Short versionWrite down what they said two years ago, what they said last year, and what they say now. Slips are normal. What matters is how big the slip is, whether it came with a reason, whether the new date held, and whether the verbs have moved from "will begin" to "has begun."
1 · The ledgerEvery plant finishes late
That is not cynicism, it is the base rate. Optimism is how anybody gets a plant funded in the first place, and I have never seen one of these finish early (if you have, I would like to meet the project manager). So a slip, on its own, is not the finding. The PATTERN of slips is.
The method is boring, and it works. For each milestone the company has named, pull the date they gave in every quarterly update and line them up. Earnings releases and call transcripts are archived by quarter, and the SEC's full-text search finds a phrase across every filing since 2001. For a private company, the Wayback Machine will show you what its news page said on almost any date.
The figure shows how to read the result. The diagonal is the day each statement was made, so a promise that reaches it has been kept. A flat line is a date that held. A staircase that climbs alongside the diagonal is a date that moves a quarter every quarter, and never arrives.
1 · The ledgerFour questions for every row
- How big is the slip, next to the size of the job? A quarter on a four-year build is noise. A year on a two-year build is a replan, and should have been announced as one.
- Did it come with a reason, and a specific one? "Customer qualification is taking longer than planned" is a reason. "Macroeconomic headwinds" is weather.
- Did the new date hold? One slip is normal. A date that moves one quarter to the right every quarter is a date nobody believes internally either.
- Did a milestone quietly disappear? A date that simply stops being mentioned is the slip that never shows up anywhere, unless you are keeping a ledger. That absence is the finding.
1 · The ledgerWhat it looks like done well
I have used MP Materials' Independence magnet plant in Fort Worth, because it has the longest public record in the field and because it is a good one. Every point on the figure is something the company said in public.1 How the chart flatters and how it doesn't is in the notes.2
Finished magnets held. From early 2022 the company said magnets in 2025. It repeated some version of that at essentially every quarterly update for four years, and it began making magnets on commercial equipment in December 2025. That is a long time to hold a date on a first-of-a-kind plant, and it is the line I would point to if somebody asked whether this team plans honestly.
1 · The ledgerMetal slipped a year, and said why
The original plan led with alloy for GM in late 2023. In November 2023 that date stopped appearing. What appeared instead was trial production of NdPr (neodymium-praseodymium) metal, and an explanation: the company had "shifted the focus to metal… because that's really more of a salable broader product," and was moving "methodically and thoughtfully" because "it's our capital risk."3
A new date (metal during 2024) came the next quarter, and commercial metal was announced in January 2025, three weeks past it. That is a slip. It is also exactly what a well-managed slip looks like: a reason, a new date, and a hit.
1 · The ledgerDeliveries are the open question
In January 2025, first deliveries were "on track for year-end." In November 2025, magnet revenue moved to "the second half of 2026," tied to qualification with GM. In August 2026 it became the fourth quarter, with the chief operating officer describing "a wide range of technical, operational, and customer-driven challenges."4
While a year is a real slip, a year to qualify a new supplier's magnets into a traction motor is realistic. Automotive qualification is slow for everybody, and it runs on the customer's calendar, not the supplier's. It is also the line to watch. The next entry in the ledger is either "has begun commercial shipments," or another quarter.
In practiceEvery plant start-up I have been part of had a schedule that was optimistic on the day it was written, and on some of those days I was the one holding the pen. What being on the inside taught me is that the first date is a guess and the SECOND date is information. By the time a team gives its second date, it has met the equipment, the vendors and the customer, and knows where the time actually goes. So I weight the second date far more heavily than the first, and I watch very closely whether it holds.
1 · The ledgerWill begin, has begun
The single most useful thing in the ledger is the verb. Announcements climb a ladder of tenses, and each rung is harder to write than the one below it: a plan, then money being spent, then equipment being commissioned, then trial production, then commercial production, and finally revenue.
MP's January 2025 release is a good example of the careful version. In one sentence it said the plant "has commenced commercial production" of NdPr metal and "trial production" of magnets.5 Two different rungs, in the same sentence, and it said so. A company that uses words precisely when precision is unflattering is a company whose flattering sentences you can believe.
The top rung is the only one the communications department cannot write. By the third quarter of 2025 the company's magnetics segment was reporting $21.9 million of quarterly revenue from what it called magnetic precursor products: metal and alloy, sold.6 Not magnets yet, and the release did not pretend otherwise.
1 · The ledgerCost estimates against outcomes
Schedules are half the ledger. The other half is money, and for a public company the check is mechanical: find the capital budget they gave, then find the line called "additions to property, plant and equipment" in the cash flow statement for the years that followed.
