A red flag in a set of accounts is a divergence between two figures that should have moved together. Revenue grew by a quarter and receivables grew by more than half; profit rose and operating cash flow fell; inventory accumulated while sales were flat. Each of those observations is arithmetic. None of them is a finding.
That distinction is the whole discipline of this piece, so it is stated before anything else. A flag is a question to put to the person who prepared the accounts, not a conclusion about the company and never an accusation of fraud. Every flag below has innocent explanations, several of them common, and an analyst who treats a divergence as a verdict has stopped analysing and started guessing. What the flags do is locate the five or six places in a filing where an answer is owed, so that the answers can be tested and the accounts either understood or set aside.
The piece sets out the form a flag takes, twelve of them written as questions with the arithmetic that raises each one and the shape of an innocent answer, the cross-checks worked on a constructed two-year statement set so the arithmetic can be followed, how to read the answer rather than the flag, the limit of the whole exercise, and the measured cost of running it.
The form of a flag
Every flag has three parts and none of them can be dropped.
The observation is a divergence, stated as two numbers and the period they cover. Not “receivables look high” but “receivables grew 59.9 per cent while revenue grew 24.0 per cent.” A divergence without numbers is an impression, and impressions are not testable by the person answering.
The arithmetic is the measure that makes the divergence comparable across companies and periods. Days sales outstanding turns the receivables balance into a number a reader can hold: 45.0 days in the first year, 58.0 in the second, an increase of 13.0 days. The measure is a convention, not a truth, and which convention was used belongs in the note beside it.
The question is the one an answer can close. “Why has days sales outstanding risen by 13 days?” has an answer; “is revenue overstated?” does not, which is why the second form produces defensiveness rather than information. A well-formed question names the figure, the movement, and the period, and asks for a cause.
An answer closes the flag in one of three ways. It explains the movement by something documented — a contract with longer terms, a large customer with a milestone payment, an acquisition mid-year — and the documentation is produced. It explains the movement arithmetically while leaving the substance open, which happens and is worth recording as an open item. Or it does not explain it, and the flag stays open, which is a state the pack should be able to represent rather than resolve by assertion.
Twelve flags, as questions
The list below is not exhaustive and is not a fraud-detection procedure. It is the set of cross-checks that most often locates the part of a filing where an answer is owed.
Table 1. Twelve red flags written as questions, the arithmetic that raises each one, the question to put to the preparer, and the shape of an innocent answer. Source: constructed for this piece; the fourth column is a description of the kind of explanation that closes a flag, not an assertion that any particular company has one.
| Observation | The arithmetic | The question to ask | What an innocent answer looks like |
|---|---|---|---|
| Revenue grows faster than receivables can absorb | Days sales outstanding rising while growth accelerates | Which customers account for the increase in days, and on what terms | A named customer on documented longer terms, or a single large invoice dated near the year end that was paid since |
| Profit rises while operating cash flow falls | Cash conversion below one, and falling | What moved in working capital this year that did not move in the prior year | A stock build ahead of a known order book, or a payables unwind, both visible in the balance sheet |
| Earnings diverge from cash over several years | Accrual ratio climbing across periods | Which accrual is growing, and when does it reverse | A genuine growth business where receivables and inventory lead revenue, with the reversal following in the next period |
| Inventory grows faster than cost of sales | Inventory days rising | What is in the stock, and what is its ageing | A new product line or a long-lead supply contract, evidenced by the order book and the ageing report |
| Capital expenditure sits below depreciation for years | Capex as a percentage of depreciation well below one | What assets are being consumed and not replaced | A business converting to a leased asset base, with the leases visible in debt |
| Depreciation falls while revenue rises | Depreciation as a percentage of revenue dropping sharply | What changed in useful lives or capitalisation | A documented revision of useful lives, disclosed in the accounting policies |
| Revenue from a customer the filing does not name | Concentration rising, disclosure not matching | Who the largest customers are, and are any of them related to the owners | A distributor or a large corporate buyer named in the notes on request, with ordinary terms |
| Payables stretch sharply | Days payable outstanding rising | Whether supplier payment terms changed, or a supply-chain finance facility was added | A disclosed reverse-factoring programme, or renegotiated terms with named suppliers |
| Exceptional items appear in every period | The “one-off” line recurring across three years | Which of these costs recur, and why are they treated as exceptional | A restructuring programme spanning several years, with the plan and its cost disclosed |
| Tax paid is well below the statutory charge | Effective tax rate against the statutory rate | What is the reconciling item, and is it recurring | Losses brought forward, credits or a rate difference by jurisdiction, each reconciled in the tax note |
| Debt is classified as long-term with covenants close | Covenant headroom against the required level | What is the headroom, and what happens if it is breached | Headroom stated by the lender, with the covenant definition in the facility agreement |
| Owner’s remuneration and personal costs run through the company | Costs that do not scale with the business | Which costs are the owners’ and which are the business’s | A documented schedule of adjustments, agreed with the owner and applied consistently |
Two habits make that list useful rather than alarming.
