Global controls
Load your own parameters
Upload an Excel workbook to replace the reference phase values, scenario multipliers, modality adjustments, therapeutic-area substitutions, commercial archetypes, crowding overlay and V cases with your own. Download the template below (it contains the current values), edit it, and upload it back. Sheets that are missing or empty are left unchanged.
Pipeline picture
Selected scenario summary
Corrected historical comparator
Phase-level computed results
Shapley decomposition of the change since 2010
Exact grouped Shapley attribution of the difference in capitalized cost between the corrected inflation-adjusted Paul et al. comparator and the selected scenario at the current discount rate. Groups: discovery/preclinical unit costs, clinical/regulatory unit costs, cycle times and transition probabilities. Contributions include interaction effects and sum exactly to the total change.
Tornado sensitivity plots
Each bar varies one parameter at a time around the selected scenario, product type and therapeutic area, while all other inputs stay fixed. Costs and cycle times are varied by a relative percentage; p(TS) is also varied by a relative percentage and capped to stay between 0.0001 and 0.9999. The NPV tornado additionally varies peak gross profit by the same percentage and the discount rate over the stated range.
Editable R&D phase values for the selected scenario
Editable scenario transformations
Forward scenarios are defined as phase-specific multipliers on reference cost and cycle time plus additive percentage-point changes to transition success (clipped to 0.0001–0.9999). AI-A uses the lower end of plausible cost and time effects with no transition-success change; AI-B uses the upper end and adds bounded transition-success improvements as an explicit upper bound; China execution changes timing only; AI-A + China uses the AI-A cost multipliers and the faster of the AI-A and China time multipliers phase by phase. All cells are editable.
Editable modality-specific R&D adjustments
Optional R&D overlay by product type. The manuscript withdrew the universal biologic R&D modifier used in the original submission, so the biologic row defaults to no change (Phase III cost ×1.0, Δp(TS) = 0) and biologics differ only through the commercial archetype. The rows for monoclonal antibodies, CAR-T, siRNA/RNAi and gene therapy substitute BIO modality transition rates for the four clinical phases and are transition-rate robustness checks, not full modality cost models. All cells are editable.
Editable therapeutic-area substitutions
Applied on top of the selected scenario and product type. The four clinical p(TS) values replace the scenario's clinical transition probabilities (a blank cell leaves the scenario value unchanged); the cost multipliers scale Phase I–III cost per program and default to 1. The first row is the canonical all-therapeutic-area reference and is a no-op. All cells are editable, and the selected therapeutic area drives every figure and table in the app.
Current scenario NPV summary
Competitive crowding and value compression — the productivity paradox
Greater R&D efficiency frees resources for more work in process, and more programs tend to converge on the same validated targets and commercially attractive indications. The crowding overlay below applies to whichever scenario is selected. V stays the perceived opportunity (the peak annual gross profit the product would earn as the sole entrant); the realized peak is V × Φ, where Φ(N,n) = Nβ · s(n,N) · (1−π)N−1. The manuscript's equal-share benchmark is the special case φ = 1, π = 0 and no differentiation window, which gives Φ = Nβ−1 (Figure 8). Order-of-entry share decay (φ), per-entrant price erosion (π) and a differentiation window (D) that ends economic exclusivity before the observed competition onset are optional extensions. Setting N = 1 reproduces the uncrowded result exactly.
Equal-share crowding sensitivity (manuscript Figure 8)
Break-even stand-alone peak gross profit required per entrant for the selected scenario, product type and discount rate, as the number of entrants N increases, for category-expansion exponents β = 0.5, 0.8 and 1.0: V*(N) = V*(1) · N1−β. The β values are illustrative sensitivities, not historically calibrated estimates.
The paradox, decomposed
Each forward scenario is built in two steps on top of the 2026 reference. Step 2 applies the scenario's R&D gains with V uncompressed. Step 3 adds the value compression implied by the crowding overlay above (same N, n, β, φ, π and D for every scenario). Read down the capitalized-cost, launch-time and p(launch) columns, then read down the NPV column. Computed at the selected product type, therapeutic area, V and discount rate.
