examples-work-out/species-verification.md

Species Verification Document

The 4 app regions are: Midwest USA, USA, India, Africa.

Each vignette maps to one row in the motivation table and demonstrates a specific PestIDBot capability. The user question should naturally trigger that capability.

Database summary (rebuilt 2026-03-03): 1410 total chunks — Midwest USA: 388, USA: 913, India: 39, Africa: 70.

Latest comprehensive example run: example_results_20260303_110427.json (6 examples covering all 4 vignettes with cross-regional demos, using updated prompt — no "Document N" citations, no general knowledge filler)

---

1. Spotted Lanternfly (Lycorma delicatula)

Table highlight: Life-stage variation (eggs, instars, adults); risk of misidentification → Life-stage robust ID; abstains when uncertain (Stage 1)

Stage 1 reference: InsectNet Figure 2 in Chiranjeevi et al. (2025) shows life-stage classification results.

Stage 2 example (AgLLM):

• Region: Midwest USA (available in DB)
• Source PDF: ISU Yard and Garden Extension — Spotted Lanternfly (3 pages: life cycle, damage, reporting)
• User question: "How can I identify and report spotted lanternfly in Iowa?"
• What PestIDBot does: Returns life-stage descriptions (eggs, nymphs with color variations, adults) from ISU extension data, plus Iowa-specific reporting guidance.
• Example output: example_results_20260303_110427.json (Example 1)

Database status:

• Midwest USA: ✅ 6+ chunks from ISU PDF (3 pages ingested)
• USA: ✅ GPT-4o generated IPM info (5 chunks)

---

2. Witchweed (Striga asiatica)

Table highlight: Africa/India (smallholder constraints) → resistant varieties, intercrops; US → eradication, quarantine → Region-filtered recommendations (Stage 2)

Stage 2 examples (AgLLM):

• Regions: Africa (primary), USA (cross-regional)
• User question: "What is the most effective way to manage Striga in my maize field?"
• What PestIDBot does: For Africa, returns smallholder-appropriate IPM (resistant varieties, Desmodium push-pull, Fusarium oxysporum biocontrol, nitrogen fertilizer). For USA, leads with integrated cultural/chemical/biological approach (crop rotation, trap crops, imazapyr/glyphosate herbicides), supplements with African biocontrol research.
• Example output: example_results_20260303_110427.json (Examples 2-3: Africa vs USA)

Database status:

• Africa: ✅
• USA: ✅

---

3. Fall Armyworm (Spodoptera frugiperda) [replaced Old World Bollworm — more regions available]

Table highlight: India: biorationals, ETL monitoring, insecticide resistance awareness; USA: Bt crops, synthetic insecticides → Resistance-aware, resource-appropriate guidance (Stage 2)

Stage 2 examples (AgLLM):

• Regions: India (primary), USA (cross-regional), Midwest USA
• User question: "What are the recommended IPM strategies for managing fall armyworm in my corn field?"
• What PestIDBot does: For India, foregrounds ETL-based pheromone monitoring, biorationals (neem, Metarhizium), and emamectin benzoate with resistance warnings. For USA, leads with Bt crops, resistance management, mechanized scouting with ISU thresholds (25% infestation).
• Example output: example_results_20260303_110427.json (Examples 4-5: India vs USA)

Database status:

• India: ✅
• Midwest USA: ✅
• USA: ✅

---

4. Sahara Mustard (Brassica tournefortii)

Table highlight: Port-of-entry screening; partial/low-light images; misidentification historically enabled spread → Abstention → triage → escalate (Stage 1+2)

Stage 1 reference: WeedNet OOD detection and conformal prediction for abstention on poor images (Shen et al. 2025).

Stage 2 example (AgLLM):

• Region: Midwest USA (available in DB), USA (cross-regional)
• Source PDF: University of Nevada Extension — Identifying and Managing Sahara Mustard (3 pages: ID, habitat/impact, management methods)
• User question: "What mechanical, chemical, and cultural control methods are recommended for managing Sahara mustard infestations?"
• What PestIDBot does: Returns Midwest USA data (early chemical applications, rodent cache monitoring) plus USA cross-regional supplements (manual removal, mowing before seed set, glyphosate/2,4-D herbicides, native vegetation restoration).
• Example output: example_results_20260303_110427.json (Example 6)

Database status:

• Midwest USA: ✅ 11+ chunks from extension PDF (3 pages ingested)
• USA: ✅ GPT-4o generated IPM info