The first is that flags cluster. A single rising days-sales figure is a question about one customer; rising days with falling cash conversion, rising inventory days and a recurring exceptional line is a set of observations that describe one thing, and the thing is worth naming once. The cross-checks are worth running as a set precisely because their agreement with each other is the information.
The second is that a flag is a question about a mechanism, and mechanisms leave documents. Longer terms leave a contract. A stock build leaves an order book. A supply-chain finance facility leaves an agreement with a bank. An analyst who asks for the mechanism rather than the motive gets an answer that can be checked, and the check is the work.
The cross-checks, worked
The two-year set below is constructed for the arithmetic and describes no company. It is small enough to recompute by hand, which is how the ratios should be tested before they are trusted.
Table 2. A constructed two-year statement set and the cross-checks computed from it. Amounts in constructed currency units; days are computed on a 365-day year. Source: constructed for this piece.
| Measure | Year 1 | Year 2 | Movement |
|---|---|---|---|
| Revenue | 1,000.00 | 1,240.00 | +24.0 per cent |
| Receivables | 123.29 | 197.10 | +59.9 per cent |
| Days sales outstanding — receivables ÷ revenue × 365 | 45.0 | 58.0 | +13.0 days |
| Cost of sales | 700.00 | 880.00 | +25.7 per cent |
| Inventory | 150.00 | 214.00 | +42.7 per cent |
| Inventory days — inventory ÷ cost of sales × 365 | 78.2 | 88.8 | +10.5 days |
| Net income | 120.00 | ||
| Operating cash flow | 62.00 | ||
| Cash conversion — operating cash flow ÷ net income | 51.7 per cent | ||
| Average total assets | 910.00 | ||
| Accrual ratio — (net income − operating cash flow) ÷ average total assets | 6.37 per cent | ||
| Capital expenditure | 60.00 | ||
| Depreciation and amortisation | 45.00 | ||
| Free cash flow — operating cash flow less capital expenditure | 2.00 |
The observations come out of that table in a particular order, and the order matters more than any one of them.
Revenue grew 24.0 per cent and receivables grew 59.9 per cent, so days sales outstanding rose from 45.0 to 58.0. Held at 45.0 days, receivables at the second year’s revenue would have been 152.88, and the balance is 197.10, so 44.22 of the receivables balance is new. The revenue increase was 240.00, of which 73.81 — 30.8 per cent — sits in receivables rather than cash.
Inventory days rose from 78.2 to 88.8 while cost of sales grew 25.7 per cent, so the stock grew faster than the activity that consumes it. At the second year’s cost of sales of 880.00, the extra 10.55 days is 25.43 of additional stock, held against a business whose revenue grew by a quarter.
