GRAIL's blood test for cancer missed its trial's main goal in February. The stock lost half its value in a day. In September an FDA panel voted for the test, and the stock now stands well above where it was before the miss. The miss says less about the test than it seemed to. The Medicare law that would pay for the test says a good deal about the price.
GRAIL's Galleri test is a good sieve with large holes. The NHS-Galleri trial enrolled 142,250 people in England for three years. When the test said cancer, it was right about half the time, and when the signal proved to be cancer its first or second guess at the organ was right about nine times in ten. It missed the trial's main goal, fewer cancers found at stage III or IV, because it sees late cancers far better than early ones. Stage IV diagnoses fell 14%, stage III rose 25%, and the two together rose 3%. Our model of the trial, like an earlier one by Ruth Etzioni's group, says more years of yearly screening would not have reached the 20% fall in stages III and IV that the trial was designed to find.
The price is a bet on a slow clock. At $143.26 on October 2, GRAIL was valued at $6.7 billion, $5.8 billion net of its cash, and on generous assumptions that needs about $1.9 billion of Galleri sales in 2032, some 2.7 million tests a year, which is growth of 50% a year for six years. Cologuard grew faster, from the same starting point. It also had Medicare for everyone from 50 to 85 and a national guideline behind it. The new Medicare law would pay $508.87 a test (or $591.92, if Medicare uses Cologuard Plus's rate) from 2029 at the earliest, only for people who were under 65 on January 1, 2029, one birth year at a time. It never reaches anyone who was 65 already. Even if one in ten of the people it reaches in 2032 were tested, Medicare would bring in about a third of the revenue the price needs.
The rest has to come from employers, insurers and patients, paying for a test no guideline recommends. Or the price is a wager on a much larger outcome, at long odds. What would change our view: test volumes on the path the price requires (about 351,000 in 2027), a large insurer covering the test without a guideline, or the guideline panel opening a review. The tests are set down below, to be scored on May 31, 2027.
This note began with a search. On October 4 I asked for a sweep of the past year of Nature and its sister journals. I wanted a finding an investor could test, one that more mathematics could carry further than the paper did. One paper stood out: "Performance of a multi-cancer early detection test in the randomized controlled NHS-Galleri trial", by Richard D. Neal, Peter Sasieni, Charles Swanton and ten colleagues, published in Nature Medicine on September 22, 2026. It reported what GRAIL's Galleri test found in three yearly rounds of screening across England, round by round, cancer by cancer and age by age. The same day the New England Journal of Medicine published the trial's main result, by Sasieni and colleagues, and it was a disappointment: the test had not reduced the number of cancers found at stages III and IV.
What made it a note for us was the stock. The market had already heard the bad news in February and cut GRAIL in half. By the time the papers appeared the shares were rising. Two weeks later they stood far above their price before the miss. Investors had decided the miss mattered less than it seemed. We wanted to know why, and whether the arithmetic agrees.
We have not found three things in this note anywhere else: the Medicare law's age rule set against the test's yield by age, year by year; a reverse valuation of the price set against Cologuard's record; and the number of tests, and of dollars, it takes to avert one death on the trial's own counts. Others got to the rest first, and they are named where their work comes up.
On February 19, 2026, GRAIL told investors that the trial had not met its primary endpoint. The next day its shares fell from $101.53 to $50.21. Half the company was gone between one close and the next.
Seven months later, on September 23, the FDA's panel of outside advisers on molecular and clinical genetics devices met to consider GRAIL's application to sell the test to adults aged 50 and over. Was it safe? Unanimously, yes. Effective? Six to four. Worth its risks? Seven to two, with one abstention. The vote is advice. The agency decides, and GRAIL expects the decision "in the coming months." The shares had closed at $80.77 on September 18. By October 2 they were $143.26, nearly three times their price on the day after the miss and 41% above it the day before.
So a trial missed, a panel said yes anyway, and the stock went around the track and past the start. The test is better than its headline. The law that would pay for it is slower than the price.
In February 1735 the Pennsylvania Gazette printed a letter from a reader who described himself as old and lame of his hands, unable to help when his neighbors' houses caught fire, and who offered advice instead. Do not carry live coals from one room to another in a full shovel, he wrote, rather than in a shut warming pan, "for Scraps of Fire may fall into Chinks, and make no Appearance till Midnight", when the stairs are already in flames and the only way out is the window. The reader was Benjamin Franklin, who was 29 and, as far as anyone knows, perfectly able to carry a bucket. The same letter carries the line that doctors still quote, that an ounce of prevention is worth a pound of cure. He was writing about fire.
