Caterpillar CAT

EARNINGS BANNER — RESULTS RELEASED THIS MORNING

Caterpillar issued its second-quarter 2026 results on 2026-08-04, the date of this dive, via an 8-K filed the same day. The figures are in the data used throughout: revenue $20.543B against a $19.335B consensus (+6.2%) and adjusted EPS $8.17 against $6.22 (+31.4%), with GAAP net income of $3.593B, up 64.9% year on year. The same 8-K furnished supplemental retail-sales-to-end-user data as Exhibit 99.2, which was not extracted into this archive.

Two consequences a reader must hold in mind. First, the analyst estimate block in this file predates the print — the FY2026 consensus of $24.783 is now demonstrably low against first-half actual adjusted EPS of $13.71 plus a $6.50 third-quarter consensus. Every forward multiple below is therefore struck on stale-high estimates of the multiple, i.e. the stock is cheaper on revised numbers than the table shows, and that is treated as an explicit variant perception in Section 6c rather than smoothed away. Second, the price used here is the closing print after the market had all day to react, which is the fairest available anchor.

Reference table

1. What the business is, and how the disclosure actually works

Caterpillar was founded in 1925, operated as Caterpillar Tractor Co. until 1986, and is now headquartered in Irving, Texas with 118,000 full-time employees and Joseph E. Creed as chief executive. NYSE-listed since 1929; CUSIP 149123101.

The segment structure was reorganised in 2025 and the vendor has not kept up

From the 10-Q filed 2026-05-06: the reportable segments are Power & Energy, Construction Industries, Resource Industries and Financial Products, plus an All Other Segment that is "included as a reconciling item between reportable segments and consolidated external reporting." The filing notes two structural facts that matter for reading any historical series:

• "Power & Energy" is the renamed successor to what the vendor's pre-2025 rows call "Energy and Transportation."
• Locomotives moved segments: "Responsibility for business strategy, product design, product management and development, manufacturing, marketing and sales and product support for diesel-electric, hybrid and battery-electric locomotives and components and other rail-related products and services, including remanufacturing and leasing moved from Power & Energy to Resource Industries."

The vendor's FY2025 seg_prod row is broken and is REJECTED in full. It contains six lines summing to $110.978B against reported FY2025 revenue of $67.589B — 1.64x — including a line labelled "Reportable Subsegments: $73.955B" which is plainly a rollup, and a line labelled "Power & Energy: −$5.058B", i.e. a negative figure for the company's largest and fastest-growing segment. No construction of these six numbers reproduces reported revenue and we do not pretend one does.

The FY2024 row, by contrast, ties exactly — $25.455B + $28.854B + $12.389B + $0.425B + $4.053B − $6.367B = $64.809B, matching reported revenue to the dollar. We use FY2024 for the historical mix and the 10-Q's prose figures for the current mix.

March-quarter 2026 segment detail, taken directly from the 10-Q's narrative (which states figures in text and is therefore extractable where the tables are not):

Segment sales include intersegment transfers, which is why they sum $1.670B above consolidated revenue — that gap is intersegment elimination plus the All Other Segment, and it is the same structure the FY2024 vendor row reflects with its −$6.367B elimination line.

Three things in that table are the whole story of this company right now.

First, Power & Energy has become co-equal with Construction Industries in size — $7.031B against $7.161B — and the filing names the driver without ambiguity: "Power Generation – Sales increased in large reciprocating engines and in turbines and turbine-related services, primarily data center applications."

Second, Construction Industries' 38% quarter is partly channel fill and the filing says so. "The increase in sales was mainly due to higher sales volume of $1.5 billion and favorable price realization of $356 million. Higher sales volume was primarily driven by the impact from changes in dealer inventories. Dealer inventory increased during the first quarter of 2026, compared with a slight decrease during the first quarter of 2025." Every one of the four regional bullets under that segment repeats "changes in dealer inventories" as the volume driver. Dealer inventory builds reverse.

Third, tariffs are visibly destroying Resource Industries' margin. Sales rose 4% and segment profit fell 39%, from $623M to $378M, taking margin from 17.0% to 10.0%. The filing: "The decrease was mainly due to unfavorable manufacturing costs. Unfavorable manufacturing costs largely reflected the impact of higher tariff costs."

Geography — clean, and one of the better vendor tables in this batch

seg_geo sums exactly to reported revenue in every year shown. FY2025: $36.609B + $12.793B + $11.199B + $6.988B = $67.589B, matching to the dollar.

North America is 54% of revenue and rising, which matters two ways: it is where the datacenter build is concentrated, and it is where the Infrastructure Investment and Jobs Act spend the 10-Q cites is landing. It is also why a $2.2-2.4 billion tariff bill on imported components is so damaging — the demand is domestic, the supply chain is not.

2. The quarter reported this morning

Read the gross-margin column and the tariff story writes itself. Margin fell from 37.3% (Q2 2024) through 33.0% (Q2 2025) to a trough of 28.6% in the December 2025 quarter — the quarter in which the first full year of tariffs landed — and has since recovered to 35.1% and 37.8%, the highest in the file. Price realization is doing the work: the 10-Q reports favourable price realization of $426 million in the March quarter alone, and describes it as positive in every one of the three primary segments.

Four consecutive beats after one miss, and the beats are accelerating: +9.3%, +9.6%, +19.1%, +31.4%. That is not a company creeping past a low bar; it is a company whose earnings are running away from the analysts modelling them.

Trailing twelve months: revenue $74.729B (a record, against the previous full-year peak of $67.589B), GAAP net income $10.833B, GAAP diluted EPS $23.22, adjusted EPS $23.82. The GAAP-to-adjusted gap is only 2.6%, because Caterpillar's sole non-GAAP adjustment is restructuring — the 10-Q states its non-GAAP measures exclude "(i) restructuring income/costs" and nothing else. That is unusually clean and means the choice of basis does not change any conclusion here.

The backlog — the single most important number in this dive

> "At the end of the first quarter of 2026, the dollar amount of backlog believed to be firm was approximately $62.7 billion, about $11.5 billion higher than the fourth quarter of 2025. The order backlog increased across the three primary segments, with the largest increase in Power & Energy. The backlog for large reciprocating engines and turbine products continues to grow within Power & Energy. Of the total backlog at March 31, 2026, approximately $24.8 billion was not expected to be filled in the following twelve months."

Three observations.

First, $62.7 billion against trailing revenue of $74.729 billion is roughly ten months of sales already contracted. For a company whose historical problem was zero visibility past the next dealer order, that is a structural change in the quality of the business.

Second, $11.5 billion of net backlog growth in one quarter is larger than the entire Resource Industries segment's annual revenue. It is also larger than the sequential revenue increase, which means orders are being placed faster than they can be filled.

Third, and this is the risk rather than the reward: $24.8 billion — 40% of the backlog — is not expected to be delivered within twelve months. Long-dated backlog is visibility, and it is also duration risk. Datacenter power orders placed in 2026 for 2028 delivery are cancellable in a way that a bulldozer order for next quarter is not.

3. Tariffs — the largest identifiable drag, quantified by the company

The 10-Q is unusually specific and we quote it in full because the numbers are material:

> "Based on tariffs implemented since the beginning of 2025 and in place over the course of 2026, we expect tariff costs to be around $2.2 billion to $2.4 billion in 2026. We remain confident that we will manage the impact of tariffs over time."

> "We expect tariff costs to be around $700 million in the second quarter of 2026. We expect about 50 percent of the tariff costs to be incurred in Construction Industries, and about 25 percent of tariff costs to be incurred in both Power & Energy and Resource Industries."