NPV table by scenario and product type
The peak-gross-profit columns are editable in the "Editable V cases" table below. With the crowding overlay active the column headings are the perceived V; the realized peak is V × Φ.
Break-even V
Editable commercial lifecycle values
Years to peak and competition onset are anchored to observed market behavior (Wouters et al., 2024a,b); the erosion parameters and horizon are model assumptions. Editing a value immediately updates all curves and tables.
Editable V cases
These values set the columns in the NPV table and the x-values in the NPV chart.
Protection-window check
Equations implemented
R&D model
OOP_i = WIP_i · C_i
m_i = Σ_{k<i} T_k + T_i/2 ; L = Σ_i T_i
PV0_R&D = Σ_i OOP_i / (1+r)^{m_i}
CAP_launch = Σ_i OOP_i · (1+r)^{L-m_i} = PV0_R&D · (1+r)^L
Scenario s: C_i = C_i^ref · m^C_{i,s} ; T_i = T_i^ref · m^T_{i,s} ; p_i = clip(p_i^ref + Δp_{i,s})
Commercial model
U(a) = min(1, a / t_peak)
E(a) = 1, a ≤ M ; E(a) = F + (1−F)·exp(−λ(a−M)), a > M ; λ = −ln(0.05)/d
M = min(competition onset, differentiation window D)
GP(t;V) = Φ · V · U(a) · E(a) · 1_{t≥L}
NPV0(V) = ∫_0^{L+H} GP(t;V)/(1+r)^t dt − PV0_R&D ; V* = PV0_R&D / PV_GP(V=1)
Crowding / value compression
Φ(N,n) = N^β · s(n,N) · (1−π)^{N−1}
s(n,N) = φ^{n−1} / Σ_{j=1..N} φ^{j−1} (equal shares when φ = 1)
Equal-share benchmark: Φ = N^{β−1} ; V*_N = V*_1 · N^{1−β}
Terms
N products launched into the crowded area; n this program's order of entry. β category-expansion exponent (β = 1 means total category value grows in proportion to entry, so no dilution under equal shares; β = 0 is a fixed pie). φ rank share decay, giving earlier entrants a durable share advantage. π net-price give-up per additional entrant. D years of economic differentiation after launch, which caps the protected period. t_peak years to peak, M competition onset after launch, F residual floor, d years to reach 95% of the post-competition decline, H commercial horizon.
Export
Frozen 2026 probabilistic reference and scenario propagation
The 24 phase parameters (8 costs, 8 cycle times, 8 transition probabilities) are treated as uncertain rather than as a single corner case. Draws use evidence-quality-specific distributions: empirical calendar-year vector bootstrap for A-rated clinical time and p(TS), a source-family mixture of lognormals for B-rated clinical costs, beta-PERT distributions for weaker discovery/preclinical inputs, and a correlated discovery-success constraint. Each scenario's transformation (and the current product-type and therapeutic-area selection) is applied to the same base draw, so scenario comparisons are paired. With the default settings (seed 20261002, 20,000 draws, 11%, small molecule, canonical therapeutic area) the results reproduce the frozen run reported in the manuscript and Supplementary Information.
Deterministic reference and probabilistic outputs — selected scenario
Scenario propagation across the same draws
Source-conditioned capitalized cost (manuscript Figure 2a)
Audited 24-parameter specification
Evidence grade: A = recent/direct data with a close definition match; B = direct empirical evidence with substantial source heterogeneity; C = indirect or imperfectly mapped evidence; D = legacy/heuristic evidence with little contemporary phase-specific public data. Ranges are evidence envelopes, not confidence intervals.
Modeling assumptions for uncertainty
Clinical time: one IQVIA calendar-year vector from 2020–2024 is sampled jointly across Phases I–III and regulatory review, preserving observed cross-phase year effects.
Clinical p(TS): one Citeline/IQVIA calendar-year vector from 2021–2025 is sampled jointly across Phases I–III and regulatory submission.
Discovery/preclinical p(TS): phase-specific beta-PERT draws are shifted together on the logit scale so their product matches a sampled discovery-to-clinic yield, modeled as beta-PERT(0.20, 0.32, 0.45). This keeps the model near the point-reference product (30.5%) while respecting the published ~34% industrial discovery-to-clinic estimate as an external calibration.