A cancer is a scrap of fire in a chink. For years it may make no appearance at all; then the symptoms come, and for many people they come late, at stage III or IV, when the disease has spread and the odds have turned. A screening test tries to see the scrap before midnight. Galleri does it from a tube of blood. Tumors shed fragments of their DNA into the bloodstream, and the pattern of chemical tags on those fragments differs from the pattern on healthy cells; the test reads the tags, says whether it sees a cancer signal, and if it does, says where in the body the signal most likely comes from.
As a sieve it is very good. Of every hundred people it called positive across the trial, 52 had cancer within a year, and in the first round, when cancers that had grown quietly for years were all there to be found, 58. For comparison, the same paper notes that when a family doctor in England sends a patient for an urgent check for suspected cancer, about six in a hundred turn out to have it. When the test said no cancer, it was right about 99 times in 100. When a signal proved to be cancer, the test's first or second guess at where it came from was right about nine times in ten. That matters in practice. A positive result has to lead to a scan of one part of the body, and a wrong guess sends a frightened person down the wrong corridor.
The trouble is the holes. Across all cancers, the test caught only about three in ten of those diagnosed in the year after a draw; for the twelve deadly cancers the trial was built around, a little over half. It caught 32 of 36 cancers of the liver and bile ducts and 4 of 157 melanomas. Nearly half of what it missed were cancers of the prostate and breast; England screens for breast cancer but not for prostate cancer. Two thirds of the misses were at stage I or II.
The sieve also catches more as people age, because there is more to catch. In the second and third rounds, the test found a cancer in one of every 698 people tested in their fifties, one in 265 in their sixties and one in 153 in their seventies. False alarms rose almost in step, so a positive meant about the same thing at every age. What changed was how often there was anything to find. The seventies yield four and a half times what the fifties do, and the law, as we will see, reaches the seventies last.
Every cancer keeps time. It starts small, at stage I, and either stays put or moves on to stage II, stage III, stage IV; at any stage it may be found because the patient feels something and sees a doctor. A test drawn once a year looks in on that clock once a year. Whether it finds a cancer depends on how long the cancer spends at each stage and how visible it is there, and Galleri sees early cancers poorly. In GRAIL's own validation study it detected 17% of stage I cancers of all kinds, 40% at stage II, 77% at stage III and 90% at stage IV (Klein and colleagues, 2021).
A test with that profile mostly catches cancers that have already reached stage III, some of which would have been diagnosed later at stage IV. Each of those it finds moves a diagnosis from stage IV to stage III. The trial's main measure counted stages III and IV together, so that move counted for nothing, and in the first round it counted against the test, because screening found stage III cancers that would otherwise have surfaced only after the trial's window closed. The round-by-round numbers fit that account. Stage III and IV together were 19% higher in the screened group in the first round, 5% lower in the second and 12% lower in the third: 706 against 688 over the three years, a rate ratio of 1.03. Stage IV fell 14%, from 397 to 342, most of all in the third round. Stage III rose 25%, from 291 to 364.
None of this is our discovery. GRAIL's Harpal Kumar wrote before the results that a shift from stage IV to stage III was where the largest gain would lie; Christin Glorioso pointed out on Substack in February that such a shift would not register as a win on the combined endpoint; and Nitzan Rosenfeld of Barts Cancer Institute wrote in BJC Reports in August that the rise in stage III had masked the fall in stage IV. Others read the numbers more harshly. Richard Houlston of the Institute of Cancer Research called the stage IV to stage III explanation "entirely speculative" and said the researchers had "presented their findings far more positively than the overall results justify." David Weinberg and Michael Bretthauer, in a letter to The Lancet, saw no convincing evidence of benefit. They put cancers of all kinds at 5.1% of the screened group against 4.8% of controls, and said the small rise in early-stage cancers could be a benefit or a sign of overdiagnosis: cancers that would never have caused symptoms or death.
The question that matters for the future is whether more years of screening would have changed the answer. Ruth Etzioni's group got there first. In a commentary by Gogebakan, Lange, Gulati and Etzioni, published online on September 16, they fitted a model to the trial's round-by-round results and found the fall in late-stage cancers leveling off at 13% to 17% a round. We built our own, separately, to split stage III from stage IV. It is the kind of model Marvin Zelen and Manning Feinleib introduced in 1969, here with four stages and two kinds of cancer, fitted to fourteen published counts and ratios for the twelve cancers. The fit is close, though not exact; the end of the note sets it beside the trial. It needed two kinds of cancer to do it. About six in ten of these cancers move slowly, spending more than a year at stage III and rarely reaching stage IV before symptoms bring them in. The other four in ten move fast, reaching stage III quickly and spending only about three months there before stage IV, so that a yearly draw has roughly one chance in four of catching one in time. How the cancers move through stages I and II, the published totals cannot tell us. We do not lean on it.