Against trailing net income of $10.833 billion, a $2.2-2.4 billion pre-tax tariff bill is roughly 15-17% of pre-tax profit. It is visible in the segment margins: Resource Industries down 700 basis points year on year, Power & Energy down 170 basis points despite 22% sales growth. Only Construction Industries expanded margin, and it did so on price realization and dealer-inventory volume.

And there is a $1.0 billion option the company has explicitly declined to book:

> "On February 20, 2026, the U.S. Supreme Court ruled that tariffs imposed under the International Emergency Economic Powers Act ("IEEPA") on goods imported into the United States were unauthorized. As of March 31, 2026, total IEEPA tariff costs were approximately $1.0 billion. The ruling did not address potential refunds, and therefore the ultimate availability, timing and amount of any potential refunds of these tariffs is highly uncertain. Based on the current facts and circumstances, we have determined that recovery of any funds is not probable."

This is the correct accounting treatment and it creates a genuine, unpriced asymmetry. Approximately $1.0 billion — about 9% of trailing net income — has been paid, ruled unlawful, and written off. No figure in this dive assumes any of it comes back. If it does, it is a one-off gain and should be treated as one, not capitalised into a multiple.

4. Balance sheet and the enterprise-value rebuild — with one correction we deliberately do NOT make

As at 2025-12-31 (the vendor's newest annual balance sheet — note this is two quarters stale relative to the June quarter reported today, and the June balance sheet is not in this file):

The correction this batch usually makes — netting short-term investments — does not apply here, and stating why is more useful than applying it reflexively. Short-term investments are zero. The $16.414B of long-term investments is not surplus cash: for Caterpillar this is the Financial Products and insurance investment portfolio, matched against insurance reserves and finance obligations, not a liquidity buffer. We deliberately do NOT net it, and we say so.

The correction that DOES apply is larger and structural: totalDebt of $43.330B is not industrial debt. Caterpillar consolidates Cat Financial, a captive finance company whose borrowings fund a portfolio of customer finance receivables and leases. The 10-K confirms Cat Financial is separately covenanted:

> "Cat Financial has also agreed under certain of these agreements not to exceed a certain leverage ratio (consolidated debt to consolidated net worth...), to maintain a minimum interest coverage ratio... and not to terminate, amend or modify its support agreement with us."

A finance company's debt is its inventory. Treating it as parent-company leverage produces a materially wrong picture: debtToEquityRatioTTM of 2.33x, financialLeverageRatioTTM of 5.29x and netDebtToEBITDATTM of 2.29x all describe a consolidated entity that does not exist as an economic unit. The 10-Q presents supplemental consolidating data separating Machinery, Power & Energy from Financial Products — but that data is presented in numeric tables, which are stripped from this extraction. We therefore CANNOT compute an industrial-only net debt figure, and we state that limitation rather than estimate around it.

The consolidated enterprise-value rebuild, presented with that caveat attached:

> Market cap $403.642B (460,644,000 shares × $876.27)

> + total debt $43.330B (consolidated, including Cat Financial)

> − cash $9.980B

> = consolidated enterprise value ≈ $436.99B

The vendor reports $442.075B, a $5.08B (1.2%) difference consistent with its using a more recent debt figure than the 2025-12-31 balance sheet in this file. Neither figure is an industrial enterprise value and neither should be used as one.

Trailing EBITDA is recomputed for the same reason as elsewhere in this batch. ratios_ttm.ebitMarginTTM of 20.10% implies trailing EBIT of $15.024B, which is $1.932B above the sum of the last four quarters' operating income ($13.092B). The gap reconstructs exactly: trailing pre-tax income of $13.930B plus trailing interest expense of $1.093B equals $15.023B. The vendor's "EBIT" is pre-tax income grossed up for interest and therefore includes Financial Products' other income. We use operating income of $13.092B plus estimated trailing D&A of ≈$2.35B = $15.44B.

That D&A figure required an estimate and we flag why. The vendor's inc_q reports depreciationAndAmortization of $0.0 for the June 2026 quarter — a plain data defect, since D&A was $595M, $598M and $570M in the three preceding quarters. We annualise the three available quarters to ≈$2.35B, cross-checked against FY2025 D&A of $2.262B from the cash-flow statement. The figure is labelled as derived.

What the balance sheet actually says, on the parts we can verify. Equity of $21.318B on $98.585B of assets is a 21.6% equity ratio, which for a company consolidating a finance arm is unremarkable. interestCoverageRatioTTM of 12.0x is comfortable. returnOnInvestedCapitalTTM of 13.2% and returnOnCapitalEmployedTTM of 20.8% are strong industrial returns. returnOnEquityTTM of 54.1% and priceToBookRatioTTM of 20.80x are artefacts of a small book — Caterpillar has bought back $22.1 billion of stock in four years — and are rejected as lenses, which is also why the vendor's composite rating scores it 1 out of 5 on price-to-book and 1 on debt-to-equity, dragging the grade to B− / 3.

Working capital is the real balance-sheet risk and it is large. daysOfInventoryOutstandingTTM is 152.4, daysOfSalesOutstandingTTM 117.4, and the cash conversion cycle is 193.6 days — more than six months of working capital tied up. Inventory rose from $16.827B to $18.135B in FY2025 and the March quarter saw dealer inventories build further. In an order-book expansion that is normal; in a backlog plateau it becomes a charge.

5. Cash flow and capital returns

Note the capital-expenditure line: $1.465B in FY2025 against $3.215B in FY2024 — a 54% cut. That is the single most surprising number in the cash-flow statement, and it deserves a flag rather than a smoothing. capexToRevenueTTM is 0.59%, which for a manufacturer of mining trucks and gas turbines is implausibly low as a run-rate. We cannot explain it from the extractable filings and we do not try — it may reflect a reclassification between capital expenditure and equipment leased to others, which Caterpillar reports separately. The consequence is that FY2025 free cash flow of $10.274B and the resulting freeCashFlowYieldTTM of 3.25% may be flattered, and no conclusion in this dive rests on the free-cash-flow figure alone.

Share count is falling steadily: 548.6M diluted (FY2020) → 469.0M (FY2025) → 462.5M in the June 2026 quarter. That is a 15.7% reduction in six years, roughly 2.5% a year, and it is a meaningful part of per-share growth.

Dividend — and a labelling correction from the filing. The vendor reports lastDividend: 6.16 and dividendPerShareTTM: 6.16, a 0.70% yield. That is the trailing-twelve-month total. The 8-K filed 2026-06-11 gives the actual current rate:

> "On June 10, 2026, the Board of Directors declared a quarterly cash dividend of one dollar and sixty-three cents ($1.63) per share of Company common stock, payable on August 19, 2026... The per share dividend of $1.63 is an increase of twelve cents from the previous quarterly dividend of $1.51 per share."

Forward run-rate is therefore $6.52, a 0.74% forward yield, against the vendor's $6.16 / 0.70%. The increase is 7.9%. Payout ratio on trailing GAAP EPS of $23.22 is 28.1% — the vendor's dividendPayoutRatioTTM of 26.0% corroborates on its own basis.

Total shareholder yield is approximately 2.0% — 0.74% dividend plus roughly 1.3% from a $5.19B annualised buyback pace. This is not an income story and should not be sold as one. Caterpillar's return comes from earnings growth, not from distribution.