In April 2022, MP described "a $700 million investment the company will make over the next two years," spread across its separations upgrade, the Fort Worth plant and heavy rare earth separation.7 Two years later the chief financial officer said they remained "within the margin of error on that assessment despite enduring inflation," while "some of that spend will slip to 2024."8 The budget held and the calendar stretched. For several plants built through the inflation of 2022 and 2023, that is a good outcome, and a rarer one than it should be.
Two cautions. Spending UNDER guidance is not automatically good news: money not spent is usually work not done, and it tends to reappear a few quarters later as a schedule slip. And watch whether the estimate keeps getting restated. One that stops being mentioned, like a milestone that stops being mentioned, has usually been replaced by a bigger one.
1 · The ledgerCount the announcements, too
This one is rougher. It is less a test than a way to develop gut feel. As Lil Wayne said, “Real G’s move in silence” — and it holds up in magnets better than it has any right to. In my experience, companies that are moving in a serious way make fewer statements, and the claims in them are much more measured. Actually making something happen does not tend to leave a lot of energy for burning up the newswire.
So scroll back through a year of a company's news page and sort it into two piles. One pile is things that HAVE happened: a shipment, a qualification, a commissioning, a hire into a plant role. The other is things that WILL: a memorandum of understanding, a "strategic partnership," a site selection, an award, a panel appearance. A busy news page that is mostly the second pile belongs to a company whose main product, for now, is announcements.
Two cautions. Listed companies are required to disclose material events, so some of the volume from a public company is obligation, not promotion. The quarterly results and the current reports filed with the SEC are not what I am counting. And knowing how to promote yourself is not a crime.
Is it dispositive? No. Some very capable companies are also very good at press releases, and a quiet company can be quiet because nothing is happening. But when the newswire gets louder while the ledger above stops moving, I pay attention.
1 · The ledgerBig sound, narrow substance
Be especially wary of a company that makes a lot of splashy announcements that SOUND good and turn out, read closely, to be much narrower than the first impression. There is a structural reason it happens. People who deal with investors want to sound as grand as possible. People who deal with customers have to set expectations correctly, because the customer is going to hold them to it. A press release is usually written by the first group.
The habit to build is to read to the last paragraph and ask what the smallest true version of the headline is. The figure has the translations I make most often. None of them is dishonest; each headline is usually literally true. Judge the company on the smallest true version, and notice which companies announce the small true version themselves. Those are generally the ones with customers.
Short versionPretend you have a drawing and a budget, and try to buy a magnet. Count the clicks to a human who can quote it. Established producers make that almost impossible to miss.
2 · The websiteCould a customer actually buy something?
The companion page has the quick version of this test. This is the long version, which I run on every company I follow, about once a year. Start on the home page with a real part in mind, say a sintered block for a servo motor that runs warm. Is there a path organized around what I am building? A grade I could put on a drawing? A named way to reach a person who sells? And how far down the menu is "Investors"?
Arnold Magnetic Technologies is a good model. The top menu leads with Products and Markets, Markets breaks out fourteen industries, and every product page ends in a "Contact a specialist" form that asks where you are and what your project is, and takes a drawing upload.9 VAC's contact page adds an "Experts by Region" finder and a list of distributors.10 TDK runs its magnet pages like an electronic-component catalog, with part-number search and a distributor inventory check.11 None of it is glamorous. All of it says: we have product, and we would like you to buy some.
Compare the site that leads with a mission statement, a groundbreaking photo and a video; puts "Investors" in the top menu and "Contact" in the footer; and whose only commercial language is that the company is "partnering with strategic customers." That phrase earns its own flag. It usually means one or two anchor buyers, and no product yet for anybody else. That is a legitimate stage. But it is a stage, and at that stage the one or two customers matter far more than anything on the website.
2 · The websiteThe leadership page, in miniature
An established producer's leadership list is weighted toward the people who run plants and the people who sell what comes out of them: operations, quality, and commercial leads by region or market. A company whose leadership page is mostly finance, strategy, government affairs and communications is telling you, accurately, what its job is right now. Count the titles. The ratio is the finding.
In practiceHaving run one of these businesses, I can tell you what an established magnet maker worries about most on its own website: a customer with a drawing who cannot find a salesperson. A lost inquiry is the most expensive thing that can happen on the site, because the customer does not try twice. They send the drawing to the next name on the list. That is why the contact button follows you around from page to page, and why its absence tells you something.
Short versionThe foundational neodymium magnet patents expired years ago. The live ones cover process. Ask what a company owns, what it licenses, and what happens if a customer gets sued, because the last time this was fought in the United States, most of the defendants were customers.