Scenarios: the multipliers and transition-success increments of each scenario are applied to the base draw at runtime; the break-even values propagate the draw through the selected commercial archetype with commercial assumptions held fixed.
Export uncertainty results
The analyses on this tab reproduce manuscript Figures 4 and 5 and the China timing bounds in Supplementary Part S2. They use the current reference phase values and scenario transformations, and the selected product type and therapeutic area, at the discount rate shown in each panel.
Cells are capitalized cost ($B) when all phase durations are scaled by a common multiplier and the rate is varied. The reference cell is in bold.
Phase I transition success is raised by 10 percentage points while Phase II success is progressively reduced; cost and time are held at the selected scenario's values.
Every therapeutic area in the editable table, with cost and time held at the selected scenario's values. Values are editable on the Pipeline tab.
Not a full modality cost model. CAR-T, RNA and gene therapy have materially different manufacturing, preclinical and clinical economics. Monoclonal-antibody preclinical cost is varied 1–2× as a robustness check.
Additional nontechnical attrition is represented as an extra continuation factor after the technical-regulatory chain, solely to show sensitivity.
China execution timing bounds (Supplementary Part S2)
Exploratory timing-reduction bounds; the midpoints generate the China execution scenario. No success-rate or cost benefit is assigned to China.
Instructions manual
This page explains what the tool computes, how to use each tab, how to load your own parameters, and how to reproduce the numbers in the manuscript. Costs are in 2025 US dollars (millions in the phase tables, billions in the summaries). Every result on every tab is recalculated from the current parameters whenever a control or an editable cell changes.
Contents
Quick start
- Leave the global controls at their defaults (2026 reference, small molecule, no therapeutic-area substitution, 11%, V = $5B/year). The Pipeline tab then shows the manuscript's reference case: $2.33B out-of-pocket, $5.25B capitalized, 14.4 years, 2.31% launch probability.
- Change the scenario to see the AI and China cases; change the product type or therapeutic area to apply overlays.
- Open Tab 2 for program-start NPV, break-even value and the crowding overlay.
- Open Tab 3 and press Run Monte Carlo (about 10–20 seconds) for uncertainty envelopes.
- Edit any yellow cell to change an assumption; press Reset to manuscript defaults to return.
1. What the model does
The tool implements the eight-stage pipeline model of the article. Each stage i (Target-to-Hit, Hit-to-Lead, Lead Optimization, Preclinical, Phase I, Phase II, Phase III, Submission-to-Launch) has three parameters: cost per program entering the stage (Ci), start-to-next-stage duration (Ti) and transition success (pi). From these it computes, for one expected launch:
- Programs per launch at each stage (1 / product of downstream transition probabilities) and the launch probability (product of all eight pi).
- Out-of-pocket cost per launch: the sum over stages of cost per program times programs per launch.
- Capitalized cost per launch: the same cash flows compounded to the launch date at the discount/capitalization rate, assuming each stage's spend occurs at its midpoint. PV of R&D at Target-to-Hit is the same quantity discounted to program start.
- Program-start NPV and break-even peak gross profit (V*): the commercial model turns a mature annual gross profit V into a gross-profit curve (linear uptake to the peak year, flat until the observed competition onset, then exponential decay to a residual floor), discounts it to program start, and subtracts the PV of R&D. V* is the V at which NPV is zero.
The 2026 reference is the audited parameter set of Table 1 in the article. The four forward scenarios (AI-A efficiency, AI-B predictive validity, China execution, AI-A + China) are not separate parameter sets: each is a vector of phase-specific cost multipliers, time multipliers and additive transition-success changes applied to the reference at run time. Editing the reference therefore moves every scenario. The Paul et al. (2010) corrected comparator is an absolute-value scenario (inflation-adjusted, 13.5-year cycle) kept for historical comparison; it does not inherit reference edits. The Equations panel on Tab 2 gives all formulas.
