In FY2023, Super Micro Computer reported $640M of profit while operating cash flow ran $640M negative. Twelve months later its auditor resigned. This is what an accrual-anomaly screen saw in the filings — and, just as importantly, what it didn't flag.
Scored by the production engine on Super Micro's own annual filings — nothing hindsight-fitted, no retraining after the fact:
Combined AAAS scores, 0–100, five risk tiers. Gate-confirmed operating accrual spike in the flagged years.
The two flagged years are the two accrual-manipulation years later charged by the SEC. The year between them scored Low — correctly. That discrimination is the point: a screen that cries wolf on every fast-growing hardware company is a siren, not an instrument. On the same production panel, twelve consecutive giant-cap company-years (Apple, JPMorgan, Church & Dwight, FY2021–24) all score Low, and the Critical tier's false-positive rate on clean names is 0.4%.
Set Super Micro beside a company of similar profitability that year — Church & Dwight — and the accrual anomaly is visible to the naked eye once you know where to look:
| Metric · FY2023 | Church & Dwight | Super Micro | Read |
|---|---|---|---|
| Net income | +$756M | +$640M | Both look profitable |
| Operating cash flow | +$1,044M | −$640M | One profit produced negative cash |
| Total accruals (NI − OCF) | −$288M | +$1,280M | $1.28B of earnings without cash |
| Cash conversion (OCF / NI) | 1.38× | −1.00× | Healthy vs inverted |
| TATA (accruals / assets) | −0.034 | +0.270 | 27% of the balance sheet is accrual |
| Beneish M-Score | −2.56 · clean | −0.79 · flag | Crosses the −1.78 manipulation line |
Computed from FY2023 10-K figures. Across FY2022–23 combined, Super Micro reported ~$1.09B of net income against ~−$1.13B of operating cash flow — a ~$2.2B gap between earnings and cash.
Accruals are where legitimate accounting judgment lives — and where income inflation hides when it happens. A large, persistent gap between profit and operating cash is not proof of anything. It is a question the filings are asking, and the one pattern this screen is built to rank companies by.
Anyone screening with AAAS had the FY2022 flag before the short report, the resignation, and the delisting scare. That is the entire value proposition of systematic screening: the filings speak years before the headlines do.
Super Micro's special committee reported finding no evidence of fraud or misconduct, and the company's filings were subsequently brought current. We include that fact deliberately, in the case study we lead with.
An accrual-anomaly flag is a reason to investigate, never a conclusion. High accruals have legitimate causes — acquisitions, hypergrowth, classification changes. AAAS exists to tell you where to spend investigation hours, and it is calibrated so that attention is not wasted: in out-of-time testing, the top decile of scores contains 79.3% of flagged company-years (7.9× random), the top fifth contains 97.2%, and the Critical tier runs 72.7% precision with a 0.4% false-positive rate on clean names.
Performance figures are measured against pattern-based income-inflating labels, since enforcement actions lag conduct by years — stated here the same way it is stated in our evidence pack.
AAAS runs six forensic layers — Beneish, Altman, Dechow, Jones discretionary accruals, Roychowdhury REM, SOM clustering — under an ML ensemble trained on 938 SEC enforcement cases across a 207,457 company-year panel. Upload one to five annual filings as CSV; get the score, the layers, the full show-the-math audit trail, and an auditor investigation checklist phrased per AU-C 240 / AS 2401.