Official Judging Rubric.
100 points across five criteria. Every format (app, essay, film, design, or otherwise) is scored on the same scale, and judges are instructed to score the thinking, not the medium. The 300-word Rationale carries equal weight for every entrant.
| Criterion | Points | Core question |
|---|---|---|
| 01 · Insight & Originality | 25 | Does this submission contain an idea we haven't seen fifty times? |
| 02 · The Acceptance vs. the Refusal | 25 | What would you automate, and what would you refuse to automate? |
| 03 · Depth of Reasoning | 20 | Has the entrant thought past the first-order effects? |
| 04 · Execution & Craft | 20 | Is the work well-made for its chosen format? |
| 05 · Communication | 10 | Can we understand it, quickly? |
| Total | 100 |
Score bands, criterion by criterion.
01 · Insight & Originality
25 pointsDoes this submission contain an idea we haven't seen fifty times?
| Score | Descriptor |
|---|---|
| 21-25 | A genuinely novel angle, or a familiar idea reframed so sharply it feels new. Judges want to argue about it afterward. |
| 15-20 | A solid, specific idea with at least one original element or unexpected connection. We're looking for nuance. |
| 8-14 | Competent but familiar. Ideas the judges have encountered in mainstream ed-tech discourse. |
| 0-7 | Generic. “AI tutor personalizes learning; teachers provide human connection” with no further development. |
02 · The Acceptance vs. the Refusal
25 pointsWhat would you automate, and what would you refuse to automate?
| Score | Descriptor |
|---|---|
| 21-25 | The refusals and acceptances are specific and surprising, defended with real reasoning (not sentiment), and structurally connected to the design. Remove them and the whole submission changes. |
| 15-20 | A clear, specific acceptance and refusal with genuine argumentation, though the connection to the design may be loose. |
| 8-14 | A refusal is named but defended with platitudes (“human connection matters”) rather than reasoning, or it reads as a list rather than a position. |
| 0-7 | The refusal is missing, an afterthought, or so broad it's meaningless (“I'd never automate teaching”). |
03 · Depth of Reasoning
20 pointsHas the entrant thought past the first-order effects?
| Score | Descriptor |
|---|---|
| 17-20 | Engages seriously with trade-offs, failure modes, or counterarguments. Acknowledges what the design costs, not just what it gains. |
| 12-16 | Some awareness of trade-offs or limitations; addresses at least one obvious objection. |
| 6-11 | Purely first-order thinking. The design is presented as having only upsides. |
| 0-5 | No evidence of reasoning beyond the initial idea. |
04 · Execution & Craft
20 pointsIs the work well-made for its chosen format?
| Score | Descriptor |
|---|---|
| 17-20 | Exceptional craft: polished, deliberate, and complete for its medium. For apps: it works. For essays: it's well-written. For films: it's well-made. |
| 12-16 | Solid execution with minor rough edges that don't obscure the idea. |
| 6-11 | Noticeable gaps in execution (broken features, unclear writing, unfinished sections) that get in the way. |
| 0-5 | Execution problems make the idea hard to evaluate at all. |
Judge's note: Craft is scored relative to the format's demands, not its production cost. A tightly argued 1,500-word essay can earn 20/20; a feature-rich but confused app can earn 8/20. Judges may also weigh how thoughtfully the entrant used AI (per their AI Use Disclosure) as part of this criterion.
05 · Communication
10 pointsCan we understand it, quickly?
| Score | Descriptor |
|---|---|
| 9-10 | The core idea lands within minutes. The Rationale is sharp. Nothing requires re-reading. |
| 6-8 | Clear overall, with occasional confusion or clutter. |
| 3-5 | The idea is in there, but the judge has to dig for it. |
| 0-2 | Unclear what is being proposed. |
- The higher score on Insight & Originality.
- Judges' panel discussion and vote.
How scoring runs.
- 01
Screening
Organizers check each entry for completeness (the work, the Rationale, and the AI disclosure) and rules compliance. Incomplete entries get one email and 48 hours to fix.
- 02
Scoring
Every eligible entry is independently scored by at least two judges using the rubric, and the scores are averaged. If two judges' totals differ by more than 20 points, a third judge scores the entry and the outlier is dropped.
- 03
Finalist panel
The top 10 entries are re-read by the full judging panel together, and judges may adjust scores after discussion. Finalists may be invited to a brief 10-minute live or video Q&A, used to verify authorship and probe reasoning, not to re-pitch.
Conflicts of interest. Judges recuse themselves from scoring any entrant they know personally or professionally.
The panel. 5 to 7 professionals from the AI and education fields.
Submissions due September 25, 2026 (11:59 PM Pacific) · Results October 3, 2026 · aivanguard.org/competition