Methodology version history

Modelling Deal Terms, rNPV Model

M10 documents term-model mapping v5: "Model these terms" now values a term set in which your organisation pays, on the licensee basis of rNPV model 4.1.0, and refuses any paid term the model cannot value. Every set with no paid term keeps its scenario inputs, and every issued export still verifies.

• Modelling Deal Terms: an in-license or cash purchase is valued on the revenue inside the license grant your organisation receives, with the development cost share the terms state, net of the upfronts, milestones, royalty and cash purchase price it pays. With a profit share, paid terms are subtracted within the profit split.
• Modelling Deal Terms: every term your organisation pays is valued or refuses the set, never listed as not modelled: an untimed paid milestone, a paid royalty with a step-down, cap or floor or on part of the licensed revenue, stock or debt in a purchase price, and paid kinds with no method each refuse with their own reason. A paid approval milestone lands at the review stage, or is refused when there is none.
• Modelling Deal Terms: new assumptions name the subtraction of paid terms, paid lump sums at the valuation date with no tax saving, the paid royalty's tax saving and whole-forecast term, and the buyer bearing development cost after a cash purchase. The four tax disclosures are shown wherever a value with paid terms is.
• rNPV Model: the note that "Model these terms" does not send paid terms is replaced; from mapping v5 it does.

• Modelling Deal Terms as of M10 →
• rNPV Model as of M10 →

rNPV Model

M9 documents rNPV model 4.1.0, which can value a deal from the side that pays: it subtracts upfronts, milestones and a royalty your organisation pays, the royalty after its tax saving. Every existing input keeps its value, and every issued export still verifies.

• rNPV Model: a deal can hold terms your organisation pays, on a new licensee basis (the whole revenue, your stated share of development cost, net of the terms you pay) or under a profit split. Paid upfronts count in full at the valuation date and paid milestones follow the timing of received ones, neither with a tax saving. A paid royalty is counted after tax, its saving capped at operating profit.
• rNPV Model: a paid milestone whose stage gate names no stage, or that has neither a gate nor a sales threshold, is refused rather than timed at the last stage. Received milestones are timed as before.
• rNPV Model: four new limitations. Received terms stay before tax, so the licensor's and licensee's values for one deal do not add up to the full-ownership value; paid lump sums get no tax saving; the royalty saving is capped at operating profit; a paid royalty runs for the whole forecast. The asset modeler and Modelling Deal Terms do not send paid terms yet.

• rNPV Model as of M9 →

Modelling Deal Terms, Holding Valuation

M8 documents term-model mapping v4, which refuses to model a term set in which your organisation pays an upfront, milestone or royalty, because valuing only what it receives would overstate its side. Holding Valuation now covers secondary share purchases and position reviews. No rNPV model value changes, and every issued export still verifies.

• Modelling Deal Terms documents term-model mapping v4: a set in which your organisation pays an upfront, milestone or royalty on the modelled asset is refused with a fixed message, where mapping v3 listed that payment as not modelled and valued the rest. Profit share and cost share are still modelled from either side, and every set mapping v4 still models gets the same scenario inputs as under v3. Scenarios modelled under earlier mappings keep their values.
• Holding Valuation: an equity holding bought from an existing shareholder is valued, ledgered and looked through like one bought in a priced round. Its implied post-money is the company's value at the price paid, and because the look-through estimate treats every share alike it can overstate common shares, which rank behind preferred shares.
• Holding Valuation: a position review shows what your organisation already holds in a company or asset, each at its own method's value, and proposes or changes no holding. There is no market-price method, so a listed company's quoted share price is not used.

• Modelling Deal Terms as of M8 →
• Holding Valuation as of M8 →

rNPV Model, IRA Terminal-Value Framework, Modelling Deal Terms, Holding Valuation

M7 documents model 4.0.0, which applies the IRA price cap to US revenue only, as the IRA section already stated. It also states more of what the model and the look-through estimate leave out, and records two workspace valuation rules. Scenarios computed on earlier models keep their values, and every issued export still verifies.

