| PoS selection (phase × area) | Pulls the correct therapeutic-area row and current phase, multiplies the right remaining transition probabilities (incl. the NDA/BLA->approval step) to get the cumulative PoS, and cites the BIO/Wong-Siah-Lo source. | Uses the wrong area or phase, forgets the regulatory step, mixes up transition vs. cumulative PoS, or invents a success rate not in the table. |
| rNPV mechanics | Risk-adjusts the commercial value by the cumulative PoS AND discounts it to present value at the asset's discount rate, nets out (discounted) remaining R&D, and avoids double-counting risk between PoS and discount rate. Arithmetic is right. | Forgets to risk-adjust or to discount, double-counts risk, ignores remaining R&D, or makes arithmetic errors that move the rNPV materially. |
| Assumption quality | Uses sensible, stated assumptions for peak sales, the revenue ramp/years factor, discount rate (appropriate to stage), margin, and LoE timing — consistent with the returned inputs and the assumption conventions. | Picks an unjustified discount rate or revenue horizon, ignores LoE/margin, or makes assumptions inconsistent with the asset's returned parameters. |
| Deal recommendation & comparables benchmarking | Translates the rNPV into a concrete upfront + biobucket (+ royalty) range, anchors the upfront as a sensible fraction of rNPV for the phase, and benchmarks against the comparable deals returned for the area/phase. | Gives no concrete numbers, an upfront untethered from the rNPV, or ignores the comparable deals entirely. |
| Evidence faithfulness | Every PoS, peak-sales, cost, and comp figure traces to a tool output; no fabricated numbers; the model is stated transparently and is reproducible. | Hallucinates PoS/peak-sales/comp numbers, contradicts the returned inputs, or hides an opaque/irreproducible valuation. |