10-K forensic · Fraudtracker

10-K forensic analysis — following the money through the footnotes

The composite scores (Beneish, Altman, Dechow) only see the income statement and balance sheet. Every meaningful accounting story lives in the footnotes: operating leases, pension assumptions, capex-vs-opex categorization, language patterns in MD&A, and the cash-flow tie-out. This page walks through each layer on the most interesting names and flags where the disclosed numbers diverge from the underlying economics.

1. Operating leases — what's on, off, and shifted

ASC 842 (effective 2019) brought most operating leases onto the balance sheet as right-of-use (ROU) assets and lease liabilities. But the operating-vs-finance classification, the discount rate, and the "service contract" loophole still let companies move billions in real obligations across reporting lines. Watch the gap between ROU asset and lease liability, and whether undisclosed long-term commitments (cloud compute, take-or-pay agreements) show up only in the future-cash-payments table.

Tesla (TSLA)

Gigafactories + Supercharger sites + retail stores

Item	FY24	FY23	Note
Operating ROU asset	$3.2B	$2.8B	Real estate, equipment
Operating lease liability	$3.1B	$2.7B	Closely matches ROU
Finance lease ROU	$0.6B	$0.7B	Mostly Solar systems
Discount rate disclosed	5.2%	4.8%	Reasonable for credit
Undisclosed (purchase commitments table)	$8B+	$5B+	Nickel, lithium, raw materials

Watch: Lease balances tie out cleanly. The thing to read carefully is the Purchase Commitments note — Tesla has long-dated raw-material commitments at fixed prices that aren't lease accounting but are economically similar (locked-in future cash outflows). $8B+ of these is comparable to the entire balance-sheet lease portfolio.

Amazon (AMZN)

Largest non-government real estate footprint in North America; air freight fleet

Item	FY24	Note
Operating ROU asset	$76B	Massive (warehouses + offices)
Operating lease liability	$77B	Closely matches
Finance lease ROU	$8.5B	Aircraft leases (mostly Boeing 767F)
Future undiscounted operating payments	$232B	Through ~2050
Implied weighted lease term	12.0 years	Long

Watch: $232B in future undiscounted lease payments — more than 60% of Amazon's market cap as of FY20 was contracted into a single liability stream. ASC 842 records the present value, not the gross. Investors who anchor on "balance-sheet liability" miss the $156B gap between gross commitments and discounted carrying value. Recession-vulnerable since lease rentals are nominal-fixed.

Walgreens Boots Alliance (WBA)

Distressed retailer; lease load > market cap

Item	FY24	Note
Market cap	~$10B
Operating lease liability	$22B	2.2× market cap
Future undiscounted	$32B	10-year average tenure
Stores closing 2024–2025	~1,200	Many on long leases

Flagged: The lease portfolio is the equity. WBA's true enterprise value sits inside the lease accounting — closing a store doesn't extinguish the lease. Distressed retailer math: when lease liability > market cap and stores are closing, you are looking at a structural impairment / restructuring trigger.

Nvidia (NVDA)

Capital-light by GAAP; massive economic commitments off the lease line

Item	FY24	Note
Operating lease liability	$1.5B	Office leases only
Long-term supply / hosting commitments	$11B+	Disclosed in Note 5 of 10-K
Inventory purchase obligations	$15B+	TSMC wafer commitments

Watch: The "small lease portfolio" headline is structural — Nvidia outsources fabrication and hosts on others' clouds. But the Long-term supply commitments footnote is where the real obligations live: TSMC wafer commitments at fixed prices, hosting commitments to Coreweave/Microsoft. These aren't leases under ASC 842 but are economically similar take-or-pay structures. Read Note 5 every quarter.

2. Pension obligations — discount-rate alchemy

Defined-benefit (DB) plans report a Projected Benefit Obligation (PBO) discounted at management's choice of rate. Each 25 basis points the discount rate is raised lowers the PBO by ~3–5%. For companies with $30B+ PBOs (Boeing, GE, Lockheed, Ford, GM), 25 bps = $1B+ in reported obligation. Combined with the "expected return on plan assets" assumption that drops straight to pension expense, this is the single largest discretionary line in industrial accounting.

Boeing (BA)

$80B+ pension obligations · perennial funded-status concern

Item	FY24	FY23
Projected Benefit Obligation (PBO)	$54B	$52B
Plan assets fair value	$53B	$50B
Funded status (over/(under))	($1B)	($2B)
Discount rate (US plans)	5.2%	4.9%
Expected return on plan assets	7.5%	7.5%
OPEB obligation (separate)	$4.5B	$4.7B

Flagged: Expected return of 7.5% is above the LDI/peer median (~6.5%). Each 1pp drop in expected return raises pension expense by ~$500M annually. The funded status appears nearly neutral, but that depends on the discount rate assumption — at the rates Lockheed uses (4.7%), the PBO would be ~$57B and the funded status would be ($4B). Pension is a cyclical headwind for Boeing's reported earnings.