6. Valuation — priced in or room?

At $876.27 (market cap $403.642B, ~460.6M shares, consolidated EV ≈$436.99B, trailing revenue $74.729B, trailing GAAP EPS $23.22):

Estimate coverage is good on the near years — 17 and 18 analysts on FY2026 and FY2027 EPS — and thin beyond: 6 on FY2028 and 4 on FY2029. The FY2029 row shows revenue rising 1.4% and EPS 2.9% and is used for nothing.

Critically: these estimates PREDATE the print released this morning. First-half 2026 actual adjusted EPS is $5.54 + $8.17 = $13.71. The third-quarter consensus is $6.50. That gives $20.21 through September, requiring only $4.57 in the December quarter to reach the FY2026 consensus of $24.783 — against $5.16 actually delivered in December 2025 and a rising trend. The FY2026 consensus is very likely 5-10% too low, and FY2027 with it. We handle this explicitly rather than adjusting silently: the multiples above are computed on the stated consensus, and Section 6c treats the revision as a variant perception.

The est EBITDA and EBIT rows are tested and NOT arithmetically impossible. ebitAvg is below ebitdaAvg in every forward year and neither is negative. However they are clearly low against actuals: FY2026 ebitAvg of $17.744B against a trailing operating income run-rate of $13.092B implies a plausible-looking step, but FY2025 ebitdaAvg of $16.592B against actual FY2025 operating income of $11.210B plus D&A of $2.262B ($13.472B) is 23% too high. We run all forward valuation on epsAvg per the data contract.

Peer context. The vendor peer set contains real comparables — Deere ($166.6B), PACCAR ($71.6B), AGCO ($7.5B), Hyster-Yale ($0.6B), Astec ($1.2B), REV Group ($3.1B) — and several that are not (Eaton, GE Aerospace, Parker-Hannifin, RTX). Caterpillar at $403.6B is 2.4x the size of Deere and larger than every other peer combined. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. What can be said, and is informative, is that GE Aerospace at $391.5B is in the same batch and is the closest analogue for "industrial re-rated on a structural demand shift" — that comparison is drawn qualitatively in the portfolio note.

6a. What today's price assumes (the inversion)

At $876.27 — 28.7x FY2027 consensus and 37.7x trailing GAAP — the price embeds the following falsifiable claims:

• Revenue reaches $77.008B in 2026 (+13.9%) and $85.231B in 2027 (+10.7%). (Consensus-derived; 13 analysts each, and PRE-print.) First-half 2026 actual revenue was $37.958B, so the second half must deliver $39.05B against a third-quarter consensus of $19.684B — which itself implies a 4.2% SEQUENTIAL DECLINE from the June quarter's $20.543B. The street is modelling a plateau starting immediately, and the June quarter beat its own number by 6.2%.
• Adjusted EPS reaches $30.580 in 2027, i.e. 23.4% growth on a 32.8% growth year. (Consensus; 18 analysts.) Caterpillar's FY2021-FY2025 EPS compound rate was roughly 18% and included a −18% year in FY2025. The price assumes two consecutive years of 20%-plus EPS growth from the most cyclical industrial in the index.
• The $62.7 billion backlog converts at current margins. (Filing-stated backlog; our assumption on conversion.) Approximately $24.8B of it is beyond twelve months. This is the single most important assumption and the one with the most evidence behind it.
• Tariff costs of $2.2-2.4 billion in 2026 are absorbed by price realization and then roll off. (Company-guided cost; our assumption on the offset — and management framing is half-weighted per house rule.) The evidence so far supports it: gross margin recovered from 28.6% in December 2025 to 37.8% in June 2026 while the tariff bill ran. But Resource Industries' margin is still 700 basis points below where it was.
• The market keeps paying roughly 28x forward for a business with a 1.605 beta. (Our number.) This is the most fragile assumption in the price. Caterpillar has historically traded in the mid-teens on forward earnings. At 20x FY2027E the stock is $612; at 25x it is $765; at 32x it is $979. The multiple, not the earnings, is the widest source of variance, and the multiple is at an all-time high for this security.

Where our own signal has a reading: the knowledge base's most specific claim on this name (Visser, 2026-05-03) states directly that "its 36x forward multiple will stay elevated, not mean-revert." That is an explicit bet on the fragile assumption above, made by the dominant voice in the lane, and it is worth flagging that our library's view and our own most-fragile-assumption flag are pointed at exactly the same number.

6b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+23.4%, FY2026E $24.783 to FY2027E $30.580) + multiple drift (COMPRESSION, from 28.7x to roughly 24x on the respective forward year, −16%) + shareholder yield (+2.0%) ≈ +7% to +9%, against a base fair value 0.4% above spot.

Our base assumes meaningful multiple compression and we should be explicit about why it is not a rolldown artefact. Today the market pays 28.7x on the next fiscal year. In twelve months, paying 24x on the then-next year would be a genuine 16% de-rating. We assume it for two reasons. First, consensus growth decelerates on its own path: EPS +32.8%, +23.4%, +20.5%, then +2.9%. A multiple appropriate to 23% growth is not appropriate to 3%. Second, and more important for a cyclical, a backlog that grew $11.5 billion in a quarter cannot keep growing $11.5 billion a quarter. The moment backlog growth flattens, the market re-prices Caterpillar from a visibility story back toward a cycle story, and the cycle multiple for this company is in the teens.

Almost all of the expected return is earnings growth, and the multiple is fighting it. If the multiple simply HELD at 28.7x on FY2028E of $36.844 the price would be $1,058 (+20.7%); if it compressed to 20x, $737 (−15.9%). The spread between those outcomes is far wider than for any other name in this batch, which is what a 1.605 beta means in practice.

6c. Variant perception (where we differ, what would surprise)

• We differ from the street on the ESTIMATES, not on the business — and this is our clearest variant perception in the batch. The consensus figures in this file were struck before this morning's 31.4% EPS beat. First-half adjusted EPS of $13.71 plus a $6.50 third-quarter consensus leaves only $4.57 needed in the December quarter to hit the full-year $24.783, against $5.16 delivered a year earlier on lower revenue. We think FY2026 lands nearer $26 and FY2027 nearer $33, which would put the stock at roughly 26.6x rather than 28.7x and move our base fair value from $880 to approximately $950. This is an arithmetic observation about stale data, not a forecast, and it is labelled as such. It is also the single strongest argument for the upside case and it is why the verdict is Hold rather than Watch.
• We differ on the composition of Construction Industries' growth. The consensus framing is "38% growth in construction." The filing says the volume came "primarily" from dealer inventory builds, repeated in all four regional bullets. Dealer inventory is a loan from the future. Watchable number: the retail-sales-to-end-user data Caterpillar furnishes as Exhibit 99.2 with every earnings 8-K — which is exactly the disclosure that would settle this, and which is NOT extracted into this archive. We flag that as a real gap in our evidence rather than reasoning past it.
• We differ on the tariff refund, mildly. The company has written off approximately $1.0 billion of Supreme-Court-invalidated IEEPA tariffs as "not probable" of recovery. That is conservative and correct accounting, and it means roughly 9% of a year's net income sits as unrecognised optionality that no estimate in this file contains. If it reverses it is a one-off and must not be capitalised — but it is real money and the price does not include it.
• We do NOT differ on the AI-power thesis, and that is worth saying plainly. Our knowledge base articulated it early and specifically, the filings confirm it, and the market has priced it. A correctly-identified, correctly-priced thesis produces no excess return. The edge our library had on this name was captured between June 2025 and today, during which the stock rose 104%.
• Positive surprise that would force a further re-rate: backlog growing again at the September quarter, particularly if Power & Energy leads it a third consecutive time; FY2027 revenue guidance above roughly $88 billion; a tariff resolution or an IEEPA refund becoming probable; or Resource Industries' margin recovering toward 15% as mining capital expenditure turns.
• Negative surprise that would break the thesis: a flat or declining backlog print — this is the primary falsifier and it is checkable in three months; the retail-sales-to-end-user data showing dealer inventories rising while end demand is flat; Power & Energy segment margin falling below 20%; a September quarter below the $19.684B consensus, which would confirm the sequential-decline model the street is already using; or any large order cancellation from the $24.8B of beyond-twelve-month backlog. Any of these hits a stock at 37.7x trailing earnings with a 1.605 beta and a falling 50-day average.

