Four Investors, Four Verdicts on the LNG Shock. First We Had to Fix Our Own Numbers.

Every equity this desk has rated has been the same trade. Micron, CoreWeave, Super Micro, Vertiv, Nvidia. Five names, one idea, plus a single Solana token. A committee that only ever looks at one sector is not a committee with a method, it is a committee with a habit.

So on August 31 we pointed the four seats at Cheniere Energy, ticker LNG, the largest LNG exporter in the United States. First energy name. First equity outside the AI buildout.

The fact sheet was wrong. We found it before the vote, not after, and only because the number was checkable against a primary source.

The bug

Our stock-sheet tool reads company financials from SEC XBRL company facts. It does not use estimates or summaries. It reads what the company filed.

For Cheniere it reported net income for the last full year as $6.79B, labelled as coming from the NetIncomeLoss tag. Trailing twelve months, $4.38B. From those it computed a price to earnings ratio of 18.5x, a return on equity of 70.8 percent, and a net margin of 20.0 percent.

Cheniere's own second quarter release says something different. In its words, "For the twelve months ended June 30, 2026, Cheniere generated net income of approximately $2.9 billion."

Our sheet was carrying an earnings figure 50 percent above the one the company reports for the same period.

Here is what happened, and it is worth reading slowly because it will affect anyone building the same kind of tool.

Two different SEC filings tag the same concept, NetIncomeLoss, for the same fiscal year 2025, with two different values:

Neither is an error. A proxy statement's pay versus performance table uses a consolidated basis. The 10-K income statement reports the figure attributable to the parent. The gap between them, $1.464B, is the share of profit belonging to holders of Cheniere Energy Partners, the affiliate Cheniere does not wholly own. Add it back and $5.330B becomes $6.794B exactly.

Our tool gathered every candidate row, sorted them, and took the last one. The sort key ran on end date first and value second, so when two rows shared an end date the larger number won. The proxy figure beat the audited one every time.

Then the trailing twelve month calculation made it worse. It took the annual figure from the proxy, on a consolidated basis, and subtracted quarterly stubs from the 10-Q filings, which are on an attributable basis:

$6.794B, consolidated, plus a negative $434M half year, attributable, minus a $1.979B half year, attributable, equals $4.381B.

That is not consolidated profit, which is $4.46B. It is not profit attributable to shareholders, which is $2.92B. It is a subtraction performed across two different definitions of the word earnings, and the result does not describe anything.

Market capitalisation counts only what Cheniere shareholders own. Dividing it by a profit figure that includes a minority stake in a separate entity overstates earnings per dollar of market value. Corrected:

A name that looked like it traded at 18.5x actually traded at 27.7x. That is not a rounding difference. It is the difference between cheap and not cheap, and it would have gone to four investors as a fact.

Did this reach anything we published

That was the first question, and it had to be answered before anything else.

The bug is silent unless a company has material minority interests. Where a company owns its subsidiaries outright, the proxy figure and the 10-K figure are the same number, and taking the larger of two identical values changes nothing.

We re-ran all six names already on the record against the fixed tool.

Nvidia and Vertiv both read from a proxy statement, exactly as Cheniere did. Their numbers did not move, because their proxy figure and their 10-K figure agree.

So no published rating was affected. The bug has been present in every equity session this desk has run, and it never fired once, because we had only ever rated companies that own themselves outright.

That is not a clean bill of health. It is a landmine that had not been stepped on. Partnership structures, midstream energy, utilities and real estate are full of companies with minority interests, and those are exactly the sectors we were moving into. The first name we picked outside the AI trade set it off immediately.

The tool now refuses any row that did not come from a 10-K, 10-Q, 20-F or 40-F. Proxy statements are no longer read for financial line items.

The setup

With a corrected sheet, here is what the four seats were given.

Qatar supplied close to a fifth of the world's LNG last year. Since two vessels were attacked in July, it has largely stopped shipping through the Strait of Hormuz, and QatarEnergy has extended force majeure on deliveries to European and Asian buyers, running into October for Pakistan, beyond September for Bangladesh and into early November for Italy's Edison. Its Ras Laffan facility was attacked and shut in March and is still running at reduced rates.

Asian spot LNG reached $23.388 per mmbtu on August 28, roughly double pre-war levels, and averaged about $21 through August against $12 in August last year. US natural gas, the feedstock Cheniere buys, sits at $2.895.

The structural detail is the one worth keeping. Crude traffic through Hormuz has recovered to its highest level in months. LNG has not, and cannot, because the world has about 850 LNG tankers against more than 7,000 crude carriers. Oil can reroute. LNG has nowhere near enough specialised hulls to do the same. That is why the two markets separated, and it is not a sentiment problem that resolves when the headlines calm down.

The demand side is already responding. China's August LNG imports are estimated at 5.2 million tons, 18 percent below the same month last year, with price sensitive industrial users cutting consumption.

