Ormeval provides decision-support intelligence derived from public sources. It is not investment, medical, legal or regulatory advice, and states nothing about the value of any security. Forward figures are deterministic model estimates, not guidance.
Ormeval · Pharma intelligence · 28 SEP 2026
Track record
Every dated call we publish and every forecast we make, scored in public against what companies actually reported - misses included.
5,395forecasts frozen, first out-of-sample score with FY2027 reporting
60%of 736 scored asset-years land within 25% of the reported actual
Forecast models as of 2026-04-29.
Our record is published so you can check it, ask any forecaster you rely on for the same two tables. Two records, kept separately: from now on the ledger fixes every new call the moment we publish it, so it cannot be quietly revised (the record opened 2026-07-22; earlier calls are backfilled and flagged as such below); the calibration scores our modelled revenue against what companies actually reported, misses included. Measured accountability is the point, in a field where the consensus most teams rely on is, by the published evidence, frequently and substantially wrong.
For every asset-year where a company reported a full-year actual and we model at least 80% of its approved indications, we compare the sum of our indication models against that reported figure. Below that coverage bar the comparison is not fair to either side, so those 31 points are excluded and counted here rather than dropped silently.
Like-for-like. Excluding the 22 comparisons distorted by something other than the model (15 where the reported figure is only a share of the drug; 8 where a model this asset does not own is in the sum; 2 where two indication models overlap; 3 of these sit in more than one category, so the distinct count is 22), 714 asset-years score a median absolute error of 17.5% against 18.1% on the full set, and run under in 542 of 714. The headline stays the full set; this is shown so both numbers are visible and the difference between them is explainable rather than asserted.
A single median is the wrong summary. A few million either way on a small product reads as a triple-digit percentage; the same miss on a $5bn brand is a rounding error. These are the same points as above, separated rather than blended.
Ranked by the dollar size of the gap, not by percentage, because percentage ranking fills the table with tiny products where a few million reads as a triple-digit error. The Error column is the honest scale: on a multi-billion-dollar brand a large dollar gap is usually a modest percentage. Every actual is a company-reported figure; every modelled figure is a model estimate. Follow the asset link for the reported series and its sources.
| Asset | Year | Reported actual | Our model | Gap (USD) | Error | Models summed |
|---|---|---|---|---|---|---|
| Humira receipts | 2022 | $21.2B | $11.5Bmodel estimate | $9.74B | -46% | 9 |
| Skyrizi receipts | 2025 | $17.6B | $8.42Bmodel estimate | $9.14B | -52% | 4 |
| Mounjaro receipts | 2025 | $23.0B | $14.4Bmodel estimate | $8.57B | -37% | 2 |
| Ozempic receipts | 2022 | DKK 59.8B | DKK 103.1Bmodel estimate | $6.29B | +73% | 1 |
| Humira receipts | 2023 | $14.4B | $8.59Bmodel estimate | $5.81B | -40% | 9 |
| Jardiance receipts | 2025 | $3.43B | $9.14Bmodel estimate | $5.71B | +166% | 4 |
| Jardiance receipts | 2023 | $2.75B | $8.22Bmodel estimate | $5.48B | +200% | 4 |
| Jardiance receipts | 2024 | $3.34B | $8.80Bmodel estimate | $5.46B | +163% | 4 |
| Rinvoq receipts | 2025 | $8.30B | $2.88Bmodel estimate | $5.42B | -65% | 7 |
| Jardiance receipts | 2022 | $2.07B | $6.89Bmodel estimate | $4.83B | +234% | 4 |
The engine models each indication separately and sums them. On 2 assets some of those models overlap, a broad programme counted on top of the narrower ones inside it, so the sum exceeds what the company reported. Jardiance is currently the largest: 4 models summed against one reported figure overstate it by $5.71B (166%), though where an asset also appears in the lists below the distortions compound rather than choose. These are flagged, not auto-excluded: dropping whatever overshoots would tune the model to the answer and make this whole page circular.
Some drugs are given together, and only one partner books the regimen's revenue. On 8 assets the sum above includes at least one model belonging to the OTHER drug, so part of the error measures a mismatch of denominators rather than the forecast. Imjudo shows the mechanism at its cleanest: 2 of its 2 summed models belong to another company's drug, and its latest scored year nets -9%. These stay in the scored set for the same reason as the overlaps above. Removing a comparison because we can explain why it looks bad would improve this page's number without improving the engine.