• Model 4.0.0: each geography of a scenario is matched to a territory, and the IRA cap applies to the United States, to worldwide revenue, and to rest of world only when no other geography covers the US. Earlier models capped every geography. For an asset with revenue outside the US, the rNPV rises and the IRA revenue at risk falls; in one example with 50% US, 30% EU5 and 20% rest-of-world revenue, rNPV moved from $281.6M to $308.0M. A scenario with no geographic split, or one whose geography names match no territory, keeps the cap as before.
• rNPV Model now states that development and launch costs are not tax-deductible in the model, that revenue is discounted at year end and costs at stage midpoints, that each geography's exclusivity runs from its own launch, which basis bears development cost, that a launch cost is left out of the yearly deal rows, and what the Monte Carlo simulation samples: stage PoS and costs, not sales, timing or margins.
• Modelling Deal Terms documents term-model mapping v3: an upfront or milestone not allocated to one indication plan is not modelled when the asset has several included plans, because the asset's value adds each plan's scenario.
• Holding Valuation: the look-through estimate now also says it ignores license payments the company owes and territory-limited rights, and that a licensor holding equity finds its own royalty inside the estimate. A license or royalty holding is not valued from a scenario of the whole asset, and valuation bases that cannot be identified are never added together.

• rNPV Model as of M7 →
• IRA Terminal-Value Framework as of M7 →
• Modelling Deal Terms as of M7 →
• Holding Valuation as of M7 →

Holding Valuation (new)

M6 adds a Holding Valuation section describing how the workspace values each position an organisation holds, and adds the look-through estimate for equity in a company. No model value changes, and every issued export still verifies.

• New section at /methodology/holding-valuation documenting the five methods (scenario, share of asset value, look-through estimate, cost and not valued), which holding kinds allow each, their inputs, the reasons a holding is not valued and when a value is partial.
• Totals add values only within one basis and one currency, never convert currencies, and count a holding that cannot be valued instead of adding zero.
• The look-through estimate values equity in a company as the ownership share of the full-ownership values of the assets the company owns or sponsors. It is labelled as an estimate, not an equity valuation, ignores cash, debt, preferences and dilution, and is left out of totals unless included.
• No existing section changed, and no rNPV model value, dataset or default changed.

Probability of Success Calibration, rNPV Model

M5 restates the probability-of-success section as reference data and changes no model value. The figures come from published sources, prefill the asset modeler and can all be changed; the model computes with the values the user keeps. Exports stamped M4 or earlier keep their values and still verify.

• PoS Calibration now opens by saying what the figures are: reference baselines from named published sources, not a prediction for any asset, updated when the sources change.
• A new Section 5 explains how the figures are used: they prefill the asset modeler, every value can be changed, saved scenarios record which values are benchmark defaults and which the user set, and the multipliers are optional.
• Removed the internal cohort analyses, the multiplier governance discussion, the outside-review note and the research roadmap. Table 2 labels each multiplier by its basis (published source or PhaseFolio assumption) instead of an evidence tier. Biomarker quality keeps its published Schwaederle 2016 values and applies only after review.
• Phase 1 response-rate bands and the other drug-specific flags are described together as flags shown beside the valuation; they do not change PoS or the rNPV.
• rNPV Model: the probability paragraph says each stage's PoS is prefilled from the reference matrix and can be changed in the asset modeler.
• The research pages now describe the May 2026 signal evaluations as a dated record of two internal checks, not a validation of the model. The verdict records and full report are unchanged.

• Probability of Success Calibration as of M5 →
• rNPV Model as of M5 →

Probability of Success Calibration, Backtest Methodology, rNPV Engine, Modelling deal terms

M4 documents rNPV engine 3.0.0 and PoS dataset pos-benchmark@2026-09-27, which change engine values for the same inputs. Every Phase I to NDA/BLA probability is now derived from named rows of BIO/QLS/Informa 2021, the CAR-T modifier is removed, orphan gene therapy gets one calibrated modifier, and deal royalties and sales milestones follow the geography-adjusted revenue timing. Exports stamped M2 or earlier keep their values and still verify.

