Level 2

Sequencing beats one-shot prompting.

One-shot prompting asks AI to jump straight to the answer. Sequencing isolates each phase of a serious project so the final recommendation has evidence behind it.

The bad prompt

"Tell me the best competitions to apply to."

This prompt is too compressed. It asks for discovery, eligibility, historical analysis, fit assessment, and strategy in one move.

The better sequence

Build the competition database.
Analyze the last five or more years of winners.
Upload the student's resume and current project history.
Match the student to realistic opportunities.
Turn the best match into a roadmap.

Worked example

High school STEM competition roadmap.

Database phase

Collect competitions, eligibility, grade limits, citizenship rules, deadlines, costs, submission requirements, judging criteria, and links.

Winner phase

Find winners from recent years. Record project titles, methods, institutions, mentor involvement, datasets, and visible judging signals.

Pattern phase

Ask the harness to identify what separates ordinary submissions from winning submissions, with confidence ratings and source gaps.

Student-context phase

Upload the student's resume, project history, skills, school constraints, equipment access, and calendar.

Roadmap phase

Generate project directions, choose one, and convert it into milestones, datasets, GitHub commits, mentor needs, and verification checks.

Prompt template

Ask for the phase, not the whole future.

Alfred, we are in the database phase.

Do not recommend a project yet.

Create a competition database with these columns:
- competition name
- eligibility
- deadline
- cost
- required artifact
- judging criteria
- past winner links
- source URL
- confidence rating

Return the database structure first.
Then tell me what sources you need to search.

Good sequencing creates proof.

A database, memo, repo, deployed site, dashboard, or roadmap should come out of the process.

Good sequencing slows down false confidence.

The model has fewer chances to hide gaps when every phase has its own verification step.

Good sequencing teaches judgment.

Students learn why a recommendation is strong instead of accepting whatever the model says first.