Synthos fair values

All three anchors are multiples of the FY2027 consensus adjusted EPS distribution (mean $30.580, low $27.388, high $34.471, 18 analysts — the deepest coverage in the file), cross-checked against FY2028E of $36.844 (6 analysts, used only as a check).

• Bear ~$550 — 20.1x the FY2027 consensus LOW of $27.388. Cross-check: 14.9x FY2028E; 25.7% below the 200-day moving average of $741.34 and 34.9% above the 52-week low of $407.79. The scenario: the backlog plateaus, dealer inventories unwind, mining stays weak, tariffs do not roll off, and the market re-prices Caterpillar as the cyclical it has always been — at the mid-teens forward multiple it has always carried. −37.2%.
• Base ~$880 — 28.8x the FY2027 consensus MEAN of $30.580. Cross-check 1: 23.9x FY2028E of $36.844 gives $880 — the two anchors agree exactly. Cross-check 2: on our own view that FY2027 is revised to roughly $33, this is 26.7x, which is the same price at a lower multiple. Sensitivity, stated openly: 25x gives $765; 32x gives $979. The entire answer lives in a 25-32x band, which is wide, and we do not claim more precision than that on a 1.605-beta cyclical. +0.4%.
• Bull ~$1,170 — 34.0x the FY2027 consensus HIGH of $34.471. Cross-check: 31.8x FY2028E; 9.9% above the 52-week high of $1,064.90 and 3.9% below the street's target high of $1,218. The scenario: backlog keeps compounding, power-generation revenue heads toward the three-times-by-2030 path the knowledge base cites, tariffs resolve with a refund, mining turns, and the market accepts that Caterpillar's earnings are now structurally less cyclical than its history. +33.5%.

Base is 0.4% above spot; asymmetry roughly 0.90:1 — near-symmetric and very wide (37.2% down against 33.5% up). That width is the honest description of this security: a 1.605-beta industrial at an all-time-high multiple with an all-time-high order book. The reason it is a Hold rather than a Watch is the stale-estimate observation in 6c; the reason it is not a Buy is that our base equals the price and the stock just failed to hold an 11% gap on record numbers.

7. Knowledge base — 33 hits, every text match a collision, and one claim the filing confirms

Raw hits: 33. Entity matches: 17. Text matches: 16. Used as name-level conviction: 15. Discarded: 18. Search terms: CAT, Caterpillar, word-boundary matched across entity lists and claim text over all 51,928 distilled claims.

Discard class 1 — the animal and the bounce (14 claims). Every single text match is a collision and this is the cleanest example of ticker-string failure in the batch. The three-letter ticker "CAT" matches "cat", and financial commentary is saturated with the phrase "dead cat bounce." Representative discards:

• "Sentience (self-model + inner world) is distinct from general intelligence; a cat is sentient." (joscha_bach)
• "Adversarial images expose that ML isn't human-like: a cat is a thin string in million-dimensional space." (lex_fridman)
• "Machines lack the common sense of a house cat." (lex_fridman)
• "Treat the economy like a cat, not a washing machine." (nassim_taleb)
• "To preserve society roughly unchanged, we shouldn't build anything smarter than a cat." (theories_of_everything)
• "The relief rally / dead cat bounce failed today." (Erik Townsend, 2020)
• "Bounce is a pathetic dead-cat bounce." (Patrick Ceresna via darius_dale, 2022)
• "the recent bounce is a dead-cat bounce; the dollar is weakening." (jordi_visser, 2025 — a DOLLAR claim)
• "Housing bounce is a dead-cat." (forward_guidance, 2025)
• "labor uptick is a dead-cat bounce." (forward_guidance, 2026)
• "expect a dead-cat bounce, not a real software recovery." (jordi_visser, 2026 — an EQUITY ROTATION claim)
• "recent strength partly as an oversold-software dead-cat bounce." (luke_gromen, 2026)
• "cat's out of the bag on what to own then." (empire) and "cat's out of the bag" on compute demand (ml_street_talk)

Discard class 2 — a different company (1 claim). 2026-03-16, mikkel_rosenvold: "Leaning toward Red Cat Holdings as a war-supply-chain trade idea." That is RCAT, a drone company. Discarded.

Discard class 3 — a customer mention, not a view (1 claim). 2026-07-09, crux_investor, speaker Darren Bowen, speaker_role: management of a gold miner: "at $4,000 gold it easily funds the Nicaragua build, with a Caterpillar equipment lease now approved for extra flexibility." This is a Caterpillar customer describing a lease. It assigns no view to Caterpillar and it is management voice of a third party. Discarded.

Discard class 4 — a stale sector claim (1 claim). 2021-10-06, luke_gromen, on Evergrande's real-economy transmission to global industrials. Five years old and superseded.

Discard class 5 — a superseded price target (1 claim). 2023-01-30, warren_pies: "Would add to CAT position around $250 with the 50-day near $240." Retained in the record for honesty but assigned zero weight: the stock is $876.

The 15 claims that carry weight

The two that the filings independently confirm — these are the most valuable items in this entire batch's knowledge-base work:

> 2025-09-21 · bullish · conviction 75 · entities: CAT · thinker: jordi_visser · skill 2.0

> "Caterpillar is purely an AI company now — power-generation revenue is 14.5% of sales growing 28% YoY, with datacenter generators outpacing construction machinery for the first time."

> 2026-05-03 · bullish · conviction 65 · entities: CAT · thinker: jordi_visser · skill 2.0

> "Caterpillar has a record $62B backlog and expects power-gen sales ~3x larger by 2030 from AI data centers; its 36x forward multiple will stay elevated, not mean-revert."

> Test: "Caterpillar power-generation sales roughly 3x larger by 2030"

The 10-Q filed 2026-05-06 — three days after the second claim — states firm backlog of $62.7 billion. The claim said $62B. It is exact. And the March-quarter segment table shows Power & Energy at $7.031B against Construction Industries' $7.161B — the two segments now within 1.8% of each other in size, which is the quantitative form of "outpacing construction machinery for the first time." A dated, specific, checkable claim in our library was confirmed by the primary source. That is what the knowledge base is supposed to do and it is worth recording when it happens.

The rest of the Visser sequence — five further claims, escalating, no reversal:

> 2025-06-22 · bullish 55 — "Caterpillar (generators) and Generac benefit from AI electricity buildout."

> 2025-07-20 · bullish 92 — "Pounding the table on these megacap energy/industrial 'power' names [XOM, CVX, FCX, CAT, DE, PH] — they've underperformed and trade very cheap two years out." The highest-conviction claim in this lane at 92, and CAT is up 104% over the twelve months since.

> 2025-09-14 · bullish 68 — "The AI power buildout is a macro theme spanning transformers, switchgear, power distribution — Caterpillar, Cummins, eventually Paccar."

> 2025-11-02 · bullish 64 and 55 — two near-identical same-day claims on Caterpillar's turbines, generators, compressors and autonomous mining vehicles. Collapsed to one.