Cash conversion of 51.7 per cent means 58.00 of the 120.00 of net income did not arrive as operating cash. The accrual ratio of 6.37 per cent is the size of that gap against the asset base, and a positive figure of that size is not by itself alarming: growth that sits in receivables and stock produces one. It is the direction and the combination that matter, not the level.
Capital expenditure of 60.00 against depreciation of 45.00 is a ratio of 1.33, which is a business investing more than it consumes and is consistent with growth. Free cash flow of 2.00 is what remains of 62.00 of operating cash after that investment, and it is the figure that would have to fund debt service, tax already paid and dividends.
Put together, the observations describe one mechanism: growth is being funded by the balance sheet. Receivables rose 73.81 and inventory 64.00, so working capital absorbed 137.81 against an operating cash flow of 62.00. The decomposition says where the absorption came from: 44.22 of the receivables increase and 25.43 of the inventory increase — 69.65 between them — are attributable to the two day-counts moving, and the remaining 68.16 to growth at last year’s collection and stock-holding efficiency. Whether that mechanism is a problem depends entirely on the answers to the questions: a company winning larger customers on longer terms and stocking ahead of a contracted order book is doing something ordinary and expensive, and a company that is not is doing something else.
Reading the answer
The answer is where the analysis actually happens, and it is worth having a method for it rather than an instinct.
Ask whether the explanation names a document. “Payment terms lengthened with a large customer” is a claim that has a contract behind it. “These are timing differences” is a claim with nothing behind it, and it is the most common non-answer in the exercise.
Ask whether the explanation is testable from the accounts the company already has. A stock build ahead of an order book can be tested against next period’s revenue and inventory. A payables unwind can be tested against the next period’s cash flow. A reverse-factoring programme can be read in the notes if it is disclosed and cannot be read at all if it is not, and the second is information.
Ask what the explanation would mean if it were true. If the receivables growth is one large customer on extended terms, then the concentration is now a risk in its own right and the flag has been exchanged for a different one. If the inventory build is a new product line, then the margin assumption in the forecast rests on a product with no history. The answers do not remove the questions; they convert them into the next set.
Record what was asked and what came back. A pack that shows six observations, the questions put, the answers received and the two that remain open is a document a reader can act on. A pack that shows six observations with a conclusion attached to each is a document that has taken a position the evidence does not support.
Where the flags stop
The limitation is worth stating as plainly as the principle at the top. No arithmetic in this piece identifies fraud, and no combination of these measures does either.
Three reasons.
Every flag has an innocent explanation, and the innocent ones are more common. Growing businesses routinely absorb cash into receivables and inventory; that is what growth costs. Exceptional items recur because restructurings take years. Tax rates differ from statutory rates for reasons a tax note explains in a table. An analyst who reads a divergence as evidence has mistaken a base rate for a finding, and the base rate is that the accounts are what they say they are.
The measures are conventions. Days sales outstanding can be computed on revenue, on credit sales, or on an average balance; inventory days on cost of sales or on revenue; the accrual ratio on total assets, on net operating assets or on cash flow. Each choice moves the number, and two analysts computing the same flag from the same accounts can disagree by several days without either being wrong. The convention belongs in the note, and a flag whose size is smaller than the convention’s range is not a flag.
The absence of flags is not assurance. These cross-checks read a set of accounts against itself. An account set that is internally consistent and wrong will pass all twelve, because the divergences they look for are divergences between reported figures and the figures are what a prepared set controls. What the checks do is make a filing reviewable in an afternoon and name the places where a document is owed. What they cannot do is verify a document that does not exist.
What a read costs
The window into a filing is a chat endpoint, and the arithmetic behind these flags fits in a handful of calls.
The published compute is the cross_verify operation at $0.006125 on Base: four of them — market capitalisation, return on equity, the earnings bridge and the per-share bridge — are $0.02450. That price is published rather than measured, because the analyst services are pending review and return no 402 quote today.