Run forward six years, our model settles where the trial's later rounds already were: stages III and IV together about 9% below the unscreened level, and stage IV about 21% below. Refitted 24 times to the trial's numbers, each time redrawn within their sampling error, the first figure ranged from 1% to 14% and the second from 12% to 27%. The trial's designers, in a simulation published in 2022, had expected about 20% fewer cancers at stages III and IV, which is outside that range. It was never close. With this test and a yearly draw, both models say, that goal was out of reach however long the trial ran. The fall in stage IV looks real. In our model it lasts.
Two cautions. The model is fitted to published totals, never to patients' records, and its two kinds of cancer are a convenient summary of twelve diseases that differ in every way that matters. The trial cannot tell this mixture from others that fit as well. And a shift from stage IV to stage III is worth very different amounts for different cancers, which is the next question.
What matters in the end is deaths, and the trial will count them in stages: deaths from the twelve cancers at its main data cut, not yet published and still early, and deaths from all cancers three and six years after the last visit, around mid-2027 and mid-2030. The trial registry lists completion in January 2031. Until then the stage counts give an estimate. It is smaller than the rally suggests.
In the second and third rounds, a net 59 fewer people in the screened group were diagnosed at stage IV with the ten of the twelve cancers on the trial's stage IV slide (anus and myeloma, with at most one stage IV case in either group, are left off). Three of them, pancreas, lymphoma and ovary, went the wrong way. Suppose each of the 59 was diagnosed at stage III instead and gained the full difference in five-year survival between the two stages: 53 points for a cancer of the colon or rectum, 12 for lung, under 7 for pancreas. That comes to about 14 deaths averted over two years among some 71,000 people, roughly 10 for every 100,000 screened each year. Starasvetskay and Lopes made the same kind of calculation first, in a preprint on September 23, and counted it in years of life. Now add a term far more generous than the evidence allows. Treat each of the 88 extra early-stage cancers as a stage III cancer caught in time, each worth 40 points of survival, though some were surely cancers that would never have harmed anyone. The total rises to about 26 per 100,000 a year. Lead time pulls the other way: a cancer found earlier looks as if it lives longer even when the end comes on the same day, and survival gaps between stages in a registry are not the effect of moving one person from one stage to another. Treat 10 to 26 as a generous estimate.
Counted per test actually drawn, that is between about 3,500 and 9,300 tests for each death averted, before anyone counts the scans and biopsies that follow a positive. At today's price of about $705 a test, that is $2.5 million to $6.5 million of tests per death averted; at Medicare's price, $1.8 million to $4.7 million.
The trial's designers expected the fall in deaths to be 40% to 45% of the fall in late-stage cancers. On the 9% that yearly screening appears able to deliver, that would be about 4%. Whether the trial can see so small a change is disputed in GRAIL's own camp. Kumar wrote last November that it was "not designed or statistically powered to detect a reduction in cancer-specific mortality." Its designers, writing in 2025, said it had good power to study cancer deaths. Our rough count sides with Kumar. Even the generous estimate would sit about one and a half standard errors from zero in a count of deaths five years on.
That arithmetic is why the next door matters. The U.S. Preventive Services Task Force decides which screening tests most insurers must cover with no charge to the patient. Its procedure manual says it gives "greater weight to evidence of an effect on health outcomes," and for cancer screening the benefit it looks for is fewer deaths. It has graded on modeling before: the stool DNA test entered its colorectal recommendation in 2016 on its accuracy and modeled years of life, with no trial of deaths of its own. But it has no review of multi-cancer tests under way. A trial that cannot show fewer deaths for years gives it little to grade.
On February 3, 2026, the President signed the Consolidated Appropriations Act, 2026. Its Section 6221 is the Medicare benefit that cancer advocates had sought for years under the name of Nancy Gardner Sewell. It is a careful piece of drafting. It covers tests furnished on or after January 1, 2029, and only tests the FDA has authorized. Medicare must first use its national coverage process and find the test "reasonable and necessary." It has not started. A test is paid only if the last one was more than eleven months earlier. The patient pays nothing. The price is fixed. Through 2030 it is the amount Medicare paid for a stool DNA screening test on the day the law was signed, which for Cologuard was $508.87, and from 2031 that amount or the test's own fee schedule rate, whichever is lower.
Then come the ages. No payment for anyone under 50, and none for anyone who, on January 1 of the year, has reached a cap: 65 for 2029, rising one year each year after. Because the cap and everyone's age rise together, the only people who will ever qualify are those who were under 65 on January 1, 2029. Nobody older ever joins. A woman who turns 66 in March 2029 is outside the benefit for the rest of her life, unless the Task Force grades the test A or B and Medicare then covers it as a preventive service, which lifts the age and frequency limits. The House report on the bill describes exactly this design, covering seniors from 65 and keeping them covered for life, and Tori Lazerson, writing for Berkeley Law's Life Sciences Forum in August, called the age rule counterintuitive. Some summaries still describe the bill as introduced, which started in 2028 with a cap of 68; the enacted text says 2029 and 65.