2. Global controls
| Control | What it does | Default |
|---|---|---|
| R&D scenario | Selects the parameter set shown on Tabs 1 and 2, the scenario whose histograms are shown on Tab 3, and the scenario used by the grid, risk-shifting, therapeutic-area and modality panels on Tab 4. Tables that compare scenarios always show all of them. | 2026 reference |
| Product type / commercial archetype | Chooses the commercial lifecycle archetype (small molecule or biologic) and any R&D overlay defined for that product type in the modality table on Tab 1. Small molecule and Biologic carry no R&D overlay by default; Monoclonal antibody, CAR-T, siRNA/RNAi and Gene therapy substitute BIO modality transition rates for the four clinical stages. | Small molecule |
| Therapeutic area | Replaces the four clinical transition probabilities with BIO/QLS/Informa 2011–2020 values for that area (cost and time unchanged). The first entry is a no-op. | All-TA canonical reference |
| Discount / capitalization rate | The rate used for capitalization, discounting and NPV on Tabs 1, 2 and 4. Tab 3 has its own comparison rate. | 11% |
| Selected peak gross profit V | The mature annual gross profit of a single product used for the NPV summary, the gross-profit lifecycle, the NPV tornado and the paradox table. It is the perceived value when the crowding overlay is active. | $5B/year |
The line under the controls summarizes the current selection, including the crowding overlay if it is active. Overlays stack in a fixed order: scenario transformation, then product-type adjustments, then therapeutic-area substitution.
3. Tab 1: Pipeline model
| Panel | Contents and use |
|---|---|
| Pipeline picture | The Paul-style funnel for the current selection: capitalized and out-of-pocket cost per launch by stage, share of total cost, cycle time, cost per program, programs needed per launch and p(TS). The header gives the totals. Download as PNG with the button. |
| Selected scenario summary | The five headline metrics (OOP, capitalized cost, PV of R&D at Target-to-Hit, launch probability with Phase I starts per launch, launch time) and, below them, the corrected Paul et al. comparator at the current rate for side-by-side comparison. |
| Phase-level computed results | Effective (post-overlay) parameters and the stage-by-stage accounting, with totals. |
| Shapley decomposition | Exact grouped Shapley attribution of the change in capitalized cost from the Paul comparator to the selected scenario, in four groups: discovery/preclinical unit costs, clinical/regulatory unit costs, cycle times, transition probabilities. Contributions include interaction effects and sum to the total. With the reference selected this reproduces Figure 2b (+$1.51B transitions, +$0.74B clinical costs, +$0.34B cycle times, ≈0 discovery). |
| Tornado sensitivity plots | One-at-a-time perturbations of all 24 parameters around the current selection. Three outputs: capitalized cost, PV of R&D at Target-to-Hit, and program-start NPV at the selected V (which also varies V and the rate). The percentage perturbations and the rate range are editable; press Recalculate or edit any field. Light bars are outputs below the baseline, dark bars above; the twelve largest spans are shown. The cost and NPV tornados reproduce Figure 3 at the defaults. |
| Editable R&D phase values | With the reference or the Paul comparator selected: the absolute cost, time and p(TS) per stage, editable. With a forward scenario selected: a read-only view of reference value, multiplier and effective value for each stage. |
| Editable scenario transformations | For each forward scenario and stage, three stacked cells: cost multiplier, time multiplier, Δp(TS) in percentage points. Edit these to redefine a scenario. |
| Editable modality-specific R&D adjustments | Per product type: commercial archetype, preclinical and Phase III cost multipliers, optional clinical p(TS) substitutions (blank = keep scenario value) and additive Δp(TS) for the clinical stages. Use this, for example, to restore a biologic Phase III cost premium. |
| Editable therapeutic-area substitutions | Per therapeutic area: the four clinical p(TS) values and Phase I–III cost multipliers. |
4. Tab 2: V / NPV model
| Panel | Contents and use |
|---|---|
| Current scenario NPV summary | NPV at Target-to-Hit for the selected V, PV of gross profits, PV of R&D, break-even V and launch timing, all for the current selection and crowding overlay. |
| NPV versus peak annual gross profit | One line per scenario at the selected product type and therapeutic area; x-values are the V cases. The marker on the zero line is each scenario's break-even V. Reproduces Figure 7a. |
