Every plant identification now asks for your confirmation before it's treated as ground truth — even when confidence is high. The app would rather ask and be right than assume and be wrong. A single misidentification that goes uncorrected can cascade into bad care advice. The confirmation step closes that gap.
Appears on every scan result, regardless of confidence level. "Yes" confirms and saves — the ID is marked with a check badge in history. "No" opens the correction flow — candidates if available, or free text entry.
When the app's candidates don't match your plant, you're no longer stuck. Tap "None of these" and type the actual plant name. Or tap "I don't know" to save it as Unknown with a follow-up task so you can come back to it.
Your confirmed name replaces the model's guess everywhere in the app — history, garden profile, care advice.
After confirming an identification, a quick thumbs up/down prompt appears. Same after a diagnosis result. These aren't surveys — they're the signal that tells us what the app is getting wrong in your specific zone and season.
Thumbs down on an ID prompts you to enter the correct name. Thumbs down on a diagnosis lets you add a one-line comment. Fire-and-forget — never blocks the app, fails silently if you're offline.
Daily Garden Brief, Companion Planting, What to Do This Week, Community Insights, and Ask Me Anything all shipped gradually over the past two weeks. As of Build 20, they're on for all testers. Ask Me Anything now knows about every new feature — ask it "what's new?" or "how does feedback work?" and it answers like a knowledgeable friend, not a manual.
The confirmation gate, escape hatch, and thumbs up/down aren't just UX polish. They're the foundation of a feedback loop that makes the app more accurate over time. Every correction is stored anonymously with your USDA zone and the current month. When patterns emerge — plants consistently misidentified in your region during a specific season — they inform targeted updates to how the app reasons about identification.
The corrections don't auto-update anything; a human reviews the patterns first. But the signal is real, and it compounds. The privacy contract hasn't changed: no user IDs, no precise location, no device fingerprints. Just zone-level, month-level aggregate data that belongs to the community.