Run a nine-salt SNAr screen and rank by measured conversion
This is a task package rather than a model case study. The agent must execute a real screening protocol on a simulated laboratory platform, one batch at a time, then rank nine candidate salts by their measured 10-minute conversion. The verifier accepts only answers tied to an actual recorded batch, so a literature-based ordering cannot pass.
The task
The reaction is an aromatic nucleophilic substitution (SNAr, addition–elimination): an aniline nitrogen attacks the activated aromatic carbon through a Meisenheimer intermediate, followed by defluorination. The substrate is activated by an ortho cyano group and a meta nitro group. Nine candidate salts are supplied as blind codes under fixed, pre-registered conditions.
- Screen
- Nine candidate salts, single 10-minute time point
- Readout
- LC quantification against an internal standard, reported as conversion
- Output
- A ranking of all nine salts by measured conversion
- Platform
- A simulated instrument: readings are synthetic task data, not real wet-lab measurements
The salts act on ion pairing and anion solvation in a single-phase system; this is not a liquid–liquid phase-transfer setup.
How the platform is used
- The agent must use the platform's own operation interface to execute batches; it cannot simply declare a ranking.
- One vial corresponds to one time point; the same vial cannot be sampled twice.
- Each executed batch receives a unique run identifier and LC values with small, deterministic batch-to-batch variation.
- The verifier requires the submitted ranking to match one recorded batch, and that batch must cover all nine blind codes within 10 minutes.
- An ordering derived only from the literature is not accepted.
Status
The task, simulator and verifier are complete. No model evaluation is published in this release. When a run is published, it will be reported with its batch record and conversion measurements.
Availability
The task specification and evaluation protocol are available to reviewers on request. The full task package is not published here to keep it available for future evaluation.