3 · The IPWho got sued last time
In 2012, Hitachi Metals (now Proterial) filed a Section 337 complaint at the US International Trade Commission over sintered rare earth magnets and products containing them. There were 29 respondents. Three were magnet makers. The other twenty-six were companies that put magnets into things: motors, audio speakers, headphones, cordless tools, computer hard drives, golf ball markers.12
A Section 337 case can end in an exclusion order that stops infringing goods at the border, and the goods at the border belong to the importer, which is usually the customer. That case ended in 2013 without a ruling on the merits,13 but the lesson stuck with every purchasing department that lived through it. When a magnet infringes, the exposure lands on whoever brought it in.
3 · The IPWhat is still live
The core composition patent, from Sumitomo Special Metals' 1982 filing, expired in 2014, and the four patents asserted at the ITC ran out in 2021.14 What remains live is process, above all grain-boundary diffusion: putting dysprosium or terbium only at the grain surfaces, where it does its work. It is how the industry now makes most of its high-temperature motor grades, and some of the patents on it are still running. One of Shin-Etsu's core diffusion patents, for example, is listed as active until March 2027.15 (The heavy rare earth page explains why those two elements matter so much.)
3 · The IPFour questions
- What patents do you own, granted or only applied for? Good: numbers you can look up. Noveon Magnetics holds granted US patents on its own grain-boundary engineering process,16 which is the right kind of IP to own: it sits where the fights are.
- What do you license, and from whom? Good: a named licensor, or "we can't say who, but yes, and it covers these grades." Many licenses are confidential, so the lack of an announcement proves nothing. Concerning: "we don't need one," said about a diffusion grade, with no reason given.
- How did you design around what you don't license? Good: a specific answer about a specific step. Concerning: "our process is proprietary," which answers a different question.
- Will you indemnify me? Good: an IP indemnity clause in the supply agreement, offered without a fight. This turns everything above into money, and if I were buying, it is the one I would ask first.
Credit where it is due on openness. Magnequench, the bonded-magnet powder business inside Neo Performance Materials, publishes a patent guide, runs a monthly program testing magnets from the market for infringing material, and says in plain words that "users of magnets or products containing magnets should contact us directly if they have any patent related questions."17 That is IP treated as a customer-service problem, which is exactly what it is.
Short versionAsk for the grade list. Not the roadmap: the grades they will quote and ship this quarter, each with a datasheet.
4 · The gradesThree lists behind "our grades"
Grades we are shipping to paying customers. Grades we have made and measured. Grades we are targeting. A company early in its ramp will have a short first list and a long third one, which is fine, as long as nobody blurs them.
The established producers make this easy to check. VAC's VACODYM page is a table of about seventy-five grades, each linked to its own datasheet.18 Arnold publishes a datasheet for each neodymium grade, with normal and intrinsic demagnetization curves from −40 °C to 120 °C.19 If you want to know what they make, it is written down, with the conditions attached.
4 · The gradesWhat to look for in a newer list
Where it starts. Starting in the middle of the grade ladder, with the workhorse grades that have modest coercivity, is normal and sensible. Launching with the full ladder, up to the high-coercivity grades traction motors need, is a claim about process maturity AND about IP, and both deserve a question. The NdFeB grade chart shows how far apart the two ends are.
Whether "automotive-grade" names a material or a status. A grade becomes automotive in the only sense that counts when a carmaker has approved the part through its production part approval process (PPAP), which commonly takes a year or more. Read it as "intended for," not "approved by," unless a customer is named.
What the datasheet was measured on. Production parts, both curves, more than one temperature, a revision date. The headline page's first question covers it properly.
4 · The gradesTwo legitimate shortcuts
One is an established parent. eVAC's plant in Sumter, South Carolina shipped its first US-made magnets in December 2025, with a stated capacity of 2,000 metric tons a year by early 2026,20 and its product pages point to VAC's existing grade catalog rather than a new one. The hard problem for a plant like that is transferring a process, not discovering one.
The other is buying a business that already operates. USA Rare Earth closed its acquisition of Less Common Metals, a UK maker of rare earth metals and alloys with more than three decades of history and an existing customer base, in November 2025.21 Buying a step that already works is a perfectly good way to turn "will" into "has" for that step.
Short versionAcquiring know-how is good. Acquiring a supply chain one company at a time can leave you with a plan that makes no sense on a map. Draw it, count the ocean crossings, and note which step each one happens at.
5 · The globeKeep a globe handy
The cautionary tale here is Molycorp, which reopened Mountain Pass in California and then set out to own every step from mine to magnet by buying them. In 2011 it bought a controlling stake in Silmet, a rare earth processor in Sillamäe, Estonia, and Santoku America, an alloy maker in Tolleson, Arizona, and formed a magnet joint venture in Japan.22 In 2012 it bought Neo Material Technologies for its separation plants in China and Magnequench's powder plants in China and Thailand, a deal its chief executive said "enhances our Mine-to-Magnets vision and places Molycorp in all steps of the vertical rare earth supply chain."23
Every one of those was a real business with real know-how. The trouble was the map.
