AI Discovery Lab
Purpose
The Discovery Lab is the controlled intake for unfamiliar tools, research, techniques, claims, and one-off questions. It prevents two opposite failures:
- ignoring something useful because it is not yet in the curriculum;
- rewriting the curriculum around every new launch.
Entry questions
- What exact claim or question are we investigating?
- Who needs the answer and what decision will it support?
- What is the consequence of being wrong?
- What evidence is already available?
- What is the smallest safe investigation that could change the decision?
Investigation routes
| Route | Use when | Typical effort | Output |
|---|---|---|---|
| Quick scan | Need orientation or vocabulary | 15–30 min | Signal note |
| Evidence brief | Need a source-backed explanation | 1–3 hr | Evidence brief |
| Controlled experiment | Need behavior evidence | 2–8 hr | Experiment report |
| Deep dive | Decision is complex or consequential | Multi-session | Decision package |
Standard workflow
Capture → Clarify claim → Triage consequence → Gather evidence → Test → Interpret → Decide → Route
Decision outcomes
- Adopt: evidence supports use now under stated controls.
- Pilot: promising but needs bounded local evidence.
- Monitor: potentially relevant; current evidence or readiness is insufficient.
- Disregard: not relevant, not credible, or not worth the cost.
- Escalate: high-stakes or specialist review is required.
Routing after investigation
- Durable principle → propose a core-path update.
- Domain application → add to a domain collection.
- Product behavior → add or revise an ecosystem tutorial.
- Uncertain development → retain in the Lab with a reassessment date.
- Unsupported hype → record the dismissal rationale; do not create curriculum noise.
Cadence
Maintain a lightweight inbox continuously. Review signals monthly, conduct a quarterly synthesis, and escalate urgent security, deprecation, or policy changes immediately.