Comet spreads six sticky notes across the table. Each one names a privacy fix for the log.
"No names. Changing IDs. Totals only. Delete after 14 days. No training on questions. Tell students everything," she reads. "Let's add them all!"
Teacher Dev looks worried. "If I only see totals, how do I know which student is stuck on slope?"
Wren nods slowly. "Every fix protects something. What does each one give up?"
Comet pulls the pencil from behind her ear. "So it is a trade, not a free win."
Nova projects a table with three columns. "Would you like a hint?" she asks. "For each fix, write what it protects and what it gives up."
Wren hands you the first sticky note. "Our reviewer gets the last word on each one."
NIST names two ways to keep fewer details: de-identification and aggregation.
De-identification takes out or replaces the details that say who a record is about, such as names.
Aggregation combines records into totals. NIST says these methods can support privacy-enhanced AI systems.
But NIST also warns that, in some cases, privacy methods can lower accuracy.
These are the crew's own proposals in the story, with their trade-offs.
| Fix | What it protects | What it gives up |
|---|---|---|
| No names in the log | Anonymity: a row does not say who asked | Teachers cannot see which student needs help |
| IDs that change each week | A pattern cannot be followed for weeks | Nobody can see if one student is stuck for a month |
| Keep topic totals, not times | Late-night habits stay private | The builder cannot see when students study |
| Delete the raw log after 14 days | Old rows cannot leak later | No long history to look back on |
| Never train SH-1 on student questions | Questions cannot be memorized and repeated | SH-1 does not learn from how students ask |
| Tell students exactly what is kept | Control: students know and can choose what to ask | Some students may ask fewer questions |
| Statement | True or false? |
|---|---|
| De-identification and aggregation can support privacy. | ? |
| Privacy methods never change how accurate or useful data is. | ? |
| Every privacy fix gives up something. | ? |
| Removing names alone always stops anyone from working out who asked. | ? |
Thoughtful trade-offs, reviewer. Tomorrow is Impact Friday, and the panel has a hard question for you.