Qwen routes each request through six real tools, then deterministic Python rules
arbitrate the library evidence to one of three defensible levels:
GREEN
Entity locked — evidence is sufficient to act on the top candidate.
YELLOW
Direction locked — no single compound, but the chemical family holds; a next-step lab plan is generated.
RED
Refuses to guess — candidates diverge; the agent reports honest uncertainty.
Evidence Arbitration — deterministic
Candidates
Deterministic Rule Checks
— same input, same output
Peak Assignments
— literature-backed via FTIR.fun knowledge graph
🧪
Next Steps for the Analyst
The agent identifies the evidence boundary. The analyst runs additional tests and makes the final call.