• PoS baselines: each Phase I to NDA/BLA rate is the BIO 2021 therapeutic-area row, shifted in log odds by the BIO 2021 modality row and biomarker row. Rows with very few transitions are smoothed so a 100% on a handful of programs does not become certainty. Preclinical rates are unchanged and labelled PhaseFolio assumptions; BIO 2021 has no preclinical stage.
• All 264 indication × modality × biomarker cells are now filled. Before M4, 125 cells had no values of their own and used the area's 'other' modality row.
• Hematologic oncology now uses the BIO 2021 hematologic cancers group (7.5% from Phase I) instead of a row that mixed in other cancers. Small-molecule, no-biomarker likelihood of approval from Phase I moves from 17.2% to 6.9% for hematologic oncology, 5.1% to 4.1% for solid tumours, 14.9% to 15.8% for rare disease and 7.4% to 14.3% for metabolic disease; the PoS calibration page lists every row.
• Biomarker strategy is built into each matrix cell from BIO 2021's biomarker rows instead of being listed as a multiplier. A companion diagnostic roughly doubles the likelihood of approval in solid tumours, in line with BIO 2021; enrichment takes half the companion-diagnostic shift, a labelled PhaseFolio assumption.
• There is no CAR-T modifier. Phares, Phillip and Trusheim (2025) report hematologic CAR-T and TCR therapies at 7.6% from Phase I against 7.5% for comparable drugs, and the derived hematologic cell therapy cell is already 7.25%. Before M4 a 3.0× CAR-T factor applied to every cell therapy.
• Gene therapy with orphan designation gets one modifier, a per-stage odds ratio of 1.13 on Phase I and II, calibrated to the 18.5% that Phares, Phillip and Trusheim (2025) report for rare-disease gene therapies. It replaces orphan designation for gene therapy instead of stacking with it.
• First-in-class 0.85× is kept and relabelled as a PhaseFolio assumption; BIO 2021 gives no figure for it.
• Deal terms: royalties, royalty tiers and the profit-split pro-forma now read the same geography-adjusted revenue series as the headline, and the pro-forma adds up to the deal value. A milestone can carry a net-sales threshold; it pays in the first forecast year whose sales reach the threshold, weighted by the probability of approval and discounted like revenue, and pays nothing if the forecast never reaches it.
• Backtests retired: the back-test library at /research/backtests, its rheumatoid arthritis, NSCLC and antimicrobial cohort pages and the in-app backtest link are removed, and Backtest Methodology is now a retirement notice. The cohorts were run on engines before 3.0.0 and are no longer maintained. No outcome backtest is currently published; the last published backtest methodology stays at ?version=M2. New PDF and Excel exports no longer cite the backtests in their PoS provenance caption.
• Modelling deal terms: term-model mapping v2 no longer names the assumption that all deal revenue is timed from the lead launch, because engine 3.0.0 times it from each kept region's own launch. Scenarios modelled under v1 keep their engine version and values.
• The PoS dataset is stamped pos-benchmark@2026-09-27. Exports stamped pos-benchmark@2026-04-08 or earlier keep that stamp.

• Probability of Success Calibration as of M4 →
• Backtest Methodology as of M4 →
• rNPV Engine as of M4 →
• Modelling deal terms as of M4 →

Modelling Deal Terms (new)

Added a new methodology section covering how licensing deal terms map onto a scenario's deal structure. No existing methodology section's body changed and the rNPV engine is unaffected.

• New "Modelling Deal Terms" section describes "Model these terms": mapping a proposed or agreed licensing term set onto a full-ownership scenario, which pins a new partner-case scenario to that term-set version. The source scenario is unchanged.
• Names what is modelled (upfront payments, milestones, one royalty, one profit share, one cost share) and how each maps onto the deal structure, what is not modelled with a reason, and what is refused outright with its message, including that a non-USD term is refused until a conversion rate is set.
• Discloses the territory restriction: a territory-limited term is modelled only against the scenario's geographies inside that territory, scaled to their share of peak sales, and states this is an approximation, not an exact re-derivation of a per-territory deal.
• Discloses milestone timing (a modelled milestone always lands at the end of a development stage), the cost-bearing rule (never assumed; set only by an explicit term), and which fields a modelled scenario keeps locked versus editable.
• Methodology M3 documents rNPV engine 2.9.0 and changes no engine value. No existing signed export is invalidated. This is a new section, so no earlier page needed a snapshot.

