CREW & READINESS / 2026 FIELD GUIDE
Find your role.
Build your team.
You don't need six people who do everything.
You need clear ownership, complementary skills and a plan everyone can follow.
Human roles own decisions. AI agents help with the work.
01 / BEFORE THE WEEKEND
Get oriented. Get set up.
The official site lists 14–15 November 2026. Confirm your event's exact start, deadline and timezone. Sources checked 17 September 2026.
Register and agree on the event
Create your participant account, find your local or Universal Event, and agree on the event with teammates. Record the deadline where everyone can see it.
2026 event hub ↗Choose a challenge you can finish
Read its full statement and resource list when available. Compare data access, team capabilities and one clear demo. Use challenge-scout with challenge-selection; use prior-art-analyst and award-targeting for context, not predictions of winning.
Official participant resources ↗Practise using the harness
Choose Claude, Codex or another model in the first-review tutorial. Read the shared rules and run the planted-error review. A correct answer alone does not prove a plugin loaded.
Nine AI roles and eleven skills are available. Use only what the task needs.
Test only the data access you need
Data owners should fetch and inspect a small real sample. Check units, missing values, quality flags, issue time and whether you need observations or forecasts.
Earthdata Login and NASA API access are optional unless your selected sources require them. Never put keys in this board.
Check submission rules and rights
Read the current 2026 participant terms: review eligibility, permissions, source attribution, licensing and protections for minors. Cite your data and third-party work. Do not expose private information.
The submission guide and judging guide may still show the previous edition. Confirm current instructions before building or submitting; do not apply last year's limits automatically.
Agree on what “good” looks like
Use Impact, Creativity, Validity, Relevance and Presentation as review lenses from the historical rubric. Name the beneficiary, use real data, explain limitations, and show a working path.
Never invent users or results. Label mock-ups and proposed behavior honestly. Use science-validator and submission-auditor before recording.
02 / WHO OWNS WHAT
Six responsibilities. Your way to contribute.
Small teams can combine roles.
Keep Narrative owned from the beginning.
03 / CAPABILITY SELF-CHECK
Start with what you can show.
Rate what you can do today. “Not assessed” stays unknown. Suggestions are a starting point for a team conversation, not a test score or personality diagnosis.
04 / THE TEAM BOARD
Make readiness visible.
Saved only in this browser on this device. Anyone using this browser can edit these rows. Export a file to share a snapshot; importing replaces the current board. Nothing here is sent to a team server.
Up to six people in this planner; check the current event's team-size rule. “Ready” means these self-reported setup checks are complete—not official registration or submission approval.
What does each readiness check mean?
- Registration: personally checked registration, team event and deadline.
- Harness practice: completed one useful workflow and reviewed its output.
- Tooling: tested the tools required for the assigned role on the actual device.
- Shared docs: read the shared rules and team runbook.
- Data access: mark Earthdata/API as not needed, needed but pending, or tested. Only needed accounts affect readiness.
05 / WORK BACK FROM THE DEMO
A weekend plan that can shrink.
These are planning checkpoints, not official event times. Fit them to your actual deadline, availability and rest breaks.
- AgreeOne challenge, one promise, named owners. Narrative starts now.
- Prove the dataGet a small slice on disk. If it fails, change source or cut scope.
- Connect the pathMake one complete example work before expanding features.
- Freeze & reviewLead calls the cut. Science review happens before filming.
- Submit & confirmFinish the page and demo, check public links, keep a backup and confirm submission before the local deadline.
When behind: cut unseen features, extra model complexity, extra sources, then polish. Protect truthful evidence, a usable demo and the submission itself.
Explore the AI toolkit ↗