5 · The globeWhat actually moved where
By the company's own filings: separated oxides went from Mountain Pass to Estonia and to Arizona to be made into metals and alloys, and most of what Arizona sold went to a single customer in Japan.24 From 2012 and 2013, unseparated rare earth concentrate went from California to Estonia (about two-thirds of Silmet's feed in 2013) and to the Chinese separation plants, while the magnet business got its oxide from those Chinese plants and had it made into metal at a partner in China.25
Put that on a globe and it is a mine in California feeding separation on two other continents, metal-making on a third, and customers on a fourth, with the bulkiest material taking the longest trips.
5 · The globeWhy the top of the funnel is the wrong place to travel
Here is the arithmetic, and it is mine. Mountain Pass concentrate runs a little over 60 percent rare earth oxide, of which about 15 percent is neodymium and praseodymium.26 A sintered magnet is about 30 percent rare earth by weight.27 So a metric ton of finished magnet starts as something like 3.8 metric tons of concentrate, most of it cerium and lanthanum the magnet does not want. Ship the concentrate across an ocean and you move roughly ten times the tonnage of the NdPr oxide you were after.
And a cost picked up at the top of the funnel does not stay the size it was. Every downstream step loses some material, so every dollar of freight on the concentrate is carried through every loss after it, and lands on the finished magnet bigger than it started. (The yield calculator shows how that compounding works.) That is a very unfavorable place to concentrate your costs.
5 · The globeWhat to take from it
While the rare earth price collapse after 2011 is what finally sank Molycorp (the company said so itself when it idled Mountain Pass in 2015),28 the map was, in my view, one of the fatal flaws in the plan, and a big part of why it could not survive the price. Mountain Pass's own capital estimate went from about $511 million at the 2010 IPO to about $1.45 billion by early 2013,29 and the company filed for Chapter 11 in June 2015.30
The lesson is not "never import anything." Where the crossings happen matters more than whether they happen. Metal and alloy late in the funnel are dense in value, and moving them is a small cost per kilogram of magnet. Concentrate at the top is the opposite. So when a company builds its chain by acquisition, get the globe out: put each step on it, draw the arrows, count the oceans, and note which step each one happens at. Acquiring know-how is good. What you are checking is whether the overall plan is coherent.
Short versionList the former senior people: which function, how long they stayed, where they went. Some turnover is healthy. A revolving door in the chair that runs the plant is not.
6 · Who leftRead LinkedIn like a roster
This is the check people skip, because it feels like gossip. It isn't. The people who have left a company know more about it than anybody. They will not tell you what they know (and should not; they signed the same kind of NDAs I have), but their public career histories tell you a fair amount on their own.
Search the company on LinkedIn, open its People tab, and look for former vice presidents, directors and heads of. Note the function, the tenure, and where each went next. Then pull the leadership page from a year or two ago off the Wayback Machine and compare. Whose name is missing, and was the chair refilled?
Function matters more than count. A couple of finance or marketing leaders in a few years is a normal young-company life. A third head of operations, or a third head of quality, in three years is a plant problem, because those are the people who find out first whether the process works.
Tenure tells you whether it was the plan. Senior departures inside a year of being hired are the signal. People who stay two or three years and leave at a change of stage, construction to operations or operations to scale, are usually the plan working. The people who build a plant and the people who run one are different people, and good companies change the team on purpose.
Where they went. To a competitor or a customer is a functioning labor market. Out of the industry altogether, several at once, is worth a quiet question. And do not mistake a transition for an exit: a founder stepping back or a new chief executive at a financing round is how companies grow up.
In practiceThe most honest reference check for a company is the same as for a person: ask somebody who used to work there. Not about anything confidential. Just "would you go back?" The length of the pause before the answer is most of the answer.
Short versionJob postings are a map of what the company is about to do. The mix of roles tracks the stage of a plant closely enough to date it.
7 · Who's hiredThe careers page dates the plant
Job postings are the one corporate document nobody edits for narrative. Builders come first (project engineers, construction, procurement), then commissioners (controls, equipment technicians, maintenance planners), then the ramp (process engineers, operators on several shifts, metrology), then production (quality, customer quality, application engineers, inside sales). The figure lays out the sequence.
Shifts. "Second shift" or a rotating twelve-hour schedule means the plant expects to run around the clock, which is a plant with a product. Customer quality. A customer quality engineer with PPAP experience is the most specific tell on the whole careers page: a customer is qualifying parts right now. Reposts. The same senior process role posted for six months is hard to fill, which is common here, because the people in the West who have run a sintered magnet line would fit in a medium-sized conference room. A staffing risk, not a red flag.
Finally, look at what is NOT posted. A company that says it is in production and is hiring for none of the production roles is either fully staffed, which is possible, or not yet where it says it is.