Methodology and what it does not claim.

  • Every output traces through a visible formula from an input you can see and question. There are no hardcoded numbers standing in for calculations, and every figure in your readout can be traced back to its source.

• The models self-check. Probabilities must sum to 100%. Valuations must stay in order. Milestones achieved cannot exceed milestones that exist. Every benchmark lookup must resolve to a real row. Violations are flagged on the summary page.

• Valuation inputs are derived, not chosen. Scenario values are looked up from a benchmark table keyed to device class, regulatory pathway and milestone stage.

• Stage-level magnitudes are corroborated against published, dated industry sources, cited on the face of the model.

• One module refuses to produce a number at all when the asset cannot support one. A discounted-value model on a pre-clinical device is arithmetic dressed as evidence, so it gates itself off and names the missing evidence instead.

 • At Series A, the same discipline prices a round instead of a cap: reimbursement probability is built from a coverage-pathway base rate and adjusted on the evidence you can show, and a risk-adjusted (rNPV) valuation carries it to a pre-money.
• The valuation returns a sensitivity range, not a point. A single figure would imply a precision the four inputs behind it — peak share, exit multiple, discount rate, probability of success — do not support.
• The round is priced on your actual preferred stack, and the preference terms are run through the exit waterfall, so dilution and what the terms cost you at exit are derived, not asserted. 

   Every engagement logs what it estimated against what happened: at seed, the residual risk, the probability and price we supported, the price agreed, and the milestone outcome; at Series A, the reimbursement read, the pre-money the evidence supported, the terms, and how the round actually cleared.
The two stages calibrate separately — a seed cap outcome doesn't validate a Series A rate, or the reverse. So the thresholds are per stage: at roughly thirty engagements in a stage, we can say where those estimates run hot or cold; at a hundred, the base rates for that stage stop being borrowed from the literature and start being ours. We publish what each shows, including the results that don't flatter us.

That is the honest answer to the obvious objection. The model is not validated yet. Here is the validation program, here is where it stands, and here is what we do not claim in the meantime.