| Commercial gross-profit lifecycle | The gross-profit curve for the selected V on the absolute program timeline (year 0 = Target-to-Hit). The note below gives launch, peak and competition years. |
| Competitive crowding and value compression | The crowding overlay. Entrants N is the number of products launched into the class (1 = no crowding). Expansion β sets how total class value grows with N (Nβ). With share decay φ = 1 and price erosion π = 0 (the defaults) every entrant receives an equal share and the multiplier is Φ = Nβ−1, the manuscript's benchmark. Setting φ < 1 gives earlier entrants a larger share (then Order of entry n matters); π > 0 adds per-entrant price erosion; Differentiation window D ends economic exclusivity D years after launch if that is earlier than the observed competition onset; the floor and years-to-floor overrides replace the archetype's erosion parameters. The metrics, the NPV-versus-N chart and the share-by-rank chart update with every edit. |
| Equal-share crowding sensitivity | Break-even stand-alone V per entrant for N = 1 to 20 and β = 0.5, 0.8, 1.0, for the current selection; reproduces Figure 8 and the $1.76B/$1.95B (β = 0.8) and $3.02B/$3.90B (β = 0.5) values at N = 6 and 10. |
| The paradox, decomposed | For each forward scenario: (1) the reference, (2) the scenario's R&D gains with V uncompressed, (3) the same R&D with the crowding overlay applied. The R&D columns improve from step 1 to 2 while the NPV column falls from step 2 to 3 when N > 1. Raise N in the overlay to see the effect. |
| NPV table and Break-even V table | All scenarios × all product types at the V cases, and the break-even values with their components (PV of R&D, PV of gross profit at V = $1B). |
| Editable commercial lifecycle values | Per archetype: years to peak, competition onset after launch, residual floor, years to reach 95% of the decline, horizon. Defaults are the Table 4 values (8/12.6 years for small molecules, 13/20.3 years for biologics). |
| Editable V cases | The V values used as NPV table columns and chart x-values. |
| Protection-window check | Launch year, peak year, competition onset and effective protection for each product type; a bold protection end means the differentiation window binds. |
| Equations implemented | The R&D, commercial and crowding formulas as coded. |
5. Tab 3: Uncertainty
The Monte Carlo treats the 24 reference parameters as uncertain. Each draw samples all 24 jointly (clinical costs from one of three source families, clinical times and transition rates from one calendar-year vector each, discovery-stage parameters from beta-PERT distributions with a correlated discovery-to-clinic yield), then applies every scenario's transformation and the current product-type and therapeutic-area overlays to that same draw, so scenario comparisons are paired. Break-even values propagate each draw through the selected commercial archetype with commercial parameters held fixed.
| Element | Use |
|---|---|
| Draws, seed, comparison rate | Defaults 20,000 / 20261002 / 11% reproduce the frozen manuscript run. Fewer draws run faster but change the percentiles slightly. A 20,000-draw run takes about 10–20 seconds; the status line reports completion. |
| Deterministic reference and probabilistic outputs | Headline metrics and a summary table (point value, mean, P05, P25, median, P75, P95) for the selected scenario. Change the scenario in the global controls to switch without re-running. |
| Histograms | Capitalized cost, cycle time and launch probability for the selected scenario, with the median (solid) and 5th/95th percentiles (dashed). |
| Scenario propagation | Dot-and-interval chart and table of medians and 5–95% envelopes for capitalized cost, time, launch probability and break-even V across all scenarios; ticks mark deterministic values. Reproduces Figures 6 and 7b and Tables 3 and 5. |
| Source-conditioned capitalized cost | Box plots of capitalized cost by clinical-cost source family (ASPE/ERG, DiMasi, Wouters-derived) for the selected scenario; reproduces Figure 2a. |
| Audited 24-parameter specification | Reference value, evidence envelope, quality grade, sampling distribution and rationale for each parameter (Table 1 and Supplementary Part S1). |
Results are cached until a parameter that affects them changes (reference values, scenario multipliers, product-type or therapeutic-area tables, commercial values, the product-type or therapeutic-area selection, or the comparison rate). Editing V, the V cases or the crowding overlay does not invalidate the run.