"So teach us to number our days," the Psalmist prayed, "that we may present a heart of wisdom" (Psalm 90:12). Section 6221 numbers them for us, one birth year at a time, and it begins with the ages where the test finds the least.
| Year | People 65 to 79 who qualify | Share of people 65 to 79 | Share of the cancers the test would find at those ages |
|---|---|---|---|
| 2029 | 2.0 million | 4% | 3% |
| 2030 | 6.0 million | 11% | 8% |
| 2032 | 13.3 million | 25% | 17% |
| 2034 | 20.4 million | 39% | 28% |
| 2036 | 27.7 million | 52% | 44% |
| 2039 | 37.2 million | 72% | 67% |
| 2044 | 50.3 million | 100% | 100% |
People by single year of age from the Census Bureau's 2023 projections; the test's yield by age from the trial's second and third rounds. Medicare's beneficiaries under 65, who qualify from age 50 if they are on Medicare through disability, are left out because the trial says nothing about their yield.
In 2032, the year our valuation looks to, the law reaches about 13 million Americans over 65, a quarter of those between 65 and 79, holding about a sixth of the cancers the test would find at those ages. The people in their seventies, where a test is most likely to find a cancer, arrive last: the first of them in 2034, the last in 2044. We have not seen Congress's reasons. A cap that starts low and rises slowly is the cheapest way to say yes, and it is also the way that keeps the test longest from the people it serves best.
In 2008 we spent most of a year trying to buy smaller competitors. We signed confidentiality agreements, they opened their books, and they all said the same thing. As I wrote in my book, The Scholars' Treasure, the refrain was: "next year we'll surely be at $5 million in sales with massive profits, so, therefore, you should pay us at least $5 million." One of them took us out on a boat to talk about a merger, and the main thing I remember is how long the boat ride lasted. None of the deals made sense. Then I stood before a buyer's executives, at a mahogany table in a room full of microphones, and offered them my own company for $40 million to $50 million on a trajectory from $15 million of revenue to $50 million in three years. "Yes, I realize this 'pretend and hope for the future revenue-based valuation' is the same technique our not-so-flattering imitators employed on us," I wrote afterward. The buyer offered one times revenue. I said yes.
A share price is that conversation held in public, all day, with buyers and sellers trading places by the minute. A reverse valuation asks the price to say out loud what future it is paying for. Here is GRAIL's. At $143.26, with 44.7 million shares and 2.0 million prefunded warrants that are shares in all but name, GRAIL's market value was $6.7 billion on October 2. Less $862 million of cash and short-term securities at the end of June, the business itself was valued at $5.8 billion, about 35 times its revenue of the last four quarters. GRAIL still loses money on its tests under standard accounting; on its own adjusted measure, which leaves out non-cash charges, its gross margin was about 48% in the second quarter. Suppose it funds its losses to 2029 (it guided to burning less than $300 million in 2026), reaches a 30% operating margin by 2032, which would be a fine margin for a diagnostics company, and is then worth 20 times operating profit, with every year discounted at 10%. Then the price needs Galleri sales of about $1.9 billion in 2032.
| What the price needs in 2032 | At today's price, about $705 a test | At Medicare's $508.87 |
|---|---|---|
| Tests a year | 2.7 million | 3.7 million |
| Growth a year from today's 234,000 | 50% | 58% |
| Across 8% to 12% cost of capital, 15 to 25 times profit, 25% to 35% margins | 1.6 to 4.8 million | 2.3 to 6.6 million |
| At a 15% cost of capital | 3.5 million | 4.8 million |
Galleri's price per test is screening revenue divided by tests in the first half of 2026; GRAIL reports tests as "more than" a count, so this is the most it could be. None of the inputs is a forecast.
Ten percent is a kind rate for a company this far from profit. At 15%, nearer what investors usually ask of one, the price needs about 3.5 million tests. The base path runs from 234,000 tests a year today to about 351,000 in 2027, 527,000 in 2028, 792,000 in 2029, 1.2 million in 2030, 1.8 million in 2031 and 2.7 million in 2032. Can GRAIL do it? Its tests grew 35% to 50% a year in the first half of 2026, while the price per test fell 8% to 9%, and the company has guided to sales growth of 22% to 32% this year. Chan, Karceski and Lakonishok found in 2003 that very few companies sustain high growth for long, and no more of them than luck alone would produce. Fifty percent a year for six years is the low end of what my own firm grew from 1995 to 2000, before it settled at 20% to 30% a year from 2002 to 2007. Every adoption curve bends. Mine did.