General Electric / GE Aerospace (GE)

Legacy industrial pension; the canonical cautionary tale

Item	FY24	Pre-spin (FY22)
PBO	$31B	$60B
Plan assets	$26B	$50B
Funded status	($5B)	($10B)
Discount rate trend (10-year)	3.4% (2020) → 5.4% (2024)

Flagged historically: 2017 — GE took a $6.2B pre-tax charge after auditors questioned long-term care reserves. PBO assumptions had been understated for years. The 2020-2024 discount-rate rise mechanically reduced PBO by ~$15B without any change in underlying obligations. Read the discount-rate sensitivity table; the assumption is doing 80% of the work.

Ford (F) and GM (GM)

Legacy auto pensions; healthcare obligations on top

Company	PBO	Plan Assets	OPEB	Total Obligation
Ford	$45B	$43B	$5B	$50B
GM	$58B	$55B	$13B	$71B

Watch: GM's OPEB ($13B) is concentrated in retiree healthcare for older UAW cohorts. Healthcare cost-trend assumptions move OPEB the same way discount rates move pensions. A 1pp ↑ in healthcare trend = $1.2B ↑ OPEB. Both companies have well-funded pensions but the OPEB obligation is structurally underfunded since plan assets aren't dedicated to OPEB.

3. Capitalization vs. expense — where the WorldCom playbook lives today

Internal-use software (ASC 350-40), R&D, leasehold improvements, and cloud-infrastructure costs all sit on a continuum where the company chooses whether to expense (hits margin today) or capitalize (smooths over future years via depreciation/amortization). Capitalization-rate creep — same activity, more of it routed to capex — is the textbook earnings-management lever.

Meta (META)

Largest capex among public tech ($35–40B AI buildout)

Item	FY24	FY22
Capex (total)	$38B	$32B
R&D expensed	$45B	$36B
Capitalized internal-use software	$3.6B	$2.1B
Capitalized SW / total R&D	8%	6%

Watch: Capitalization rate creeping up. Each 1pp ↑ in capitalized share of R&D adds ~$450M to current-year net income (smoothed over 3 years). Direct WorldCom analogue is line-cost capitalization — Meta's variant is cloud-infrastructure-software capitalization. Read the Property and Equipment note for the "internal-use software" line and watch its growth rate vs. R&D growth rate.

Tesla (TSLA)

"Tooling and equipment" capitalization; FSD R&D treatment

Item	FY24
Capex	$11B
R&D expensed	$4.5B
Capitalized tooling & equipment	~$3B
Average useful life - tooling	7 years
Average useful life - vehicles fleet	10 years

Watch: 7-year tooling useful life is at the upper end of auto-industry peers (5–7 typical). FSD development costs run through R&D; the unit-economics shift if capitalization were elected (it isn't, currently). The reason FSD deferred revenue grew to $4B is that collected FSD revenue is deferred — the costs are expensed as R&D up front, the revenue recognized only on feature delivery.

Salesforce (CRM)

Acquired-intangibles amortization is structurally permanent

Item	FY25
Capitalized purchased software (gross)	$45B
Accumulated amortization	$22B
Net intangibles	$23B
Annual amortization (in non-GAAP addback)	~$2B
Useful lives (range)	3–10 years

Flagged: Salesforce capitalizes purchased software from acquisitions (Slack, Tableau, MuleSoft, Demandware) over 3–10 years and adds the amortization back to non-GAAP earnings as "non-cash". For a serial acquirer, this means the addback is structurally permanent — every year there's "non-recurring" amortization. Multiple expansion based on the non-GAAP figure ignores ~$2B/year of real economic cost. The corollary: when Salesforce stops acquiring, the "non-GAAP advantage" shrinks.

Palantir (PLTR)

Internal-use SW capitalization is small; SBC is the real story

Item	FY24
Capitalized internal-use software	~$0
R&D expensed	$510M
SBC (in non-GAAP addback)	$760M
SBC / Revenue	~26%

Flagged: Palantir doesn't have the WorldCom-style capitalization issue (internal SW is essentially zero on the balance sheet) — it has the larger problem of SBC being the path to non-GAAP profitability. Treating $760M of equity issuance as "non-cash" for a growth multiple is the same illusion in different costume.

4. Language patterns — what evasive disclosure looks like

Every word in MD&A and footnotes is chosen by counsel and signed off by the auditor. Specific phrases have specific GAAP and SEC meanings; others are weasel constructions. Learn the markers.