Independent corroboration — 3 claims from 3 different sources, which matters more than the seventh Visser claim would:

> 2026-03-02 · bullish 60 · thinker: odd_lots — "Massive data-center buildout (Stargate) boosts old-economy industrials like Caterpillar; power is a key AI bottleneck."

> 2026-05-29 · bullish 70 · thinker: invest_like_the_best · speaker: Colin — "Caterpillar is in a super-cycle from AI data-center buildout plus rare-earth and central-bank gold-mining demand; a buy-and-hold-forever compounder he's owned since 2016, up ~600%." Note the disclosed position — an owner talking his book, and we say so.

> 2026-07-27 · bullish 55 · speaker: Dan Ives, guest on anthony_pompliano — "AI's second-third-fourth derivatives extend to names like Caterpillar and, on the defense side, Lockheed Martin." Dan Ives is a sell-side voice and is weighted accordingly; the channel is one where house policy discounts the host's own voice, but this is a guest claim and is treated as first-class subject to the speaker's own promoter-class discount.

The claim requiring a promoter disclosure:

> 2026-05-03 · bullish 75 · speaker: Josh Brown · thinker: compound_and_friends

> "Caterpillar is the quintessential Halo company — in the CAPEX boom, software embedded in manufacturing, can't be ripped out and replicated by AI."

> Elsewhere in this same knowledge base (2026-07-28) the same speaker discusses "the Halo (LOHA) ETF" with an explicit "(disclosed involvement)" note. "Halo company" is not a generic description; it is the selection criterion of a product the speaker is associated with. This claim is therefore weighted LOW with the frame stated, per house policy on promoter-class voices. The disclosure is the speaker's own and we honour it rather than treating the claim as independent analysis.

Half-weighted commercial-partner voice — 2 claims, excluded from the conviction pool:

> 2025-10-28 · bullish 85 · thinker: jensen_huang — "Every manufacturing company will run two manufacturing systems: one to build the machine, one to build the AI that powers it."

> 2026-03-27 · bullish 70 · thinker: jensen_huang — "Power and generation equipment (incl. small modular nuclear) are key downstream beneficiaries of AI datacenter buildout."

> Management voice of a company whose product creates the demand described. Quoted, half-weighted, excluded from the conviction pool.

The bear case — 1 claim, and it is specific:

> 2025-05-14 · bearish · conviction 70 · entities: Caterpillar, Tesla, Apple, BYD, Huawei · thinker: forward_guidance

> "US blue chips are losing share to Chinese national champions in China and soon globally — the Caterpillar-displacement pattern repeating with better Chinese products."

> This is the only substantive negative claim in the lane, it names the company as the archetype, and NOTHING in the extractable filings measures Chinese competitive share. We flag it as an unmeasurable risk rather than dismissing it.

And two genuinely old name-level claims, retained for the record:

> 2023-01-30 · neutral 58 · thinker: warren_pies — "Hold CAT: strong margins across segments and forward P/E below industrials despite all-time highs, but elevated expectations set up miss risk into earnings."

> The forward P/E is no longer below industrials. That is precisely what has changed, and stating the old claim makes the change visible.

The concentration, stated plainly

Jordi Visser supplies 7 of the 15 usable claims — 47% of the lane, the highest single-speaker share in this batch. He is also the highest-skill voice in it (2.0 against 1.0 for everyone else). Unlike the BAC lane, he does not reverse: the sequence runs June 2025 → July 2025 → September 2025 → November 2025 → May 2026, consistently bullish, with rising specificity and two claims that the filings later confirmed. That is a coherent tested thesis rather than noise, and it is why net conviction is positive-high. But 47% from one speaker is disclosed, not smoothed, and the independent corroboration from odd_lots, invest_like_the_best and the Huang partner claims is what makes the lane a lane rather than an opinion.

Net conviction: positive-high. Breadth after collapsing by speaker: 7. Claim count used: 15. The knowledge base was early and right on Caterpillar. The relevant question for today's reader is whether that edge has been captured, and the stock's 104.4% twelve-month return against SPY's 24.3% says it has.

8. Data integrity — what we rejected from the vendor file and why

Twelve findings. The geographic table is clean and the income statement ties to the filings; the segment table, the headline P/E and one D&A field are broken.

1. ratios_ttm.priceToEarningsRatioTTM of 56.39 CONTRADICTS the vendor's own netIncomePerShareTTM — REJECTED. At $876.27, a 56.39x multiple implies trailing EPS of $15.54. But netIncomePerShareTTM is $23.53, and the sum of the last four quarters' net income ($3.593B + $2.549B + $2.401B + $2.290B = $10.833B) divided by revenuePerShareTTM-implied shares of 460.4M gives $23.53 — the two agree. The correct trailing P/E is 37.2x on that basis, or 37.7x on our per-quarter diluted EPS sum of $23.22. The vendor's 56.39x is wrong by 51% and would make Caterpillar look far more expensive than it is. priceToEarningsDilutedRatioTTM of 56.61 fails identically. Notably, every OTHER price ratio in the file is internally consistent — priceToBookRatioTTM 20.80 against book value per share $42.124, priceToSalesRatioTTM 5.401 against revenue per share $162.313, priceToFreeCashFlowRatioTTM 30.74 against FCF per share $28.519 — so the defect is isolated to the earnings ratios.

2. The FY2025 seg_prod row does not tie to revenue and is REJECTED in full. Six lines summing to $110.978B against reported revenue of $67.589B — 1.64x — including a rollup line ("Reportable Subsegments: $73.955B") and a negative $5.058B figure labelled "Power & Energy", which is the company's largest growth segment. No reconstruction we attempted reproduces reported revenue. The row is discarded and the current segment mix is taken from the 10-Q's prose figures instead. The FY2024 row ties exactly ($64.809B) and is used for history.

3. seg_prod uses a superseded segment taxonomy before FY2025. The 10-Q confirms "Energy and Transportation" was renamed "Power & Energy" and that locomotives and rail moved from Power & Energy to Resource Industries. Pre-2025 and post-2025 segment revenue are not comparable and no growth rate in this dive is computed across that boundary.

4. inc_q.depreciationAndAmortization is $0.0 for the June 2026 quarter — a plain data error. The three preceding quarters report $595M, $598M and $570M. We annualise those to ≈$2.35B for the trailing EBITDA calculation and label the figure as derived, cross-checked against FY2025 cash-flow D&A of $2.262B.

5. Trailing EBITDA is not an operating figure — RECOMPUTED. ebitMarginTTM of 20.10% implies EBIT of $15.024B, which is $1.932B above the sum of the last four quarters' operating income ($13.092B). The gap reconstructs exactly as trailing pre-tax income ($13.930B) plus trailing interest expense ($1.093B). The vendor's EBIT includes Financial Products' other income. We use $13.092B + $2.35B = $15.44B and compute EV/EBITDA at ≈28.3x rather than 26.34x.

6. totalDebt of $43.330B and netDebt of $33.350B conflate industrial and captive-finance debt — flagged, NOT corrected, and the reason is stated. Caterpillar consolidates Cat Financial, whose borrowings fund a finance-receivables portfolio and which the 10-K confirms is separately covenanted on its own leverage ratio. debtToEquityRatioTTM (2.33x), financialLeverageRatioTTM (5.29x) and netDebtToEBITDATTM (2.29x) therefore describe a consolidated entity that is not an economic unit. The 10-Q presents supplemental consolidating data splitting Machinery, Power & Energy from Financial Products — but that data is in numeric tables, which are stripped from this extraction. We cannot compute industrial-only leverage and we do not estimate it.