Probability of Success Calibration, Backtest Methodology, COGS Benchmark, SG&A Benchmark, Tax Rate Benchmark, rNPV Engine, JHTV Portfolio Scoring and JHTV Deep-Dive Production (retired)

Methodology releases are now numbered: M1 is the release first stamped methodology@2026-09-27, and new exports stamp M2. M2 is a correction with no engine value changed. It publishes the PoS baseline table the engine computes with, removes an unsupported Citeline attribution, corrects author and source records, and corrects how the rNPV page describes deal royalties when geographies are set. It also retires the out-of-date JHTV public-data case study.

• The PoS calibration page, its machine-readable copy and the signed-export appendices now show the baseline table the engine computes with, beside the nearest BIO/QLS/Informa 2021 group for each area. The rows are described as a PhaseFolio composite informed by BIO 2021, not a transcription of it.
• For five therapeutic areas, exports stamped before M2 were computed on the engine table while the page showed different values. The engine table was always the one used; the displayed table was wrong. Pre-M2 exports are unaffected and still verify.
• Earlier versions attributed the modality rows to Citeline 2024 pipeline data. PhaseFolio has never licensed or used Citeline data. The rows for modalities other than small molecule are PhaseFolio assumptions set relative to each area's small-molecule row, and are now disclosed that way.
• New exports stamp the PoS dataset as pos-benchmark@2026-04-08, the date its values last changed, instead of 2021. Exports stamped pos-benchmark@2021 keep that stamp.
• Author and source records corrected: the BIO 2021 author list; the NEWDIGS analysis is linked directly; the $95,780 per-dose CAR-T figure is attributed to Harrison et al. (2019), and the autologous CAR-T cost analysis to Spink and Steinsapir (2018); the monoclonal antibody pricing, antibody supply and ASCO CAR-T cost references now name their authors; the tax statute references are credited to the U.S. Congress; the AbbVie references are labelled as the Form 10-K they link; an archived copy replaces a Batavia Biosciences link whose certificate fails.
• The rNPV Engine page said deal royalties always read the same revenue series as the headline. That holds only with no geographies set. With geographies, the headline follows each geography's share and launch delay, while royalties use the worldwide net-sales schedule built from peak revenue without launch delays, so the two can differ. The page now says so. The engine is unchanged; applying geography timing to royalties is a planned engine change that will ship as its own engine version.
• Methodology M2 documents rNPV engine 2.9.0, which adds an optional territory key to geography inputs without changing any output. M2 itself changes no engine value.
• The JHTV public-data case study is retired. It was out of date, and its scores and rNPV figures should not be relied on. The case-study page, asset deep dives, case-study PDF and /api/v1/case-studies endpoints are removed, and old links land on the retired JHTV Portfolio Scoring page. The last published JHTV Portfolio Scoring and JHTV Deep-Dive Production pages stay available as frozen snapshots. The rNPV engine never used the JHTV rubric.

• Probability of Success Calibration as of M2 →
• Backtest Methodology as of M2 →
• COGS Benchmark as of M2 →
• SG&A Benchmark as of M2 →
• Tax Rate Benchmark as of M2 →
• rNPV Engine as of M2 →
• JHTV Portfolio Scoring as of M2 →
• JHTV Deep-Dive Production as of M2 →

SEC EDGAR Deal-Term Extraction, rNPV Engine

Retired the SEC EDGAR deal-term source. The rNPV wizard no longer suggests deal terms from SEC filings; an optional deal structure is an analyst assumption. Same-input rNPV outputs are unchanged, and scenarios that accepted a suggestion earlier keep their source attribution.

• The SEC EDGAR Deal-Term Extraction section is now a retirement notice that lists every earlier version as a frozen snapshot.
• The rNPV Engine input-provenance list no longer names SEC comparator deal terms as a source.
• New signed exports no longer stamp the sec-deal-terms dataset. Exports that already carry it keep that stamp and still verify.
• Engine 2.8.1 removes unused SEC auto-populate code. Deterministic and Monte Carlo outputs are unchanged.
• The Backtest Methodology page no longer points readers to the retired Workspace Intelligence dashboard; per-drug results are on each cohort's page under /research/backtests. This is a pointer change only: the backtest section keeps methodology@2026-06-07, and no cohort, AUC or engine output changed.

• SEC EDGAR Deal-Term Extraction as of methodology@2026-09-27 →
• rNPV Engine as of methodology@2026-09-27 →