Corrections

Every figure on this tracker comes from a primary document, and sometimes what we say about that document is wrong. When it is, we fix it and we write it down here: what was wrong, how many records it touched, and what they say now. A system that finds and discloses its own errors is more trustworthy than one that hides them. A correction you can read is the only way to tell the two apart. We correct a record in place where the underlying document has not changed. We withdraw it where we should never have published it. Either way, nothing is ever silently deleted.

3 corrections

4,321 records affected

31 July 2026 318 records withdrawn applied 31 Jul 2026

Money nobody raised: takeovers paid in shares, running totals, and one round counted twice

318 published records reported money that no company raised. A Form D reports an "amount sold". For three kinds of filing, that amount is not capital arriving to be spent on anything. We published all three here as funding rounds. Between them they overstated the money total on this tracker by $14.25bn.

The first is a takeover paid for in shares. One company buys another and hands the sellers stock instead of cash. That stock goes on a Form D, and its value appears in the same box a startup uses to report a round. So one filing recorded that Danaher is acquiring Masimo at $180 a share, for "total consideration of $9.9 billion". We published it as Masimo raising $9.9bn. Another recorded a merger of W.D. Company, Inc. with and into Dillard's, Inc., valued off a closing share price with no cash anywhere in the document. We published that one as $2.39bn. The form asks the issuer outright whether the offering is part of a business combination, and 177 published records answer yes. That answer is a field in the public data set, and we had never read it.

Seven of those 177 stay. Those seven say, in the filer's own words, that cash came in and the company then spent it on a deal. Two examples: "a portion of the proceeds of the sale of securities to investors was used to acquire", "funds are being used to acquire a hospital". That is a real raise and the record is true.

The second is an offering with no ceiling that has been selling for years. Where the amount offered is the word "Indefinite", the figure filed is everything sold since the first sale rather than a round. OPTCAPITAL LLC's $1.77bn is one of these. It is the fourteenth annual amendment to an offering whose first sale was in July 2012, so the figure covers fourteen years of sales presented as one raise.

The third is the same round counted twice. A Form D amendment restates the running total for an offering already filed, so an offering we published twice is one raise on the page twice. Fluidstack appears at $450m in January and $842m in May: one offering, one raise, and the May figure is the whole of it. Its separate $730m offering opened in June under its own file number, and it is a different raise. That is why we match these on the offering's number and never on the company.

We withdrew all 318 on 31 July 2026 rather than restating them. What was wrong was not the figure, which is what the filing says. What was wrong is that we called the figure money raised at all, and there is no smaller true number to put in its place. We deleted nothing. A withdrawn record keeps its row and the reason it went, so we can still count it here.

What we projected, and what actually happened
	Before	We projected	Measured now
Funding records	3,344	3,026	3,026
Money raised	$122.0bn	$107.7bn	$107.7bn
Records drawn from Form D	3,013	2,695	2,695
Takeovers published as raises	177 records, $8.5bn	7 records, $0.7bn	7 records, $0.7bn
Employers with a funding record	3,127	2,906	2,937

Measured on the live tracker 31 July 2026, the day this correction ran. It is a snapshot of what this correction did, not a current total, and later corrections have moved these figures again: see the note below where one has.

One row of the projection missed, and it stays visible. Four of the five figures landed exactly where this page said they would. The fifth did not: we said 2,906 employers would keep a funding record, and the answer came out at 2,937. Those 31 are not records the withdrawal missed. They are employers that arrived between the projection and the run. A historical backfill and a night of collection added 32 funding records worth $3.55bn while this correction sat in the queue. We leave the projection above rather than replacing it, because a corrections page that quietly revises its own numbers is doing the thing it exists to prevent.

Keeping the last figure filed, not the first and not the sum. An amendment restates an offering's running total, so the last filing for an offering is the whole raise and every earlier one is that same money again. Across the 66 offerings this touches, the last figure is also the largest in 65 of them. In the one exception the filer revised its own total downwards, and its latest answer is still the one we should be showing.

What each rule costs in real records, measured rather than assumed. Withdrawing on the word "Indefinite" alone would take 138 more records worth $1.70bn, including a $200m round that opened this quarter. An uncapped offering is only a running total once it has been running. So the rule also requires the first sale to be more than a year before the filing. Matching duplicate offerings on the company rather than on the offering's file number would delete Fluidstack's genuinely separate $730m. And the takeover rule is the one with a cost we cannot measure away: 115 of the 177 filings answer yes and then explain nothing. Among the 62 that do explain, 7 turn out to be cash raises. If the silent ones behave the same way, roughly a dozen real raises go with them.

A rule we considered and rejected. Eleven of these filings say the money paid a sales commission. That is what a company pays a broker to sell securities, and a merger does not need one. Using that to rescue a record would have kept five filings that state in words that the shares were merger consideration. Four of those five are bank mergers where the fee is the adviser's. The rule rescued fewer records than it wrongly kept, so we do not use it.

What we cannot promise about the records that remain. Two things. The quarterly data set that carries these fields only appears once a quarter has ended. So this pass does not cover 9 records filed this month, worth $0.09bn, and we check those when that data set appears. A filing can answer yes to the takeover question and then explain nothing. We cannot tell that filing apart from a cash placement that happens to fund a deal, so some real raises go with the rest. A survivor would look like a funding record for a company that a buyer was taking over rather than one raising money. Or it would look like a figure covering years of an evergreen fund's sales. If you see one, we would like to be told.