7. Long-term investments of $16.414B are deliberately NOT netted against debt — the standard correction is inverted here. Short-term investments are zero, and the $16.414B of long-term investments is the Financial Products and insurance portfolio matched against reserves, not a liquidity buffer. Netting it would understate leverage by $16.4B. We state the reasoning because the batch's standard correction runs the other way.

8. bal_a is two quarters stale. The newest balance sheet is 2025-12-31; results through 2026-06-30 were released this morning. No June-quarter balance sheet is available in this file and the June 10-Q had not been filed as of this dive. All balance-sheet figures are labelled as at 2025-12-31.

9. The analyst estimate block PREDATES this morning's print — flagged as a timing defect rather than a data error. FY2026 epsAvg of $24.783 requires only $4.57 in the December quarter after first-half actuals of $13.71 and a $6.50 third-quarter consensus. The estimates are stale-low and every forward multiple in this dive is correspondingly stale-high. This is stated rather than adjusted, and the consequence is treated as a variant perception in Section 6c.

10. est.ebitdaAvg is materially high against actuals — NOT USED for valuation. FY2025 ebitdaAvg of $16.592B against actual operating income of $11.210B plus D&A of $2.262B ($13.472B) is 23% too high. The rows are not arithmetically impossible — ebitAvg is below ebitdaAvg in every year and neither is negative, so this file trips none of the three known corruption signatures — but they are wrong in a consistent direction and all forward valuation runs on epsAvg.

11. lastDividend: 6.16 is the trailing-twelve-month total — corrected from the filing. The 8-K filed 2026-06-11 states: "On June 10, 2026, the Board of Directors declared a quarterly cash dividend of one dollar and sixty-three cents ($1.63) per share... an increase of twelve cents from the previous quarterly dividend of $1.51 per share." Forward run-rate is $6.52, a 0.74% yield, not $6.16 / 0.70%.

12. quote.yearHigh/yearLow disagree with tech.hi52/lo52 — we use tech. The quote block gives $1,073.46 and $405.46; the technical block gives $1,064.90 and $407.79. Discrepancies of 0.8% and 0.6%. We use tech throughout for consistency with the moving averages and relative returns computed from the same close series.

Book-equity metrics are economically low-information here and are REJECTED as lenses. returnOnEquityTTM of 54.1%, priceToBookRatioTTM and priceToFairValueTTM of 20.80x are all arithmetically correct and all distorted by four years of buybacks totalling $22.1B against equity of $21.318B. returnOnInvestedCapitalTTM of 13.2% and returnOnCapitalEmployedTTM of 20.8% are the honest return measures and are what we use. The vendor's composite rating of B− / 3, with 1 out of 5 on price-to-book, debt-to-equity and price-to-earnings, is driven by exactly these distortions plus the broken P/E from finding 1, and is rejected.

A capital-expenditure anomaly we flag but cannot resolve. Capex fell from $3.215B (FY2024) to $1.465B (FY2025), a 54% cut, taking capexToRevenueTTM to 0.59% — implausibly low as a run-rate for a manufacturer of mining trucks and gas turbines. It may reflect a reclassification between capital expenditure and equipment leased to others, which Caterpillar reports separately, but we cannot verify that from the extractable filings. FY2025 free cash flow of $10.274B may therefore be flattered and no conclusion in this dive rests on the free-cash-flow figure alone.

Not defects, correctly reported and independently confirmed: seg_geo summing exactly to reported revenue in every year; the quarterly revenue and net income series matching the 10-Q's stated growth rates (Q1 2026 operating profit of $3.085B, +20%, matches the filing verbatim); earn_cal actuals reconciling to the 8-K-reported quarter; the employee count and chief executive name; interestCoverageRatioTTM of 12.0x; and the share-count decline from 548.6M to 462.5M diluted.

Non-equity tripwire — checked and passed. CAT is common stock, NYSE, CUSIP 149123101, ISIN US1491231015. Note that the 8-K of 2026-08-04 lists a second registered security — "5.3% Debentures due September 15, 2035" — so a debt instrument does exist under this registrant. The file in hand is the common stock: price $876.27 (not a $25 or $1,000 par-like level), beta 1.605, a variable dividend just raised 7.9%, a 52-week range of $407.79 to $1,064.90 (a 161% span), and volume of 5.35M shares. This is common equity.

Extraction limits, stated rather than estimated around. Numeric tables are stripped, so the June-quarter balance sheet, the segment revenue tables, the ME&T-versus-Financial-Products consolidating data, the exact tariff and inventory figures beyond what the MD&A states in prose, and the Exhibit 99.2 retail-sales-to-end-user data — which would directly settle the dealer-inventory question raised in 6c — are all unavailable. Earnings-call transcripts are not available on this data plan.

9. Technicals

• Price $876.27. −17.7% from the 52-week high of $1,064.90; +114.9% above the 52-week low of $407.79. Position within the annual range: 71st percentile.
• Moving averages, split and this is the key signal: −4.7% BELOW a 50-day average of $919.44, which is falling; +18.2% above a 200-day average of $741.34, which is rising. The long-term uptrend is intact; the intermediate trend has rolled over.
• RSI 43.51 — below the midpoint, neither oversold nor strong. MACD −28.61 — negative, and in absolute terms the largest negative reading in this batch (3.3% of the share price).
• Maximum drawdown from peak over the trailing year: −17.7% — the current price is the worst point of the drawdown.
• Relative performance, and the split between horizons is the whole story: 3-month +0.17% vs SPY +7.6% and QQQ +7.7% — Caterpillar has underperformed the S&P by 7.4 points over a quarter; 6-month +26.8% vs SPY +11.1%; 12-month +104.4% vs SPY +24.3% and QQQ +30.8% — an 80-point outperformance over a year. This is a stock that did everything in the first eight months of the period and nothing in the last three.
• Sentiment: 31 buy, 21 hold, 4 sell across 56 analysts — the highest sell count of any name in this batch. Consensus target $943.33 (+7.7%), median $900, high $1,218 (+39.0%), low $769 — 12.2% BELOW spot. A genuinely dispersed distribution with real disagreement, which is more informative than the tight consensus bands elsewhere.

Today's move and what it does to the entry

This is the most important paragraph in the dive. Caterpillar reported second-quarter results before the open on 2026-08-04. It beat consensus EPS by 31.4% and revenue by 6.2%. The stock opened at $922 — an 11.1% gap above the $830.03 previous close — traded as high as $934.9994, and closed at $876.27, up 5.57% on the day but 6.3% below the day's high and only $9.42 above the day's low of $866.85, on 5.35 million shares against a 3.19 million average — 68% above normal volume.

A 31% earnings beat that opens 11% higher and closes near the low of the day, on heavy volume, is distribution. It says a large cohort of holders used the print as an exit. It does not say the business is bad — the business just produced the best quarter in its history. It says the news was owned before it was reported, which is exactly what a 104% twelve-month return and a knowledge-base lane that called this in June 2025 would predict.

And the stock is still 4.7% below a falling 50-day average and 17.7% below its high. A failed gap on record numbers, below the 50-day, with MACD at −28.6, is not a base — it is an unresolved distribution pattern. One session proves nothing, but the burden of proof has shifted.

What we are telling the reader plainly: this is not the day to establish a position. The two setups that would change that are below.

