01 · cash conversion
$6.83B of accruals sat between earnings and cash.
Total accruals (net income minus operating cash flow) were positive $6.83B. Cash conversion was −3.95×.
+$6.83BNI − OCF
FY2019
FY2019
Forensic case study · Archer-Daniels-Midland · production engine v17_71
Profit said yes.
Cash said no.
In FY2019 Archer-Daniels-Midland reported $1.38 billion of net income. Operating cash flow was negative $5.45 billion. AAAS scored the filing 69.3 / 100 — Critical. Seven years later the SEC settled accounting and disclosure fraud charges for $40 million.
Computed from filed financial statements · show-the-math audit trail · screening signal, not an accusation
Production engine v17_71
Filed-statement inputs
Scope disclosed
AAAS // filing x-ray
FY2019 · forensic composite
Filed statements → six forensic layers → ensemble
69.3/100
Critical
Reported net income+$1.38BProfitable on the income statement
Operating cash flow−$5.45BOperations consumed cash
Forensic layers3 / 4 classical rules flagged
BeneishM-Score −1.74
Altmanrule flag
DechowF-Score 1.29
Jones DA0.223
REM0.700
SOMStandard
207,457 company-yearstraining / calibration panel
938 SEC enforcement company-yearsconfirmed enforcement observations in panel
Six forensic layersclassic models + ML ensemble
Every value traceablefinancial input → calculation → verdict
The filing contradiction
A profitable year can still ask an uncomfortable cash question.
AAAS does not declare fraud. It ranks unusual reporting patterns so an analyst, auditor, lender, investor or board knows where verification should begin.
02 · discretionary accruals
Jones DA reached 0.223.
The discretionary-accrual reading sat above the production panel’s 95th percentile of 0.179.
0.223Jones
discretionary accruals
discretionary accruals
03 · multi-signal convergence
The warning was concentrated, not universal.
Beneish, Altman and the accrual interaction contributed rule flags, while several broad indicators remained quiet. That distinction matters.
3 / 4classical
rule flags
rule flags
Interactive evidence chain
Don’t take the score on faith. Follow how the question forms.
AAAS is most useful when a user can move from a headline anomaly to the exact accounting question — and then to the evidence that could weaken or support it. Click through the chain.
AAAS // evidence chainFY2019
Step 01 · observe
The income statement says profit. Cash flow says the opposite.
The first signal needs no model interpretation: reported net income was positive while operating cash flow was deeply negative.
Net income+$1.38B
Operating cash flow−$5.45B
ObservationProfitable year
→
ContradictionOperations consumed cash
→
QuestionWhat bridged earnings to cash?
Discipline: an anomaly is a question to investigate, not a conclusion.
Step 02 · quantify
The bridge is large enough to deserve decomposition.
Total accruals — net income minus operating cash flow — reached +$6.83B, while TATA reached +0.155 and Jones discretionary accruals 0.223.
Total accruals+$6.83B
Jones DA0.223
MeasureNI − OCF
→
ScaleTATA +0.155
→
ContextDA above 95th percentile
Interpretation: unusually accrual-heavy earnings increase the priority of reconciliation; they do not prove manipulation.
Step 03 · triangulate
The signal becomes more interesting when different methods disagree intelligently.
Beneish, Jones and other layers elevate the year while some broad indicators stay quieter. Concentrated disagreement is more informative than a dashboard where every light turns red.
Composite score69.3 · Critical
Classical rules3 / 4 flagged
Lens 1Beneish −1.74
+
Lens 2Jones DA 0.223
+
ContextQuiet signals retained
Design goal: show convergence and counter-evidence together so users can see where the model is confident — and where it is not.
Step 04 · verify
The score earns its keep only if it tells you what to inspect next.
The premium investigation layer converts the pattern into records and falsification work: accrual listings, estimate support and an independent operating-cash reconciliation.