The outside view here is Cologuard, the stool DNA test whose Medicare rate the new law borrows, now owned by Abbott, which bought its maker, Exact Sciences, in March. Cologuard went from 244,000 tests in 2016 to 1.7 million in 2019, 91% a year, faster than the path above, from almost exactly where Galleri stands now. It had help that Galleri will not. The FDA approved it and Medicare agreed to cover it within two months of each other in 2014, for everyone from 50 to 85, every three years; in 2017, Medicare paid for nearly two thirds of its screening revenue. The Task Force listed it in 2016. With that came coverage by most insurers at no cost to the patient. Galleri will have Medicare for one birth year at a time and no Task Force review in sight.
Count it in dollars, because Medicare pays less per test. If one in ten of the 13.3 million people the law reaches in 2032 were tested, Medicare would bring in about $680 million, 36% of the $1.9 billion the price needs; if one in twenty, 18%. The rest, at today's price, is 1.7 million to 2.2 million tests from employers, life insurers, health plans and patients, seven to nine times today's entire volume, for a test with no guideline behind it. Today that channel brings in more than nine dollars in ten of Galleri's revenue. The only sell-side figure we found in print points the same way. UBS's Doug Schenkel, as Dow Jones reported on September 23, puts Medicare at about $300 million of GRAIL's sales in 2030, about what one in ten of the 6 million people the law reaches that year would bring.
There is a fair case for the price, and it is worth putting at full strength. First, cadence: Cologuard is taken every three years and Galleri every year, so 2.7 million Galleri tests means about 2.7 million people, while Cologuard's 1.7 million tests in 2019 came from a larger group cycling through. Counted in people won over, Galleri needs fewer than the comparison suggests. Second, Cologuard is one case, and forecasts built on several analogies tend to beat forecasts built on one (Lovallo, Clarke and Camerer, 2012). Third, and strongest, a price is an average of outcomes, not a single path. If the losing outcome is worth about GRAIL's cash, today's market value is a coin flip on a company worth $12.5 billion, or a one-in-four chance of one worth $24 billion. Neither needs the middle path. Hendrik Bessembinder showed in 2018 that about 4% of listed companies account for all of the stock market's gain over Treasury bills; a price like GRAIL's is partly a ticket in that draw. The winning state needs FDA approval, insurers who pay, a death count that clears the bar and a Task Force grade that lifts the age cap, and those events would move together, which is why averaging them misleads (Jensen's inequality, which Sam Savage called the flaw of averages). On the evidence in this note, each link in that chain is years away, and the death count is the weakest.
One more thing the price has to carry. Since October 2025 GRAIL has sold about 5.4 million new shares and 2.0 million prefunded warrants, through a private placement, its at-the-market program and Samsung's investment, and $189 million remains on the at-the-market program. Its latest quarterly report replaced the promise of cash "into 2030" with "at least the next 12 months." A rally is the natural moment to sell shares. Every share sold divides the future among more owners.
None of these is a recommendation; each is a claim the market will test, with what would prove it wrong.
GRAIL: the price needs Cologuard's pace without Cologuard's coverage, or long odds on something much larger. The reverse valuation asks for about 2.7 million tests a year by 2032, 50% growth a year, or 3.5 million at a sterner discount rate. Cologuard grew faster with Medicare for everyone from 50 to 85 and a guideline. Galleri's Medicare benefit begins in 2029 for one birth year at a time, at a price below today's, and even at one in ten of those it reaches it supplies about a third of the revenue needed. GRAIL's filings say it is looking at Medicare Advantage plans, which can pay for more than traditional Medicare does, as a nearer road. That is the side door to watch. Falsified if Galleri's tests run at or above the path, more than 84,000 in the first quarter of 2027 and 351,000 in the year, or if a large national insurer covers the test for members over 50 within a year of FDA approval.
GRAIL: the deaths will come too slowly to open the guideline door this decade. On the trial's own counts, a generous estimate gives 10 to 26 deaths averted per 100,000 people screened each year, and on the designers' own rule of thumb the fall in deaths would be about 4%. The fuller counts of deaths are not due before mid-2027. A change that small would be hard to tell from chance in a trial of this size. Without the Task Force's A or B grade, insurers need not cover the test and the Medicare age cap stays. Falsified if any of the trial's counts of deaths shows significantly fewer deaths from the twelve cancers or from all cancers, or if the Task Force posts a draft research plan on multi-cancer screening by the end of 2027.