10. Insiders — eight transactions, all phantom-stock accruals, all meaningless

Every transaction in this file is an A-Award of Phantom Stock Units at a price of zero, in quantities between five and twenty-two units. These are dividend-equivalent accruals on deferred compensation balances — when Caterpillar pays its quarterly dividend, phantom-stock holders' balances are credited with additional units of equivalent value. The 2026-06-26 and 2026-07-24 clustering is consistent with the $1.63 dividend declared 2026-06-10.

What the file contains is therefore: zero open-market purchases, zero open-market sales, and zero discretionary transactions of any kind. This is the emptiest insider table in the batch and it carries no signal whatsoever. We report it as such rather than constructing a reading from eighteen phantom units.

Worth noting for context: chief executive Joseph E. Creed holds only 11,483 phantom stock units and chief financial officer Kyle Epley 5,961 — small balances that suggest recent appointments to these roles. The file contains no common-stock holdings for either officer, which is a genuine information gap rather than a finding: Form 4 filings in this archive cover only phantom stock, and total beneficial ownership is not extractable.

11. Verdict, kill-criteria and flip conditions

Hold.

Caterpillar just produced the best quarter in its hundred-year history and the stock closed near the low of the day. Both of those facts are true and the verdict has to hold them together.

What is genuinely excellent, and all of it is verified: second-quarter revenue of $20.543B, up 24.0%; GAAP net income of $3.593B, up 64.9%; operating margin of 20.9% against 17.6%; adjusted EPS of $8.17 against a $6.22 consensus, the fourth consecutive beat and the largest at 31.4%; firm order backlog of $62.7 billion at 2026-03-31, up $11.5 billion in a single quarter, with the largest increase in Power & Energy and approximately $24.8 billion beyond twelve months; Power & Energy sales up 22% with the filing naming "data center applications" directly; Construction Industries margin expanding to 21.4%; gross margin recovering from a 28.6% tariff trough to 37.8%; return on invested capital of 13.2%; a share count down 15.7% in six years; and a knowledge-base lane whose most specific claim — "a record $62B backlog" — was confirmed to within $0.7 billion by the primary filing three days later.

What we are declining to pay for: 37.7x trailing GAAP earnings and 28.7x FY2027 consensus on a security with a 1.605 beta that has historically traded in the mid-teens; $2.2-2.4 billion of guided tariff cost in 2026 with Resource Industries' margin already down 700 basis points and roughly $1.0 billion of Supreme-Court-invalidated tariffs written off as "not probable" of recovery; Construction Industries' 38% growth attributed by the filing "primarily to the impact from changes in dealer inventories"; a base fair value of $880 that is 0.4% above spot and 6.7% below the street's $943.33; the highest sell-rating count in this batch (4 of 56); and, most immediately, a stock that gapped 11.1% on a 31% beat and closed 6.3% below the day's high, below a falling 50-day average, with MACD at −28.6.

The distinction that matters. We are not saying Caterpillar's AI-power thesis is wrong. We are saying our library identified it in June 2025, the stock has risen 104% since, the filings now confirm it, and a correctly-identified, correctly-priced thesis produces no further excess return. The one thing that keeps this from being a Watch is the stale-estimate arithmetic in 6c: consensus predates today's beat and is demonstrably low, which means the stock is cheaper on revised numbers than the table shows. A realised thesis with a fair value at spot and a probable upward estimate revision is a Hold — own it, and let the revision or a better price make the next decision.

Pre-registered KILL criteria — what would take this to Avoid:

• A flat or declining backlog print at the September quarter. This is the primary falsifier and it is checkable on 2026-11-04. $62.7B growing $11.5B a quarter is the entire visibility argument; a plateau converts Caterpillar back into a cyclical at 37.7x trailing earnings.
• The retail-sales-to-end-user data (Exhibit 99.2) showing dealer inventories rising while end-user sales are flat, which would confirm that Construction Industries' 38% quarter was channel fill.
• Power & Energy segment margin below 20.0%, against 20.6% in the March quarter and 22.3% a year earlier. A growth segment with compressing margins is a price-taker.
• A September quarter below the $19.684B revenue consensus, which would validate the sequential-decline path the street is already modelling.
• Tariff costs materially above the guided $2.2-2.4 billion, or evidence that price realization can no longer offset them — the December 2025 quarter's 28.6% gross margin is the precedent for what that looks like.
• A break below the 200-day moving average of $741.34 on company-specific news rather than market beta.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $760 — roughly 25x FY2027E as stated, or 23x on our revised estimate, and close to the 200-day average. At $760 the base case is +16% and the risk/reward turns genuinely asymmetric. This is 13.3% below today's close and inside the range this stock has traversed in the last four months.
• The FY2027 consensus revising above roughly $33 in the weeks after this print, which is the arithmetic 6c argues for and which would move our base to approximately $950 at an unchanged multiple. This is the single most likely path to an upgrade and it should resolve within a month.
• A backlog print above $65 billion at the September quarter, particularly with Power & Energy leading a third consecutive time. That converts an order surge into a trend.
• The IEEPA tariff refund becoming probable, releasing roughly $1.0 billion — about 9% of a year's net income — with a further $2.2-2.4 billion annual drag potentially rolling off behind it.
• A reclaim of the 50-day moving average at $919.44 on volume, converting today's failed gap into a base — though on its own this would be a technical flip only and would need one of the fundamental legs with it.

Where CAT fits in the Synthos Framework Portfolio. The AI-power and electrification sleeve, industrial expression. On batch overlap, which matters acutely here: CAT and GE are in this same batch and are two expressions of the same trade — both are hundred-year-old industrials being re-rated because compute needs power, both have record backlogs, and both have most of their upside behind them. They are not diversification from each other. Within the pair, Caterpillar is the more cyclical and the more tariff-exposed; it also has the larger non-power business, which is a hedge in a datacenter slowdown and a drag in a datacenter boom. PLTR, in the same batch, is the same theme at the opposite end of the risk curve. Logged as a tracked Synthos call (Hold) as of 2026-08-04 at $876.27.

Single biggest risk: that the backlog is the order surge, not the run-rate. $62.7 billion of firm orders, up $11.5 billion in one quarter, with 40% of it beyond twelve months, is the most bullish fact about this company and structurally the most fragile. Extraordinary order books are placed once. Datacenter operators securing prime-power capacity ahead of grid interconnection are making a one-time procurement decision with a multi-year delivery tail; once the capacity is contracted, the incremental order rate reverts to something much closer to replacement. At 37.7x trailing earnings with a 1.605 beta, the first flat backlog print is a de-rating event, and there is nothing in the valuation to absorb it.

Most fragile assumption in the price: that the market keeps paying roughly 28x forward earnings for a business with a 1.605 beta and a mining cycle attached. Every other assumption on this page concerns volumes, margins or costs, and each is observable quarter by quarter. This one is about willingness to pay, and it has already travelled from the mid-teens — where Caterpillar has spent most of its listed life — to an all-time high. The knowledge base's own most specific claim on this name bets directly on it: "its 36x forward multiple will stay elevated, not mean-revert." We do not have a view strong enough to contradict that, and we do not have one strong enough to underwrite it either. When the fragile leg of the thesis is the multiple and our fair value equals the price, the honest answer is to hold and let the next backlog print decide.