6. Tab 4: Sensitivity & robustness
| Panel | Contents |
|---|---|
| Capitalized cost by rate (Figure 4a) | All scenarios at 7, 9, 11 and 13%, with break-even V at each rate in the table. |
| Cycle-time × rate grid (Figure 4b) | Capitalized cost for the selected scenario when all phase durations are scaled by 0.8–1.2 and the rate is varied 7–15%. The current cell is outlined. |
| Risk shifting (Figure 5a) | Phase I p(TS) raised by the stated amount while Phase II is reduced by each listed offset; the Phase I×II progression and capitalized cost show whether an early-stage gain survives downstream. Both inputs are editable. |
| Therapeutic-area substitution (Figure 5b) | Capitalized cost for every therapeutic area in the editable table, sorted; the selected area is bold. |
| Modality substitution (Figure 5c) | Capitalized cost for every product type; monoclonal antibodies are also shown with preclinical cost at 1×, 1.5× and 2×. |
| Commercial / portfolio discontinuation (Figure 5d) | Additional nontechnical attrition after the technical-regulatory chain, applied as a continuation factor on the final stage; the listed percentages are editable. |
| China execution timing bounds | The low/mid/high phase-timing reductions behind the China scenario (Supplementary Part S2) next to the time multipliers currently in the scenario table. |
7. Loading your own parameters (Excel)
Press Download Excel template (current values) on the main page to get a workbook containing every editable assumption as it currently stands. Edit it and upload it with Upload parameters. Sheets are matched by name (case and spacing are ignored); any sheet that is missing or empty leaves the corresponding assumptions unchanged, and rows with unrecognized names are skipped. Percentages may be entered as 0.47 or 47. The status line reports what was loaded.
| Sheet | Columns | Notes |
|---|---|---|
| Reference | Scenario, Phase, Cost_per_program_M, Cycle_time_y, pTS | Rows for "2026 reference" edit the reference. Rows with any other scenario name create or edit an absolute-value scenario (used for the Paul comparator). |
| Scenarios | Scenario, Phase, Cost_multiplier, Time_multiplier, dpTS_pp | One row per scenario and phase. New scenario names create new forward scenarios; unlisted phases keep multiplier 1 and Δp(TS) 0. |
| Modality | Product_type, Archetype, Preclinical_cost_multiplier, Phase3_cost_multiplier, pTS_PhaseI…pTS_Submission, dpTS_PhaseI_pp…dpTS_Submission_pp | Blank pTS cells keep the scenario's clinical values. Archetype must match a row of the Commercial sheet. |
| TherapyAreas | Therapeutic_area, pTS_PhaseI…pTS_Submission, Cost_mult_PhaseI…Cost_mult_PhaseIII | New names add areas to the selector. |
| Commercial | Archetype, Years_to_peak, Competition_onset_y, Residual_floor_pct, Years_to_floor, Horizon | New archetype names can be referenced from the Modality sheet. |
| Crowding | Parameter, Value | Entrants_N, Rank_n, Expansion_beta, Share_decay_phi, Price_erosion_pi, Diff_window_y, Residual_floor, Years_to_floor (blank = archetype value). |
| Settings | Setting, Value | Discount_rate_pct, Selected_peak_V_B, Selected_scenario, Selected_product_type, Selected_therapeutic_area, V_case_1 … V_case_n. |
Phase names are matched flexibly ("Phase I", "phase 1", "Lead Optimization", "LO" all work). Uploading clears any cached Monte Carlo run. Reset to manuscript defaults discards all edits and uploads.
8. Exports and figures
- ⬇ PNG buttons next to every chart save that chart at twice its on-screen resolution; file names encode the scenario, product type and therapeutic area.
- Download NPV results CSV (Tab 2): NPV, break-even and R&D metrics for every scenario × product type × V case under the current overlay.
- Download selected phase results CSV: the phase-level table for the current selection.
- Download assumptions JSON: every current parameter table and setting, for a complete record of an analysis.
