chemistry-ranked-entities-and-property-curves

# Print the named-entity ranked list and the property-versus-coordinate curve -- chemistry's two most-computed, least-reported artifacts Two chemistry deliverables are routinely computed and never printed. Plan both into the report skeleton. First, the ranked list. Whenever the method scores individual entities -- per-residue scans, per-atom or per-fragment attributions, per-pose scores, per-molecule rankings -- the deliverable is an explicit table of the leading entities by chemical identifier with their scores and units, plus the scan's bookkeeping: how many entities were scanned and the observed minimum and maximum. An aggregate ranking metric (AUC, precision@k, a correlation) does not substitute for the named list. If a per-entity file exists in your outputs, sorting its head into the report is the result. Then map those entities back onto chemistry -- which contacts, which functional groups, which charges or multipoles -- and state whether that is what a chemist would expect. Second, the curve. Where a property depends on a governing physical or protocol coordinate -- bond length, intermolecular separation, cutoff radius, temperature, training-set size, number of sampling steps -- sweep it and plot the continuous curve with the reference overlaid, then read the derived constants off it and report their errors: equilibrium geometry, well depth, barrier height, asymptotic decay exponent, sum rules and conservation checks. A bar chart of RMSE by model variant answers a d...