Provenance & disclosures

• Traceability: 33 raw knowledge-base hits on CAT and Caterpillar across 51,928 distilled claims (17 entity matches, 16 text matches). 15 used as name-level conviction; 18 discarded. ALL SIXTEEN text matches are collisions: twelve on the animal or the phrase "dead cat bounce" (speakers include Lex Fridman, Joscha Bach, Nassim Taleb, Erik Townsend, Patrick Ceresna, Luke Gromen, and — notably — Jordi Visser twice, on the DOLLAR and on an equity rotation, neither about Caterpillar), one on Red Cat Holdings (RCAT), one on "cat's out of the bag" ×2, and one on a gold miner's Caterpillar equipment lease (management voice of a third party). Two entity matches were also discarded — a five-year-old Evergrande sector claim and a superseded 2023 price target of $250. Of the 15 retained, Jordi Visser supplies 7 — 47% of the lane, the highest single-speaker concentration in this batch — and he does NOT reverse, running consistently bullish from 2025-06-22 through 2026-05-03 with rising specificity. Two of his claims are independently CONFIRMED by the primary filings: the 2026-05-03 claim of "a record $62B backlog" against the 10-Q's $62.7 billion, and the 2025-09-21 claim that datacenter generators were "outpacing construction machinery for the first time" against a March quarter in which Power & Energy ($7.031B) and Construction Industries ($7.161B) are within 1.8% of each other. One claim carries a PROMOTER-CLASS disclosure and is weighted low with the frame stated: Josh Brown's 2026-05-03 "Caterpillar is the quintessential Halo company" is the selection criterion of an ETF (LOHA) in which the same speaker discloses involvement elsewhere in this knowledge base. Two Jensen Huang claims are half-weighted as commercial-partner management voice and EXCLUDED from the conviction pool. One further claim (invest_like_the_best, 2026-05-29) carries a disclosed long position held since 2016. Independent corroboration comes from odd_lots (2026-03-02) and Dan Ives as guest (2026-07-27). The single substantive bear claim is forward_guidance (2025-05-14) on Chinese displacement, which nothing in the extractable filings measures. Breadth after collapsing by speaker: 7. All quotes are verbatim from the stored claim text.
• Data as-of: annual fundamentals to 2025-12-31; quarterly fundamentals through 2026-06-30, released via 8-K on the morning of this dive; most recent filing-verified financials 2026-03-31, from the 10-Q filed 2026-05-06 · estimates 2026-08-04 but struck BEFORE the 2026-08-04 print — see Section 8, finding 9 · prices 2026-08-04, quote timestamp 1785873599 = 2026-08-04T19:59:59Z ($876.27, +5.57%; open $922; high $934.9994; low $866.85; 50-DMA $919.44; 200-DMA $741.34; RSI 43.5; MACD −28.61) · knowledge-base claims 2026-08-04. All figures come from the Synthos vendor data file for CAT or from the SEC filings in the CAT archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-13 (fiscal 2025); 10-Q filed 2025-11-03 (September 2025 quarter); 10-Q filed 2026-05-06 (March 2026 quarter — the primary filing source for this dive); 8-K filed 2026-04-30 (first-quarter results); 8-K filed 2026-06-11 (annual meeting results and the $1.63 quarterly dividend declaration); 8-K filed 2026-08-04 (second-quarter results — the date of this dive). The 2026-08-04 and 2026-04-30 8-Ks are cover documents only: Exhibit 99.1 (the earnings release) and Exhibit 99.2 (supplemental retail-sales-to-end-user data for machines and power systems) were NOT extracted and contain no numbers in this archive. The Exhibit 99.2 omission is material — it is the disclosure that would settle the dealer-inventory question raised in Section 6c — and it is stated as a gap rather than reasoned around. Numeric tables are stripped from all extracted filings, so the June-quarter balance sheet, the segment revenue tables, the ME&T-versus-Financial-Products consolidating data and exact inventory and tariff figures are unavailable except where the MD&A states them in prose.
• Where the filings contradicted or corrected the vendor (detailed in Section 8): the FY2025 seg_prod row sums to $110.978B against reported revenue of $67.589B (1.64x), contains a rollup line and a negative figure for Power & Energy, and is rejected in full in favour of the 10-Q's prose segment figures; priceToEarningsRatioTTM of 56.39 contradicts the vendor's own netIncomePerShareTTM of $23.53 and is rejected in favour of 37.7x; the 10-Q confirms the segment reorganisation (Energy and Transportation renamed Power & Energy; locomotives moved to Resource Industries) that makes pre-2025 segment data non-comparable; lastDividend of $6.16 is the TTM total against a $6.52 forward run-rate confirmed by the 8-K's declaration of $1.63 per quarter, up from $1.51; trailing EBITDA recomputed from $16.786B to $15.44B because the vendor's "EBIT" is pre-tax income plus interest and includes Financial Products' other income; the June-quarter D&A field of $0.0 is a plain error and is replaced with a derived ≈$2.35B; totalDebt and every leverage ratio conflate industrial and captive-finance debt and are flagged as not economically meaningful, with the industrial split explicitly NOT estimated because the consolidating tables are stripped; and quote.yearHigh/yearLow ($1,073.46/$405.46) are rejected in favour of tech.hi52/lo52 ($1,064.90/$407.79). Where vendor and filing AGREED — worth recording: seg_geo summing exactly to reported revenue in every year; the Q1 2026 operating profit of $3.085B and its stated +20% growth matching the filing verbatim; the quarterly revenue series reconciling to earn_cal actuals; the FY2024 seg_prod row tying to the dollar; and the employee count and chief executive name.
• Basis note: analyst estimates are adjusted (non-GAAP); reported diluted EPS is GAAP. For Caterpillar the gap is unusually small — trailing GAAP EPS of $23.22 against trailing adjusted EPS of $23.82, a 2.6% difference — because the 10-Q states the company's sole non-GAAP adjustment is "(i) restructuring income/costs." No conclusion in this dive turns on the choice of basis, and both trailing multiples are shown (37.7x GAAP, 36.8x adjusted).
• Estimate coverage: 13 analysts on FY2026 and FY2027 revenue; 17 and 18 on FY2026 and FY2027 EPS; 6 on FY2028 EPS and 4 on FY2029. Only FY2027 is used as a valuation anchor; FY2028 is a cross-check; FY2029 is used for nothing. All rows predate the 2026-08-04 print and are treated as stale-low, with the consequence stated explicitly rather than adjusted.
• Peer note: the vendor peer set contains genuine comparables — Deere ($166.6B), PACCAR ($71.6B), AGCO ($7.5B), Astec ($1.2B), Hyster-Yale ($0.6B), REV Group ($3.1B) — alongside names that are not (Eaton, GE Aerospace, Parker-Hannifin, RTX). Caterpillar at $403.6B is 2.4x Deere and larger than every other peer combined. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.
• Fair-value caveat: the $550 / $880 / $1,170 anchors are multiples of the FY2027 consensus adjusted EPS distribution — 20.1x the low of $27.388, 28.8x the mean of $30.580, and 34.0x the high of $34.471. Stated arithmetic, not a discounted cash flow. The base is sensitivity-disclosed: 25x gives $765; 32x gives $979. The whole answer lives in a 25-32x band, which is wide, and no greater precision is claimed on a 1.605-beta cyclical. Cross-checked against FY2028E: 23.9x $36.844 gives $880, agreeing with the base exactly. On our own view that FY2027 is revised toward $33 after today's print, the base of $880 is 26.7x and a constant 28.8x would give approximately $950 — that revision path is stated as a variant perception, not incorporated into the published base.
• Timing: second-quarter results were released on 2026-08-04, the DATE of this dive, via an 8-K filed the same day, so every operating figure here is current to the hour. The next print is 2026-11-04, 92 days away. The stock's +5.57% close carried the earnings news; the reversal from an 11.1% opening gap to a close 6.3% below the day's high, on volume 68% above average, is treated as the single most informative technical fact in this dive.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.