On a co-promoted or alliance product the partners split the revenue, and the engine models the whole drug while the company reports its share. Pfizer books alliance revenue for Xtandi while Astellas books the brand, and Lilly reports its share of Jardiance while Boehringer books the rest of the franchise; the same arithmetic runs through drugs beyond this list - Pfizer books about half of Eliquis, for example, because BMS books the rest. The miss is about who signs the invoice, not about the forecast. These stay scored, like the two lists above. Note that a partial booking does not doom a comparison - Enhertu is on this list and currently lands within 2% - which is why membership records the accounting structure rather than this year's error.
This is a backtest, and it is checkable now. We run today's models back over fiscal years companies have already reported and score the result against what a reader could have got with no engine at all: the last reported actual carried forward flat, and the last reported actual grown at its own trailing rate. The fourth figure re-levels our curve - same modelled shape, but started from the company's latest reported actual instead of the curve's calibrated level - so the cost of the plateau drift diagnosed in our bias write-up is priced rather than described.
Scored on 274 asset-years of the 707 the calibration above covers - every one where the predicted year sits exactly one year past the last reported actual and a trailing growth rate is computable. Forecast models as of 2026-04-29. How the engine works.
Read it straight: on next-year asset totals, the naive baselines currently win. The cause is diagnosed, not mysterious - our curves are calibrated to a reference year and then plateau, while real franchises keep compounding (the same conservatism as the 75% understatement rate above, which our forecast-bias analysis traces to the same cause). Two things belong beside that. First, this comparison favours our engine - today's curves were curated with these actuals visible, the baselines only ever see earlier years - and the engine still does not win, which is why this section exists. Second, a trailing growth rate answers no question this product is for: it has no indication mix, no geography, no patent cliff, and it extrapolates straight through a biosimilar event. We publish it as the bound on what our accuracy may claim, not as a substitute for the model.
The fourth column is a proposal, and a tie. Carrying the engine's own year-over-year shape from the company's latest reported actual - instead of from the curve's calibrated level - lands level with the trailing-growth baseline on the median and splits the head-to-head almost exactly in half. It is not the displayed forecast, for a reason we have recorded as part of the engine's declared domain: applied automatically it would overwrite years where the engine correctly called a discontinuity trailing data missed, and it only acts one step past the last reported actual - the product's headline is a multi-year peak, where this correction decays to the raw engine.
The comparison above is one blended number, and it hides the real structure: the engine and a trailing extrapolation are good at different things. On discontinuity and youth - launch ramps, fast growers, franchises in collapse - extrapolating the trailing rate detonates, and the engine's structural knowledge (indication mix, patent cliffs, competitive entry) is the better estimator. On mature, steady franchises, a two-line trailing extrapolation currently does the one-year-ahead job better than our engine does, and this table says so rather than blending it away. We publish where our own model loses - ask any other provider for the same table. That is what being checkable means.
| Regime | n | Engine | Trailing growth | Engine closer |
|---|---|---|---|---|
| Launch ramp (age 1-2)directional - n too small to lead with | 10 | 16.7% | 154.6% | 100% |
| Growth (age 3-5) | 58 | 18.1% | 33% | 60% |
| Mature (age 6-10) | 116 | 17.2% | 5.8% | 17% |
| Late lifecycle (11+) | 84 | 25.4% | 9.6% | 27% |
| Fast growers (trailing >40%) | 76 | 23.9% | 35.5% | 59% |
| Steady (trailing 0-15%) | 74 | 17.2% | 4.3% | 16% |
| Declining (trailing <0%) | 63 | 20.6% | 9.7% | 27% |
Median absolute error per regime, same scored points as above; regimes are observable before the predicted year (asset age, trailing growth of the actuals). “Engine closer” is the head-to-head share against the trailing-growth baseline.
Everything above scores today's models against the past - a real discipline, but the models could see those years. The forward record is different: 5,395 forecasts across 413 assets are frozen in an append-only ledger (earliest snapshot 2026-08-09), and each is scored only when a fiscal year reported after the snapshot lands. Nothing is rescored, revised or backfilled.
No frozen forecast has matured yet. The earliest frozen year is 2027, so the first out-of-sample accuracy figure - one-year-out error - can exist only once FY2027 actuals are reported. A five-year figure exists in 2032. Until then this section deliberately shows nothing: quoting the retrodiction numbers above as if they were forward accuracy is exactly the substitution this page exists to refuse.