Executive outputDecision framing
Technical outputVerification plan
ObtainYear-end accruals
→
TestEstimate support
→
ReconcileOCF to source records
Outcome: benign evidence should weaken the screening thesis; unsupported reconciling items should strengthen the need for deeper investigation.
Show the math
Every red flag should have a path back to the filing.
The persuasive part of AAAS is not a score in isolation. It is the chain from reported values, to a transparent calculation, to a screening interpretation, to the record that can challenge it.
Input
Filed statement values
Start with the company’s own reported financial statements. No narrative inference is needed to establish the initial contradiction.
reported values
Calculation
Transparent transforms
Cash conversion, total accruals and model inputs are computed from those values rather than inserted as opaque labels.
NI − OCF · OCF ÷ NI
Interpretation
Forensic lenses
Independent methods ask different questions about accrual quality, reporting patterns, operating behavior and context.
signal ≠ conclusion
Challenge
Records decide
The output ends with what to obtain, reconcile and test — including evidence that should weaken the screening thesis.
verify · falsify · document
InputFormulaSignalCounter-evidenceDecision
Four filings · four verdicts
The screen scored the filings, not the headlines.
Two consecutive Critical years were followed by two Low years when earnings became cash-backed. A useful screen has to know when not to shout.
FY 2018
69.1
Critical
NI +$1.81B · OCF −$4.78B · DA 0.198 · 3 flags
FY 2019
69.3
Critical
NI +$1.38B · OCF −$5.45B · DA 0.223 · 3 flags
FY 2021
10.1
Low
NI +$2.71B · OCF +$6.60B · cash-backed earnings
FY 2022
14.9
Low
NI +$4.34B · OCF +$3.48B · no classical flags
Across FY2018–19: $3.2B of reported net income versus −$10.2B of operating cash flow — a $13.4B two-year gap between earnings and cash.
Trajectory view
Watch the signal collapse when cash backing returns.
The chart is generated from the four filing cards above. Select a filing to bring its existing evidence into focus; nothing is re-entered or re-scored here.
Select a filing above to inspect the trajectory.
69.3Critical
Beneish−1.74 · flag
Jones DA0.223 · elevated
Dechow1.29 · elevated
SOMStandard cluster
RoychowdhuryREM 0.700
Altmanfinancial-health flag
Forensic X-ray
One score. Six different questions.
The value is not the red badge. The value is seeing which independent lenses agree, which stay quiet, and what records would resolve the disagreement.
Accrual qualityDoes profit depend unusually heavily on non-cash accounting?
Financial-statement patternDo classic manipulation-risk components cross screening thresholds?
Real operating behaviorDo cash-flow and production patterns look abnormal?
Behavioral contextIs the company broadly abnormal, or is the warning concentrated?
The anomaly in plain sight
Set beside a clean filing, the difference becomes physical.
Both companies reported profits. Only one converted that profit into operating cash.
Metric
Church & Dwight FY2023
ADM · FY2019
Read
Net income
+$0.76B
+$1.38B
Both look profitable
Operating cash flow
+$1.03B
−$5.45B
Profit did not convert to cash
Total accruals (NI − OCF)
−$0.28B
+$6.83B
Large earnings / cash gap
Cash conversion (OCF ÷ NI)
1.36×
−3.95×
Healthy versus inverted
TATA (accruals ÷ assets)
−0.032
+0.155
Accrual-heavy balance sheet
Jones discretionary accruals
0.066
0.223
Above the panel 95th percentile (0.179)
Beneish M-Score
−2.30 · clean
−1.74 · flag
Crosses −1.78 screening line
Classical rule flags
0 of 4
3 of 4
Signals converge
Both columns are computed by the shipped AAAS production engine (v17_71) from each company's own filed annual statements — Archer-Daniels-Midland Company FY2019 and Church & Dwight Co., Inc. FY2023. A large accrual gap is a reason to investigate, not proof of manipulation.
Beyond the score
The output is designed to move from signal to decision.
The premium layer separates the executive question from the forensic work. One report tells a Board what deserves attention. The other turns that concern into competing hypotheses, counter-evidence and tests.