Phrase	Technical meaning	What it can hide
`substantial doubt about the entity's ability to continue as a going concern`	ASC 205-40 trigger; auditor required to issue qualified opinion	This is the canary — if it appears in the 10-K, the equity is at risk.
`reasonably possible`	ASC 450 — between 10% and 50% probability	Used for loss contingencies management thinks could happen but isn't required to accrue. The accrual goes from "reasonably possible" → "probable" (50%+) only when forced.
`material weakness in internal control over financial reporting`	SOX 302/404 finding; reasonable possibility that material misstatement won't be caught	Almost always disclosed late. Look for "remediated" language — that's the signal management knows it was a real weakness.
`previously reported as / now reclassified to`	Prior-period reclassification	Often used to move expenses out of "operating costs" and into "other" or "non-operating", improving operating margin without changing underlying economics.
`subsequent events`	ASC 855 — events after balance-sheet date but before issuance	This footnote is where post-period bad news lives. Read it last; if it's longer than usual, something happened.
`variable interest entity (VIE)`	ASC 810 — entity in which the company has a controlling financial interest without majority voting	The Enron technology. If a VIE footnote is unusually long or carries large absolute numbers, read every line. Modern examples: Tesla's solar leasebacks, Chinese ADRs (VIE structure for the entire entity).
`related-party transactions`	ASC 850 — transactions with directors, officers, family, or controlled entities	Round-tripping (Luckin), self-dealing (Tyco), siphoning (Wirecard). Always read this footnote in full.
`at the discretion of management`	Plain English; suggests management has latitude	Used in revenue recognition, useful-life choices, impairment timing. Earnings-quality red flag when it appears multiple times in a single 10-K.
`the Company is not currently aware of`	Plain English; legalistic hedge	Suggests the Company has been thinking about it. The corollary phrase to `could materially adversely affect`.
`change in segment reporting`	ASC 280 reorganization	Often signals decline in a previously-segmented business; combining segments hides the deterioration. Reconstruct prior comparable periods using the original disclosure if possible.

Concrete examples in current filings

Lucid Motors (LCID) — "substantial doubt" qualifier appeared in Q2 2024 going-concern review. Saudi PIF capital injection followed.
Boeing (BA) — repeated "material weakness in internal control" disclosures around 737 MAX certification documentation 2023–2024.
Tesla (TSLA) — "Solar power purchase agreements" footnote uses VIE accounting for residential solar leasebacks; ~$3B in consolidated VIE assets.
Nvidia (NVDA) — "we may not be able to obtain or otherwise secure sufficient supply" language in risk factors, FY24 10-K. Hedging against TSMC capacity constraint.
Alibaba (BABA) — entire entity is a VIE for the listed shareholders; the operating company is in China and the listed entity holds contracts, not equity.

5. Following the money — true free cash flow vs. reported

The reported "Free Cash Flow" or "Adjusted FCF" lines remove things they shouldn't (real economic costs) and add things they shouldn't (timing benefits). Reconstruct it from raw 10-K cash-flow-statement data: True FCF = CFO − Capex − SBC market value − Acquisitions for growth. Stock-based comp is the most-ignored component — it's "non-cash" in accounting but is real shareholder dilution at market prices.

Company	CFO	Capex	Reported FCF	SBC (FV)	True FCF	Reported P/FCF	True P/FCF
Nvidia	$65B	$3B	$62B	$5B	$57B	71×	77×
Microsoft	$120B	$78B	$42B	$11B	$31B	88×	119×
Meta	$92B	$38B	$54B	$15B	$39B	23×	32×
Tesla	$15B	$11B	$4B	$2.5B	$1.5B	300×	800×
Palantir	$1.2B	$0.05B	$1.15B	$760M	$390M	170×	510×
Snowflake	$0.9B	$0.05B	$0.85B	$1.4B	($550M)	90×	N/M (negative)
Salesforce	$13B	$0.7B	$12.3B	$3.2B	$9.1B	22×	30×
Oracle	$22B	$25B	($3B)	$5B	($8B)	N/M	N/M
CoreWeave	$4B	$22B	($18B)	$0.5B	($18.5B)	N/M	N/M

Read this carefully. The "True P/FCF" column collapses the capex-heavy and SBC-heavy stocks the most. Snowflake's true FCF is negative — its "non-GAAP profitability" depends entirely on treating $1.4B of equity issuance as free. Oracle is FCF-negative now because the $300B OpenAI deal requires capex that outpaces operating cash flow. CoreWeave is structurally cash-burning — the only path to "FCF" is if revenue commitments overshoot the depreciation schedule on $20B of GPU spend, which assumes continuing demand.

How to use this page

Lease section tells you what economic obligations sit beyond the headline balance sheet — particularly relevant for retailers (WBA), capital-light tech with hidden contractual exposures (Nvidia), and serial real-estate consumers (Amazon).
Pension section is for industrials. The discount-rate assumption is the largest single discretionary number in their reporting — a 25-basis-point change moves billions.
Capex/opex section exposes capitalization-rate creep (Meta), serial-acquirer non-GAAP gaming (Salesforce), and SBC-as-currency (Palantir).
Language section is the reading list. When you see substantial doubt, material weakness, or previously reported as, slow down and read every word of the surrounding paragraph.
Cash reconciliation is the bottom line. Subtract SBC at market value and ask: is the business actually generating cash for shareholders, or is it generating reported earnings via dilution and capitalization?

All of the company-specific values on this page come from publicly disclosed 10-K and 10-Q filings as filed with the SEC. They are illustrative of the analytical method, not investment advice. The most reliable use of this kind of analysis is to short-list ~5 names that screen poorly across multiple of the five layers and read their actual filings end-to-end.