- Download model specification JSON (Tab 3): the frozen manuscript specification, including the sampling distributions.
- Download last Monte Carlo draws CSV: one row per draw with source family, calendar-year vectors, discovery yield and the reference-scenario outputs.
- Download scenario MC summary CSV: medians and 5–95% envelopes per scenario, in the same layout as the manuscript's
scenario_mc_summary.csv.
9. Reproducing the manuscript numbers
| Manuscript item | Where in the tool | Settings |
|---|---|---|
| Table 1 / Figure 1 (reference case: $2.33B OOP, $5.25B capitalized, 14.4 y, 2.31%, 13.2 Phase I starts) | Tab 1, pipeline picture and summary | Defaults |
| Corrected Paul comparator (13.5 y, $2.67B) and Figure 2b Shapley decomposition | Tab 1, comparator box and Shapley panel | Reference selected, 11% |
| Table 2 / Table 3 deterministic scenario results (AI-A $4.47B, AI-B $2.47B, China $4.85B, combined $4.37B) | Tab 1 (select each scenario) or Tab 2 break-even table | Defaults |
| Figure 2a and Table 3 uncertainty column (reference median $4.98B, $2.50–$8.84B) | Tab 3 after Run Monte Carlo | 20,000 draws, seed 20261002, 11%, small molecule, canonical therapeutic area |
| Figure 3 tornados | Tab 1 tornado panel | ±20%, rate 9–13%, V = $5B |
| Figure 4 rate and cycle-time sensitivity | Tab 4 | Defaults |
| Figure 5 robustness panels | Tab 4 | Reference selected |
| Figures 6, 7b, Table 5 (scenario medians and envelopes; break-even medians $1.17/$0.99/$0.56/$1.08/$0.97B) | Tab 3 scenario propagation | As for Figure 2a |
| Figure 7a and Table 4/5 deterministic break-even ($1.23B reference) | Tab 2 | Defaults |
| Figure 8 crowding sensitivity | Tab 2 equal-share panel | Reference selected |
The Monte Carlo engine (Mulberry32 generator, draw order and distributions) is the same as in the Python model of the reproducibility package, so medians and percentiles agree to the third decimal. Running with a different product type or therapeutic area applies that overlay to every draw; the manuscript figures use small molecule and the canonical therapeutic area.
10. Interpretation notes and common questions
- Why does editing a forward scenario's phase values not work? Forward scenarios are transformations of the reference. Edit the reference values (select "2026 reference") or the multipliers in the scenario-transformation table.
- Why are Small molecule and Biologic R&D numbers identical? The manuscript withdrew the universal biologic R&D modifier; biologics differ through the commercial archetype (longer uptake, later competition, higher floor). Add a Phase III cost multiplier or Δp(TS) in the modality table if you want an R&D overlay.
- Why is the discovery contribution in the Shapley decomposition near zero? Discovery-stage unit costs in the reference are the inflation-adjusted Paul values, so there is no change to attribute; the manuscript labels these inputs grade D and bounds them in the Monte Carlo.
- Which rate should I use? 11% is retained for comparability with the 2010 benchmark and is an analysis setting rather than an audited parameter; Tab 4 shows every scenario at 7–13% and the grid to 15%.
- What does Φ mean? The share of the stand-alone peak gross profit a product realizes when N products share the class. With equal shares Φ = Nβ−1; β is illustrative, not calibrated, because public class histories do not identify it.
- Why do the paradox table's steps 2 and 3 agree? The crowding overlay is at N = 1. Raise N in the crowding panel.
- Why did the Monte Carlo results disappear? You changed a parameter that affects the draws (reference values, scenario multipliers, overlay tables, commercial values, product type, therapeutic area or the comparison rate). Press Run Monte Carlo again.
- Are the therapeutic-area and modality results full cost models? No. They substitute clinical transition rates only (plus an optional preclinical cost multiplier for antibodies) and are robustness checks, as stated in the manuscript.
- Is anything stored? Nothing leaves the browser. Edits are lost when the page is reloaded; use the Excel template or the assumptions JSON to keep a parameter set.