---
name: plan-garden-experiment
description: Use when a gardener wants to test a low-risk practice from a family memory or document whether it works in their own growing conditions.
---

# Turn a memory into a bounded observation

Input: a reviewed account or claim, plant/variety, broad location, season, growing medium, conditions and practical constraints. If context is missing, ask only for the details needed to avoid an unsuitable recommendation. Do not treat a story's popularity or age as proof.

Separate original recollection, proposed mechanism and actual observations. Consult current primary horticultural sources when advice depends on local climate, plant health, chemicals or regulations. Cite the source and applicable region. Never invent citations or claim a crop diagnosis from one photograph.

Propose one small, reversible, season-appropriate observation with a comparator where feasible. Specify baseline, intervention, what remains constant, duration, measurement units and stop condition. A winter record of light or soil moisture may be more appropriate than planting. Do not prescribe pesticide mixtures, medicinal remedies, food-preservation procedures, invasive-species releases or edibility decisions.

Output: a concise trial card, expected uncertainty, observation dates and the result that would change the decision. No scheduler is implied. Create a reminder only through an available scheduling tool when the user asks for one. Do not send external messages without instruction.

When results return, retain failures and null results. One success is an observation, not universal validation. Keep source account, trial and evidence separate; the current notebook cannot promote a claim to verified. No available write tool means return the card for manual storage.

Acceptance: testing a family's watering habit in two comparable containers produces a bounded measurement plan, not a universal rule. Unsafe mixtures are routed out of scope. Missing observations never become invented outcomes.