Graded for 2026-08, generated 2026-09-18: every dated snapshot in the append-only forecast ledger, summed at the asset level and compared against the reported actual that arrived afterward. Misses first, structurally - the largest-gap table is written before the closest-calls table in the underlying file, so neither can be padded to bury the other. The same table as plain text is at /track-record/scoreboard.md.
Nothing has matured yet. 2644 frozen ledger row(s) were considered for 2026-08; the newest reported fiscal year available to grade against is 2025. An empty scoreboard is the honest state until the first forecast year closes and reports, not a suppressed result.
What counts as a forecast. A row in the forecast ledger - the append-only record of the forward curve as it stood on a dated snapshot - with snapshotId on or before the last day of this month. Today's model is never consulted; that would be a backtest (the calibration module), not a track record.
What counts as an actual. A cited FY figure in the asset's `reported` record (the cited reference dataset), for a year at or before 2025. A fiscal year is reported in the year after it closes, so nothing later can have an actual yet.
Unit of comparison. One (asset, snapshot, year). Predicted is the SUM of that snapshot's program curves for that asset; the ledger is per indication program and `reported` is per asset, so a program row on its own has no comparable actual.
Known bias. The indication build-up covers eight markets and only the indications that carry a model; a reported FY figure is worldwide and covers the whole label. Both gaps push the build-up DOWN, so some part of any under-call here is definitional rather than error. The scope of the build-up is stated in the forecast reconciliation notes. Nothing is rescaled to close the gap.
Determinism. No model call, no network, no randomness. The month and the generation time are fixed inputs, so a month always regrades to the same result.
Sources: the forecast ledger, the cited reference dataset (reported actuals)
Every catalyst we publish is appended to a file that is never edited or deleted. A call that has not yet happened is pending and carries no outcome, we do not fill the answer in afterwards.
377 of these were backfilled. The ledger began on 2026-07-22, so calls published before that carry the date the ledger was created, not the date we first published them. They are flagged individually below. The ledger is only as old as it says it is.
Keytruda · Merck & Co. · 1L PD-L1-positive metastatic NSCLC (China)
Late-breaking analysis presented at the IASLC 2026 World Conference on Lung Cancer (WCLC), 13 Sep 2026; society/sponsor release, peer-reviewed publication pending.
Source →Flucelvax / Fluad (CSL Seqirus influenza vaccines) · CSL · seasonal influenza immunisation, egg-based standard-dose vaccine, US
Checked 9 Aug 2026: the openFDA NDC endpoint returns zero records for brand_name AFLURIA under any labeler, and DailyMed returns no AFLURIA SPL. The only Seqirus-labelled records remaining are Flucelvax, Fluad, Audenz (H5N1, SPL last published 30 Sep 2024) and Rapivab. We found no CSL discontinuation notice, so this is what the US listing registries show rather than a company statement, and should be confirmed with CSL before it is used commercially.
Source →Xgeva · Amgen · Xgeva-equivalent 120 mg denosumab biosimilars - approved but not marketed
Approved-but-never-marketed. Samsung Bioepis's partner Cordavis launched only the Prolia-equivalent Ospomyv (NDC start 10 October 2025), leaving the oncology presentation on the shelf. Counting approvals rather than launches overstates the erosion pressure on Xgeva by two products.
Source →Faslodex · AstraZeneca · HR-positive advanced or metastatic breast cancer, US generic market
Approved-versus-marketed check: of 15 fulvestrant ANDAs on Drugs@FDA, Apotex (ANDA 211730) and two Eugia applications (ANDA 208811, ANDA 215169) are Discontinued; in the NDC directory the Athenex listing ends 31 Dec 2026 and the Meitheal listing ends 31 Jan 2027.
Source →Enflonsia · Merck & Co. · Children under two years at increased risk for severe RSV entering their second RSV season
Source →Kerendia · Bayer · Chronic kidney disease without diabetes, adults with albuminuria
1,584 participants; registry shows the trial COMPLETED with primary completion 31 Dec 2025. A prespecified exploratory subgroup in glomerular diseases was published separately in JAMA on 21 Jul 2026. Roughly doubles the addressable CKD population if it reaches a label.
Source →Orforglipron / LY3502970 · Eli Lilly · Type 2 diabetes
Delivered on the end-of-Q2-2026 submission target Lilly set when ACHIEVE-4 read out in April.