28 July 2026 3,005 records signal_direction, talent_readthrough applied 29 Jul 2026

Form D records said "Hiring up" on filings that disclose no hiring

Every record drawn from SEC Form D filings carried the badge "Hiring up". A Form D reports money raised in a private placement. It states an amount and it states nothing at all about headcount, so the badge was our claim and not the filing's.

Those records also carried a read-through asserting that "capital raised is spent on headcount within the following two to six quarters". That sentence appears in no filing. It was a generalisation printed identically on thousands of records, presented as though it had been read off the document.

We corrected 3,005 records on 29 July 2026. The badge is now "Headcount Not Stated". Each read-through says only what its filing records: who raised how much, when, and the address on the filing, followed by the gap named plainly. For example: "The filing records the money only; it names no roles and no hiring plan."

28 July 2026 998 records withdrawn applied 29 Jul 2026

We listed entities that employ nobody as employers, and they were inflating our money totals

998 published records were not companies raising money to hire. They were single-asset property vehicles, insurance separate accounts and synthetic guaranteed investment contracts, and we published all of them as startup funding.

A large share of Form D filings come from entities that exist to hold an asset rather than to employ anyone. Examples: a limited company formed to buy one building, a numbered series vehicle, a non-traded credit fund. We published them as employers raising money, where they were useless to a recruiter or a job seeker. Because each raise is large, they distorted every money total on the tracker.

Insurance products were the same failure in a different form. A life insurer files a Form D for each variable life or annuity product it sells, and the "amount sold" is premium collected from policyholders, not capital the company raised. The largest single record on the tracker was one of these, at $7.4bn.

We withdrew all 998 on 29 July 2026. We deleted nothing. A withdrawn record keeps its row and carries the reason we withdrew it, so we can still count it here.

What we projected, and what actually happened
	Before	We projected	Measured now
Funding records	4,024	3,026	3,064
Money raised	$199.7bn	$114.1bn	$124.0bn
New York	$59.04bn	$8.44bn	$8.44bn across 294 records
Real estate	$13.16bn across 875 records	$1.00bn across 1	$1.16bn across 4 records

Measured on the live tracker 29 July 2026, the day this correction ran. It is a snapshot of what this correction did, not a current total, and later corrections have moved these figures again: see the note below where one has.

The money total has since fallen from $124.0bn to $101.4bn, and we corrected nothing downwards to make that happen. The figure above is what the live tracker held on 29 July 2026, the day this correction ran, and we leave it as measured. Two later passes moved it. We re-issued every stale company_key, which merged eleven employers we had counted as more than eleven. We also found five funding amounts off by a factor of a million, because the multiplier parser read only English. Two of those five were Danish kroner sitting in a US dollar column. This page promises we leave amounts in other currencies out rather than converting them at a rate nobody published, so we removed those dollar figures rather than restating them. The current total is on the dashboard, computed on request. This page does not restate it, because a corrections log that keeps rewriting its own history is not a log.

We published a projection of $114.1bn and the result was $124.0bn. Before this ran, this page said the money total would land near $114.1bn. It landed at $124.0bn, about $10bn higher, and that difference is not the correction falling short. $9.25bn of it is ten records added by a new national-press collector whose first run happened between the projection and the correction, including a single $8.6bn semiconductor raise. The same run is why the country count moved at the same time. The remaining $0.9bn is twenty-one records from other collectors. Those collectors were already live, but the copy of the database we computed the projection against was missing their records. We are leaving the projection visible above rather than replacing it, because a corrections page that quietly revises its own numbers is doing the thing it exists to prevent.

What we can and cannot promise about the records that remain. The withdrawal reached every record it could reach, and all 998 landed. We checked afterwards. No remaining record on the tracker matches any of the vehicle name patterns. The "Hiring up" badge now appears on 43 records across the whole tracker, rather than 4,018 records. What we cannot promise is that no inflated record survives anywhere, because a correction can only reach the records the pipeline holds. A survivor would look like a funding record whose employer is a numbered or single-address entity. It could also be an insurance separate account, or a name that reads like one building rather than one company. If you see one, we would like to be told.

A cost worth stating. We exclude Form D filings in the real-estate industry group outright, because the overwhelming majority of them are single-asset vehicles. This does drop a small number of genuine real-estate employers along with them, and the dataset offers no field that separates the two. We think carrying billions in vehicles that employ nobody is the worse of the two errors, but it is a real cost and not a free one.

We checked the fix rather than assuming it worked. A first pass at these exclusions left the four largest records on the tracker still wrong. We had written the rule from the spelled-out phrase "guaranteed investment contract", and the filings use the trade's abbreviations instead. The first two are "Synthetic GICs issued to insurance carriers of BOLI/COLI policies" at $4.21bn and "Synthetic GICs issued to IRC Section 529 plans" at $3.23bn. The other two are "Allocated Units of Precious Metals" at $2.51bn and "AGL Institutional Life" at $0.59bn. Seven filings and $12.4bn in all, found by reading the money list after the fix instead of trusting it. The 998 include them.

Spotted something wrong? Every record links to the document it came from, so you can check it. Write to the contact page and we will correct it here. Where this data comes from · Back to the tracker