Guardant Health: the covered road. Guardant is building its multi-cancer business on top of a blood test Medicare already pays for. Its Shield colorectal test, approved by the FDA in July 2024 and covered by Medicare every three years from age 45, sold about 66,000 tests in the second quarter, four times as many as a year earlier and more than Galleri's 61,000, and Guardant expects 270,000 to 285,000 this year. From April 2025 Medicare paid $1,495 a Shield test, almost three times what the new law will pay for a test that looks for many cancers. Since late 2025 a doctor ordering Shield can tick a box and receive a report on several other cancers from the same blood, with no extra testing. That report is not authorized by the FDA, it is not what Section 6221 pays for, and Guardant's filings do not mention the law. At $177.80 Guardant was valued at about $24 billion on October 2, mostly for the tests that guide cancer treatment, which brought in two thirds of its revenue in the second quarter, so the multi-cancer race moves it far less than it moves GRAIL. Our claim is that coverage, more than the number of cancers a test looks for, decides volume: the first multi-cancer reports to reach Cologuard's scale this decade are likelier to ride a covered single-cancer blood draw than to arrive through the new law. The road has its own risk. Medicare tightened its rules for blood-based colorectal tests in June 2026, and Shield must now complete the study the FDA required after approval to keep its coverage. Falsified if Galleri's tests outnumber Shield's in any quarter of 2027.
Likely, we can get some of it from primary research, which is what Vista is all about. The people who wrote coverage rules, bought benefits and graded screening tests know now what the statistics will take years to show.
As a matter of general past practice, how long did a new screening category take from FDA approval to a final coverage decision, and what evidence did it need to be found reasonable and necessary?
The answer that would change the view"Under a year, on accuracy data alone."
Would your plan have covered an FDA-approved multi-cancer blood test for members over 50 at about $700 a year with no Task Force grade, and what would have decided it?
The answer that would change the view"Yes, within a year of approval, for every member over 50."
Could a multi-cancer test be graded on stage shift and modeling, as stool DNA was in 2016, or would the Task Force wait for deaths?
The answer that would change the view"Modeling could carry it, and a review could open within two years."
For a client engagement, Vista combines this report with that primary research.
We will score this note on May 31, 2027, and publish the result on the scorecard whether it flatters us or not. Each test sides with the price:
Three or four passes and the price's reading gains ground; one or none and ours holds; two is too early to say. Four yes-or-no tests cannot settle a question about size, as Nassim Taleb has warned; they can show which way the evidence is moving.
Franklin's letter did more than give advice. It held up the clubs of a city in a neighboring province, Boston, whose members kept a fire engine, worked it once a quarter, and met to "discourse of Fires, of the Faults committed at some", and in time Philadelphia had its own fire company, which Franklin's autobiography traces to a paper on fires that his editors identify as this letter. The advice was cheap. The company was what made it work: people who agreed in advance, and kept agreeing, to pay for the looking. Galleri has its sieve, a fine one with holes. What it does not yet have is the company.
The bet. That Galleri's sales grow about 50% a year to some $1.9 billion in 2032, roughly 2.7 million tests a year, carried by FDA approval, a slow Medicare benefit and buyers who pay without a guideline. Or that the long odds of a much larger outcome are better than they look.
The payoff. If that path holds, the price returns the 10% a year we discounted at, on a generous margin and multiple; across reasonable inputs the price needs 1.6 million to 4.8 million tests in 2032, and about 3.5 million at 15%. Medicare can supply about a fifth to a third of the revenue that year, if one in twenty to one in ten of those it reaches are tested.
Our read. The test is better than its trial's headline: a positive means cancer half the time, and stage IV diagnoses fell 14% in the trial and by about a fifth in our model. The trial's design asked for a fall in stages III and IV that this test cannot deliver with yearly screening, and the deaths that would open the guideline door will be few and slow to count. The law that pays begins in 2029, with the youngest and least productive ages, and never reaches today's older Americans. The price assumes the rest of the volume arrives anyway, or that a long shot comes in.
What settles it, and when. FDA approval and GRAIL's first-quarter 2027 volumes, by May 31, 2027; the trial's counts of deaths, the fuller ones from mid-2027. Before then, we would talk with a former Medicare coverage analyst, a former benefits buyer at a large employer or health plan, and a former Task Force member.
Galleri finds late cancers far better than early ones, so most of what it changed was the stage at which late cancers were found. Stage IV diagnoses fell 14% and stage III rose 25%, and the trial's main measure counted the two together, which rose 3% (a rate ratio of 1.03).
Not before 2029, and only after the FDA authorizes it and Medicare completes a national coverage decision. A law signed on February 3, 2026, Public Law 119-75, Section 6221, pays for authorized multi-cancer tests from January 1, 2029, at the Cologuard rate of $508.87 through 2030 (or $591.92, if Medicare uses Cologuard Plus's rate). An FDA advisory panel voted 7 to 2 for Galleri on September 23, 2026; the FDA's decision is pending.
People aged 50 or older who were under 65 on January 1, 2029. The age cap starts at 65 and rises one year each year, so anyone 65 or older on that date never qualifies under this benefit, unless the U.S. Preventive Services Task Force gives the test an A or B grade and Medicare then covers it as a preventive service.