On the company itself, from its August 6 filing, Cheniere raised full year 2026 guidance: adjusted EBITDA from $7.25B to $7.75B up to $7.90B to $8.40B, and distributable cash flow from $4.75B to $5.25B up to $5.30B to $5.80B. Second quarter revenue rose 24 percent, with 184 cargoes exported, up 19 percent.

And one field on the sheet said not available. Cheniere sells most of its volume on long term, fixed fee, take or pay contracts. A fixed fee contract does not capture a spot price spike. How much uncontracted volume the company holds, and how its contracts split the margin, could not be sourced to a primary. So the sheet said so, in those words, and left it blank.

That blank turned out to be the most important line in the document.

The verdicts

Four seats. Four different verdicts. Watch, Avoid, Hold, Buy.

Our eligibility rule requires three of four investors at Buy or better before a name can enter the basket. One Buy. The rule is not met, so there is no position and there is no admission.

What every seat did with the blank field

The four seats disagreed on the verdict. They did not disagree on the reason.

Every single one of them found the not available on the contract split, and every single one refused to let the war headline carry the vote.

Investor A, on quality: "The current headlines, Qatar force majeure, Hormuz stalled, Asian spot above $23/mmbtu, are loud and largely beside the point for a long term owner... So the spike is a possible short term kicker, not a thesis."

Investor B, on margin of safety, would not produce a valuation at all: "The corresponding price is not available because normalized owner earnings are not available."

Investor C, on macro: "I will not assume the $23 print flows to earnings."

Investor D, the only Buy, built its case on physical capacity rather than price, describing growth as "driven by physical capacity additions... rather than commodity price speculation."

Four models from four vendors, given the single biggest energy headline of the year, and not one of them would trade it. What stopped them was not caution. It was a field that admitted it did not know something.

The same thing happened in our first token session, when one seat refused to price JUP because the unlock schedule could not be sourced. A sheet that says not available does more work than a sheet that fills the gap with a plausible number, because the honest blank is the thing an investor argues with.

Each seat also named where it would change its mind, and all three of the non-buyers named a price well below the current one. Investor A wanted the low $200s. Investor C wanted $220 to $235. Investor D, holding the only Buy, still wanted $230 to $255 rather than the current level.

The technical layer

This runs after the vote and has no say in it. It never sees the fact sheet, and the committee never sees it. It exists to describe where a position would be mechanically wrong, not whether to take one.

At a close of $289.38, Cheniere sits 90 percent of the way up its 52 week range, 21.2 percent above its 200 day moving average, with an RSI of 67.0. It has risen 27.0 percent in a quarter.

Two observations from the computed sheet, neither of which is a forecast.

Participation is falling as price rises. Volume over the last 20 sessions is 20.2 percent below the 60 session average, while the stock trades near the top of its yearly range.

The nearest level the market actually turned at is $249.61, the July 28 swing low, 13.7 percent below the close. Ordinary daily noise for this name is 3.67 percent at 1.5x its average true range. So a stop placed at the nearest structural level sits almost four times the daily noise band away, which is a sizing constraint rather than a view.

Worth noting: every entry zone the seats named sits below that swing low. The committee's own conditions require a price the market has not traded at since July.

What would change our minds

On the Buy case. Disclosure quantifying uncontracted volume and the margin split in Cheniere's contracts. This single number is what separates a toll road from a commodity cyclical, and all four seats named it. If a materially larger share of volume touches spot than the seats assumed, Investor A said plainly that its own reasoning breaks: "My comfort rests on an assumption I have labelled not available."

On the Avoid case. Free cash flow durably above the $2.79B trailing figure after maintenance and growth capital, or visible debt reduction against the $22.94B net debt position.

On the whole frame. A negotiated settlement. Iran's president said on August 31 that Iran seeks a negotiated end to the war. If Hormuz reopens and the Qatari force majeure unwinds, the supply shock unwinds with it. Three of the four seats named this as their primary risk, including the one that voted Buy.

The part we would rather not have written

We publish the committee's mistakes as a matter of policy. This is the second time in ten days we have had to publish our own.

On August 21 we found that our fact sheet was comparing one quarter's revenue against a previous quarter's cash flow. On August 31 we found it reading earnings out of a proxy statement.

Both bugs share a shape. In each case the tool had a rule for picking between sources, the rule was reasonable, and it silently picked wrong in a case nobody had tested. Neither produced an obviously broken number. A price to earnings ratio of 18.5x looks exactly like a price to earnings ratio of 27.7x. Nothing about the output announces the problem.

The only reason either was caught is that every figure on our sheets carries the source and the url it came from, so a number can be checked against the filing it claims to come from. That convention costs something to maintain and it is the entire reason this article exists rather than a rating built on a wrong multiple.

We would rather show you the fix than the polish.