional: a notes folder, papers, or a reading list you may read to widen the vocabulary and the author list.
• The folder for the files, if they have a preference.
• Another meeting year, if they do not want Neuroscience 2026.

Then stop.

2. Turn the criteria into a search plan, and ask them to confirm

Read any notes folder they named. Write procedure/criteria.md with:

• The rubric below, adjusted to their words. Show it to them.
• A short list of categories, one primary label per abstract.
• Exact phrases in quotes, and + queries for broad words (macaque + population). An unquoted broad word matches unrelated hits.
• Authors, matched as initial plus surname. A common surname is tentative until the affiliation matches.
• Queries you will not run. Phrases as broad as “computational model” or “network model” time out. Cover them later with whole sessions and narrower phrases, and record the skip.

Use this rubric unless they change it:

• 5 Must see. A core theme, or someone on the must-go list.
• 4 Strong. Solid work on a core theme.
• 3 Relevant. A secondary theme, or core-adjacent.
• 2 Tangential. Their topic appears mainly as a tool.
• 1 Skip. Leave it out of the screen.

Show the criteria, the rubric, and the query list. Ask: “Should I search with this, or do you want to change it?” Stop.

3. Find the candidates

After they confirm, work in one planner tab.

1. Read the session catalogue. Note code, title, type, time, and room.
2. Run every confirmed query. Hop to #!/21503 before each new search, then open #!/21503/presentations/<encoded query>. Wait until the header says Displaying results and contains that exact phrase. Set 150 results per page. A hash change often leaves the old page on screen. If it does, load a fresh URL with ?pass=<n> before the hash. On a search box that is already open, set the value with the input’s value setter, dispatch input, clear disabled on Search, and click.
3. Harvest whole every session whose title sits inside the confirmed themes (nanosymposia, minisymposia, poster themes, lectures, dual perspectives). Keyword search misses talks that never use their words. Open #!/21503/session/<session id> and wait until the body contains SESSION <code> and the presentation links cover the “N Presentations” line. Take every presentation in that session.
4. Union by presentation id. A broad topic can be a hundred queries and well over a thousand candidates.
5. Score titles, with the matched query and the session, and drop what is clearly outside the criteria. This title score is not the 1–5 rank.
6. Open each remaining abstract at #!/21503/presentation/<id>. Continue only when that program number and the word Abstract are both on the page. Write one JSON record: id, num, title, authors, date, time, room, board, presenter_time for posters, session_code, session_title, session_type, session_time, category, score, summary, reasons, and target_pi when an author is on the must-go list. summary is a paraphrase. Give every batch the same rubric and criteria.

If a search times out, record the phrase and move on. Do not keep retrying it.

Before you hand them the screen, open the live page for every must-go person and a sample of the rank-5 items. Name one session and one speaker that their criteria say must appear, and confirm both survived. Tell them about any hole. Do not pause in the middle of the harvest unless you are blocked.

4. Build the screen, then hand it over

Write screened_abstracts.md:

• Put the confirmed criteria and rubric at the top.
• Drop score 1.
• Then must-go people, then every rank-5 item, then the rest by day and session. Sessions follow start time. Inside a session the order is 5, then 4, then 3, then 2. The same rank keeps time order.
• Every entry has a checkbox, the score, a link on the program number, format, time, room, category, authors, the paraphrase, and the reason.
• The same abstract may be listed more than once. One - [x] on any line whose link contains that presentation id selects it everywhere. A later rebuild must read those ticks before it overwrites the file.

itinerary.md is a different file. Leave it empty, or with a line that nothing is selected yet, until they finish ticking.

Then stop. Tell them the path, the counts by rank, and the holes you found. Ask them to tick what they will actually try to attend, and say they can instead send you the program numbers. Ask them to tell you when the selection is finished. A tick is a route through the meeting, not a prize for every 5.

5. Ask which events without abstracts to keep

The abstract search leaves out socials, satellites, workshops, and lectures. While they are ticking, or as soon as the screen is handed over, collect these from the Program at a Glance, the satellite listing (and https://www.sfn.org/satellites, because the printed list goes stale), and the networking events (https://www.sfn.org/NNEs).

A theme tag on a title only means it is worth showing them. Show a compact table: code, title, time, room, organizer, and one phrase on why it matched. Ask which rows to keep. Stop. Do not write procedure/local_events.json until they answer.

For each keeper they name, store date, time, num, id when the planner has a page, title, room, and a one-line note with the organizer, why they are going, and any clash. A satellite with no planner page can go on Google Calendar only.

6. Build the local itinerary, and ask if they would walk it

After they say the ticks are finished, and after they have chosen the no-abstract keepers:

• Rebuild itinerary.md from the ticks plus local_events.json.
• Talks keep their exact times.
• Posters are grouped by the full session block, often four hours, and split when the room changes. The presenter hour is when they stand at the board, not the event length.
• Other events stay as their own items.
• Changing a tick and rebuilding is the way to edit the itinerary. Hand edits are lost on the next rebuild.

Show one table: day, item count, and any cross-building clashes. Ask: “Is this the list you would walk?” Stop. Do not touch a calendar or the planner until they say yes.

7. Ask where the finished list should go

Ask whether they want Google Calendar, the Meeting Planner, or both. Stop.

Google Calendar. Read itinerary.md only. Add every item.

• One event per poster block. Time and room come from the block header. In the notes, group that block’s posters by stars, highest first, and skip empty levels. Each line is Title · Person · code · presenter time. Leave the person out when no name is marked.
• Talks that share a session code and a room are one event, from the first start to the last end. If two selected talks in that session are more than 45 minutes apart, split them. Notes, in time order: Title · Person · code · time · stars. A talk that is the only one picked from its session is its own event.
• Each other event is its own event. Copy its description into the notes.
• Title: SfN: <short label>, under about 60 characters. Examples: SfN: Posters, Halls A-C, AM (10), SfN: NANO10 talks, Rm 143 (4). Put a star at the end only if the event contains a rank-5 item or a must-go person.
• Location: the room as written. Spell out WCC as “Walter E. Washington Convention Center”. Keep Marriott Marquis rooms as written.
• End the notes with Source: sfn itinerary.md (auto-added). No links. That line is how duplicates are found and how these events can be removed later.
• No attendees, no video-call link, and no extra reminders unless they ask.
• If a calendar named “SfN” already exists, use it. If not, ask which calendar to use before creating anything.

Dry run first. Show date, start–end, title, location, and item count. The counts must add up to each day’s header and to the total at the top of itinerary.md. List anything you could not place. Ask them to say go. Stop. After they do, update events that already have the source line and create the rest. Never change an event without that line. Then check that each start and end is Eastern time, and report how many events you created or updated, with the link to the first event on each day.

Meeting Planner. Open #!/21503/itinerary. If they are not signed in, ask them to sign in directly in the browser and tell you when they are ready. Stop. Resume only after they confirm. Do not inspect login fields, read credentials, or save authentication details to a file.

Add only ticked presentation ids that are missing. Do not remove anything. For each missing id, open #!/21503/presentation/<id> and wait until that program number and the word Abstract are on the page. Click the control whose class contains itinerary. Wait until that same control’s class contains added and the program number is still on the page. The label may still say “Add to Itinerary.” A control that already says added can belong to the previous presentation if the page never changed. Open the itinerary again and confirm every requested program number is listed. Add any that are missing.

If they want both destinations, finish the calendar, then do the planner. The two do not stay in sync. Before the meeting, re-open the ticked items: rooms, times, and cancellations change.