In the NHS-Galleri trial about 52% of positive results were cancer, negative results were right about 99% of the time, and when a signal proved to be cancer, the test's first or second guess at the organ was right about nine times in ten. It caught about three in ten of the cancers diagnosed in the following year, and a little over half of the twelve deadly cancers the trial focused on.
GRAIL's list price is $949, and its website offers the test for $699 when requested online. GRAIL's revenue per test in the first half of 2026 was at most about $705. The new Medicare law would pay $508.87, or $591.92 if Medicare uses Cologuard Plus's rate.
At $143.26 on October 2, 2026, and on generous assumptions, about $1.9 billion of Galleri sales in 2032, roughly 2.7 million tests a year, or growth of 50% a year for six years from about 234,000 tests a year today; or else a real chance of a much larger outcome. This is research, not a recommendation.
Every figure was computed in code (model.py and boot.py, standard library only) from the inputs below, and each input was checked against its primary document. The sieve: positive results, cancers confirmed within twelve months and people tested in each round, from Neal and coauthors (2026) and its Supplementary Table S2 for age at enrollment; the repeat-round yield by age pools rounds 2 and 3. The clock: a model of the kind introduced by Zelen and Feinleib (1969), with four preclinical stages and two kinds of cancer that differ only in the speed of progression, starting from the unscreened steady state and solved exactly year by year (a matrix exponential). A cancer moves from stage to stage at rates g, is diagnosed through symptoms at rates c by stage, and is found by a screen with probability s by stage; s starts from the stage sensitivities of Klein and coauthors (2021) for all cancers and is shifted by one fitted amount on the log-odds scale for the twelve prespecified cancers. The share of the screened arm tested in each round is the trial's (98.9%, 90.7%, 87.6%). Ten parameters were fitted by least squares (Nelder and Mead's method, five starts, three seeds reaching the same optimum) to fourteen published quantities for the twelve cancers: the control arm's stage mix, the screened-to-control ratio at each stage over three years, the stage III and IV and the stage IV rate ratios by round, and episode sensitivity by round. The fit puts 41% of these cancers in a fast kind that spends about three months at stage III and 59% in a slow kind that spends about fourteen months there; how cancers move through stages I and II, and the sensitivities there, are not identified by published totals, and nothing in the note rests on them; the fast kind's speed sits at the model's limit, which in effect means these cancers reach stage III almost at once. The projection keeps everyone who came to all three rounds coming once a year. The range comes from 24 refits, each to the targets redrawn within their sampling error: rate ratios on the log scale with the spread of their published 95% intervals, counts as Poisson, sensitivities as binomial. From stage to deaths: stage IV diagnoses in rounds 2 and 3 by cancer for the ten of the twelve cancers on the trial's stage IV slide (ASCO 2026, slide 13; anus and myeloma, with at most one stage IV case in either group, are left off), each multiplied by the difference between five-year net survival at stage III and at stage IV, net of the three that rose; per 100,000 a year divides by two years of the screened arm (71,122 people). The generous term gives each of the 88 extra stage I and II diagnoses of the three years 40 points of survival. Tests per death averted divide by tests actually drawn (126,821 in rounds 2 and 3; 197,146 in all three). The check on the five-year look estimates deaths within five years among the control arm's cases of the twelve cancers from their stage counts (stage III and IV survival weighted by the control arm's stage IV cases by cancer; stage I and II survival assumed at 70% or 80%, giving about 630 to 690 deaths) and expresses the deaths averted over three years as standard errors of the log rate ratio: 0.6 to 1.6. The law: Census single-year ages at July 1, with half of the cohort at the cap counted as under it on January 1; the yield at each age is the trial's repeat-round yield for its decade. The price: market value is the October 2, 2026 close times the shares on the second-quarter 10-Q cover plus the prefunded warrants; enterprise value subtracts cash and short-term securities at June 30, 2026. Revenue needed in 2032 solves enterprise value plus the present value of assumed losses ($130 million in the second half of 2026, then $250 million, $200 million and $100 million in 2027 to 2029, each discounted from mid-year) compounded six years at the cost of capital, divided by the multiple of operating profit and by the margin; the method is the reverse valuation of Rappaport and Mauboussin's expectations investing. Tests are that revenue over the price per test, and the path compounds from twice the first half's tests. Medicare's supply in 2032 multiplies the people 65 and over who qualify by a share tested in the year, at $508.87 a test. The two-state reading sets today's market value equal to p times a winning value plus (1 − p) times the cash, and solves for the winning value at p of one half and one quarter. Guardant: Shield tests by quarter from Guardant's releases; market value at the October 2 close on Nasdaq; the convertible notes at principal.