Source →Mounjaro · Eli Lilly · Type 2 diabetes
Source →oveporexton · Takeda · Narcolepsy type 1 (narcolepsy with cataplexy), adults
Approved ahead of the Q3 2026 PDUFA goal date, but for adults only, narrower than the 16-and-over Chinese label.
Source →quemliclustat · Gilead · Pancreatic cancer
Source →mrna-1010 · Moderna · Seasonal influenza prevention, adults 50 years of age and older
The 65-and-older portion of the indication was granted under ACCELERATED approval (21 CFR 601.41) on immune-response surrogates, not full approval; the 50-64 portion is a standard approval. BLA had been received 5 December 2025, so approval landed on the PDUFA date.
Source →Mounjaro · Eli Lilly · Type 2 diabetes inadequately controlled on metformin with or without an SGLT2 inhibitor
Dual primary endpoint split: non-inferiority met on weight (15.2% vs 15.8%) but not on HbA1c (-1.9 vs -2.2 percentage points at 68 weeks).
Source →Recorded and still ahead of us as of 2026-10-15, the date this page was built. No outcome is shown because none exists yet. A further 1819 recorded calls are past their date but have no cited outcome on file. That is a gap in our coverage, not a prediction still running, and it is counted above rather than hidden in this list.
Yiwei · Akeso · 2L+ EGFRm NSCLC (ex-China, US)
Source →Myqorzo · Cytokinetics · oHCM monotherapy (vs metoprolol)
Source →Zanzalintinib · Exelixis · 2L+ metastatic CRCbackfilled
Source →Velexbru · Ono · Relapsed/refractory primary CNS lymphoma (US)
Source →Keytruda royalty (2017 PD-1 settlement) · Ono · Keytruda royalty income
Source →Cliramitug · AstraZeneca (Alexion) · ATTR-CMbackfilled
Source →Skytrofa · Ascendis Pharma · Growth disorders in GH-sufficient children (ISS, SHOX deficiency, Turner syndrome, SGA)
Source →Jakafi · Incyte · Myelofibrosisbackfilled
Source →Gamunex-C / Xembify / Yimmugo (IG franchise) · Grifols · Secondary immunodeficiency (CLL / multiple myeloma / NHL)
Source →Cabometyx · Exelixis · RCCbackfilled
Source →Keytruda · Merck & Co. · Unresectable Stage III NSCLC (post-chemoradiation)backfilled
Source →Keytruda · Merck & Co. · 1L HR+/HER2- metastatic breast cancerbackfilled
Source →The forward curve was re-levelled on 2026-09-18. It now starts from the latest full year the company reported, carries that franchise’s own trailing reported growth into the next year, and only then follows the modelled shape. Before that date it started from a curated forward anchor, which is why the engine understated on most scored years. A re-levelled year is no longer the cited anchor and is labelled a model estimate, not a baseline. Figures below this paragraph are backtests of the indication build-up and are unaffected by the change.
The actuals are not ours. Every reported figure on this page is a number the company published in its own results, linked at the end of each row. We do not adjust them.
The model figures are ours, and labelled. Anything marked model estimate is computed by our deterministic engine from a per-country patient funnel. It is an estimate from public data, not guidance and not a consensus figure.
The comparison is only run where it is fair. An asset whose approved indications we model only partially would look wrong for a reason that has nothing to do with forecast quality, so those 31 asset-years are excluded, and counted, so you can see the denominator.
A pending call stays pending. The ledger is append-only. When an outcome becomes public we append the resolution; we never rewrite the original line, and we never mark something resolved that has not been publicly reported.
More on sourcing and the engine in the methodology.
Why a measured error beats an unmeasured one. Pharmaceutical revenue forecasting is genuinely hard: a Nature Reviews Drug Discovery analysis of analyst consensus concluded that most consensus forecasts are wrong, often substantially. The standard a decision-maker should demand is not a forecast that claims to be right, but one that is scored against what actually happened and shows its full distribution, misses and all, the way this page does. The numbers above are not a confession; they are the accountability the alternative never offers.
Source: Cha, Rifai & Sarraf, “Pharmaceutical forecasting: throwing darts?”, Nature Reviews Drug Discovery, 2013.
The number on this page is the one behind every asset, indication and scenario in the catalog: scored against reported actuals, cited to the public record, and shown with its misses. Trust the forecast because you can check the record.