The reading and the questions are measured. On Base, deepseek-v4-flash quoted $0.00700 for a short prompt at max_tokens: 1000 on 31 October 2026 and $0.00955 for a prompt carrying a filing; Qwen quoted $0.00875 at both lengths. Three years of statements are long-prompt work, and Qwen’s flat rate is 8.4 per cent below DeepSeek’s long-prompt figure there; the fourteen questions and the follow-ups are short work, where DeepSeek is 20 per cent cheaper.
One illustrative read: three long calls to extract and normalise three years of statements, at Qwen’s $0.00875, is $0.02625; fourteen short calls, one per flag, to run each check against the numbers and draft the question, is $0.09800; six short calls for the follow-ups on the answers that did not close, is $0.04200. Twenty-three calls, $0.16625, or $0.00723 a call.
Three comparisons bound it. Reading the statements on DeepSeek’s long-prompt rate takes the read to $0.16865, 1.4 per cent more, which is the smallest of the three model premiums in this series because only three of the twenty-three calls carry a document. At the metered ceiling of $0.00955 the whole read costs $0.21965, 32.1 per cent above the measured total. At the $0.00875 that circulates for this service it costs $0.20125, 21.1 per cent more than measured. The 0.50 USDC opening balance is 71 calls at $0.00700, so a read of twenty-three calls leaves 48 in the balance — two reads and most of a third.
The decision
One practice makes a flag review useful, and it is a matter of what gets written down.
Publish the open items with the questions attached, and reconcile the reported figure against the computed one rather than choosing between them. A pack that says the reported days sales outstanding is 58.0, the computed figure from the schedule is 61.2 because of a year-end invoice, and the question put to the preparer was answered with a contract reference, has done the work. A pack that says receivables look high has not. Where the two figures disagree and the difference is unexplained, that is the item to carry forward, named, with the conclusion it would change — not a reason to select the friendlier number and move on.
Where this goes
- The pillar: x402.review/2026/10/31/finance-and-markets/ — three engines, four jobs, one price
- The desk: studio.x402.press/roles/finance_desk — three views on one brief and a disagreement table that leaves the difference standing
- The role: studio.x402.press/roles/financial_analyst — the agent that runs the statements and the cross-verification
- The service it buys: x402.press/services/deepseek/ — the canonical price, per model and per chain
- The second model: x402.press/services/qwen/ — the flat rate that does not step with prompt length
- The review: x402.review/2026/10/31/deepseek-review/ — the measured handshake and what it does not promise
- The cost context: x402.review/2026/10/31/inference-endpoints-compared/ — the two endpoints against each other
Method note
The model prices were read from the live 402 challenge on Base on 31 October 2026 at 08:50 UTC, outside DeepSeek’s weekday peak windows of 01:00–04:00 and 06:00–10:00 UTC, with one short prompt and one of about 3,000 tokens at max_tokens: 1000; Qwen was read the same way and did not step between the two lengths. The cross_verify operation at $0.006125 is the published catalogue price for the analyst services, which are pending review and return no 402 quote, so it is published rather than measured. Call counts are planning assumptions stated in the cost section. The two-year statement set, every figure in Table 2 and the ratios computed from it are constructed for this piece and describe no company; the same is true of the answer described in the closing decision, which is constructed to show the form of a good one. Days are computed on a 365-day year, the accrual ratio on average total assets, and the ratios are rounded to one decimal place with the accrual ratio to two, so a recomputation from the underlying figures may differ in the last digit. Nothing in this piece identifies, or is intended to identify, fraud in any company, and no observation here is a conclusion about any set of accounts. Nothing here is investment, legal, tax or accounting advice.
x402.review is written and edited by N.P. Vincent. The publication is connected to x402.press, which sells the services reviewed here; that interest is disclosed on every piece that touches a service, and it is stated here rather than left to inference.
Filed under Jobs · Tags: financial-statements, red-flags, due-diligence, accounting