| The model beside the trial (twelve cancers) | Model | Trial |
|---|---|---|
| Control group's stage mix (I to II, III, IV) | 0.45, 0.23, 0.32 | 0.45, 0.23, 0.32 |
| Screened over control, three years (I to II, III, IV) | 1.16, 1.25, 0.83 | 1.16, 1.25, 0.86 |
| Stage III and IV rate ratio, rounds 1 to 3 | 1.20, 0.91, 0.90 | 1.19, 0.95, 0.88 |
| Stage IV rate ratio, rounds 1 to 3 | 0.94, 0.77, 0.79 | 0.91, 0.78, 0.74 |
| Episode sensitivity, rounds 1 to 3 | 0.57, 0.50, 0.50 | 0.63, 0.48, 0.51 |
| Three rounds together: stage III and IV; stage IV | 1.00; 0.83 | 1.03; 0.86 |
| Input | Value | Source |
|---|---|---|
| Positive results, cancers, people tested by round | 722, 419, 70,325; 518, 261, 64,498; 561, 257, 62,323 | Neal and coauthors (2026) |
| Cancers found, people tested, rounds 2 and 3, by age | 50 to 59: 47 of 32,783; 60 to 69: 195 of 51,741; 70 to 79: 276 of 42,297 | Same, Supplementary Table S2 |
| Twelve cancers, three years, screened vs control | I to II 647 vs 559; III 364 vs 291; IV 342 vs 397 | ASCO 2026 abstract LBA100 and slides 8 to 11 |
| Stage III to IV and stage IV rate ratios by round | 1.19, 0.95, 0.88; 0.91, 0.78, 0.74 | Same, slides 10 and 11 |
| Episode sensitivity, twelve cancers, by round | 63.4%, 47.6%, 50.7% | Neal and coauthors (2026) |
| Stage sensitivities, all cancers | 16.8%, 40.4%, 77.0%, 90.1% | Klein and coauthors (2021) |
| Stage IV in rounds 2 and 3, screened vs control | bladder 1 vs 5; liver and bile duct 4 vs 14; oesophagus 9 vs 21; head and neck 10 vs 16; colorectal 21 vs 32; lung 50 vs 73; stomach 6 vs 8; lymphoma 38 vs 37; ovary 8 vs 7; pancreas 33 vs 26 | ASCO 2026, slide 13 |
| Five-year net survival, stage III and IV, England | colorectal 64.2, 11.0; lung 16.7, 4.5; pancreas 8.7, 2.1; and seven more in model.py | NHS Digital, cancers diagnosed 2018 to 2022, as shown on ASCO 2026 slide 13 |
| Share price, October 2, 2026; shares; prefunded warrants | $143.26; 44,666,234; 1,998,573 | Nasdaq; GRAIL 10-Q for the second quarter of 2026 and 10-K for 2025 |
| Cash and short-term securities, June 30, 2026 | $861.6 million | GRAIL 10-Q |
| Screening revenue and tests, first half of 2026 | $82.5 million; more than 117,000 | GRAIL first- and second-quarter 2026 releases |
| Revenue, last four quarters; adjusted gross profit, second quarter of 2026 | $165.3 million; $21.6 million of $44.7 million | GRAIL releases |
| Medicare payment for a stool DNA test on February 3, 2026 | $508.87 (Cologuard); $591.92 (Cologuard Plus) | CMS Clinical Laboratory Fee Schedule, 2026 |
| Completed Cologuard tests | 2015: 104,000; 2016: 244,000; 2017: 571,000; 2018: 934,000; 2019: 1.7 million | Exact Sciences 10-Ks for 2015 to 2019 |
| Shield tests, second quarter; 2026 guidance; Medicare price from April 2025 | 16,000 (2025), 66,000 (2026); 270,000 to 285,000; $1,495 | Guardant releases; CMS list of advanced diagnostic laboratory tests |
| Guardant shares; share price, October 2, 2026 | 134,196,170; $177.80 | Guardant 10-Q for the second quarter of 2026; Nasdaq |
| Cost of capital; multiple; margin; year | 10% (and 15%); 20 times operating profit; 30%; 2032 | Assumptions, varied in the text |
The model, its fitted parameters, the bootstrap and its full output are kept with the note's working files; the arithmetic in the text was checked against them.
This note is research, not investment advice. It states what prices and models assume under labeled assumptions; nothing here is a recommendation to buy, sell or hold any security. The decision belongs to the reader. As of October 4, 2026, Russ Rosenzweig, Vista's founder, owns options on Abbott Laboratories, CVS Health and UnitedHealth Group. He owns no shares of GRAIL, Guardant Health, Elevance Health, Cigna or Humana. Holdings through mutual funds and exchange-traded funds are not counted. How this note was made: written by Russ Rosenzweig with Vista's AI research desk. It began with a search of a year of Nature's journals that he asked for; the desk reviewed the draft, the models were built and every figure computed in code, and each source was checked against its primary document.