AAAS · forensic accounting advisory
Executive Advisory Memo
What the Numbers Are Telling You
The bottom line
The FY2019 reading is outside the normal range and should be treated as a verify-the-past, fund-the-future matter, not a routine monitoring item.
Why the Board is here
The concern is concentrated rather than uniform: earnings quality and cash conversion require support, while several broad indicators remain quiet. The Board’s question is whether FY2019 earnings, balance-sheet movements and cash position reconcile to supported business events.
Decision framing
The report separates what appears concerning from what the evidence does not establish, then identifies the records that would materially weaken or support the screening thesis.
Preview adapted from the AAAS Premium Board Memo generated for Archer-Daniels-Midland FY2019.
AAAS · technical forensic investigation
Investigator's Thesis
Earnings-Quality Investigation — Working Hypotheses & Verification Plan
The surviving explanation is a moderate-confidence screening hypothesis of an earnings-quality unwind or related accrual and real-activity pressure, accompanied by material liquidity and cash-conversion pressure.
Competing hypotheses
MODERATEH1 · Earnings-quality unwind through accrual realization and real-activity pressureEvidence for · counter-evidence · missing evidence · falsification test
MODERATE-LOWH2 · Benign working-capital and cost-structure shiftTest whether ordinary operating explanations reconcile the cash and accrual pattern
LOWH3 · Current-period income inflation through revenue or receivables timingNormal/quiet sales and receivables readings weaken this as the sole explanation
Falsification first
Obtain the year-end accruals listing, decompose discretionary accruals, test estimates, and reconcile operating cash flow to independently supported receipts, disbursements and working-capital records.
Preview adapted from the AAAS Premium Technical Investigation generated for Archer-Daniels-Midland FY2019.
Falsification first
A persuasive tool should tell you how it could be wrong.
The strongest product claim is not “trust the model.” It is “here is the evidence that would change the interpretation.” Open each item to see what a serious investigation should try to establish.
Why this matters: this is the bridge from AI output to professional judgment. The screen prioritizes attention; the records decide the explanation.
Could ordinary working-capital timing explain the cash gap?
Reconcile receivables, inventory, payables and other operating working-capital movements to source schedules and subsequent cash realization. A coherent, documented operating explanation would weaken the anomaly thesis.
Could year-end estimates explain the discretionary-accrual signal?
Decompose major accruals and estimates, inspect support and subsequent reversals, and determine whether the unusual accrual level follows supported business events rather than unsupported income effects.
Could the negative operating cash flow be classification rather than economics?
Reconcile operating cash flow to independently supported receipts, disbursements and classification records. A clean reconciliation can materially change how the headline gap should be read.
What would increase concern rather than reduce it?
Unexplained reconciling items, unsupported estimates, weak documentation, or reversals inconsistent with the stated operating explanation would increase the case for deeper audit or forensic work.
The honesty clause
A serious tool shows you its blind spot.
This screen did not flag every year covered by the SEC's order against ADM.
The SEC order covers adjustments in 2019, 2021 and 2022. AAAS scored FY2019 Critical — but FY2021 and FY2022 Low. That fact belongs on the page, not in fine print.
ADM reported that the corrections concerned intersegment profit shifting between its business segments and had no impact on its consolidated balance sheet, earnings or cash flows. AAAS screens consolidated accrual anomalies. A transfer that nets to zero at the consolidated level is outside what this instrument measures.
What AAAS did flagTwo extraordinary consolidated accrual years, including FY2019.
What AAAS did not claimThat it “caught the fraud” or detects every accounting mechanism.
Scope: a reason to investigate, never a conclusion of wrongdoing.
Calibration
A flag is only useful if attention is not wasted.
Measured out-of-time — on fiscal years beyond the training cutoff — the tier ladder increases sharply from Low to Critical.
0.953ROC-AUC · out-of-timeGini 0.906 · KS 0.794
79.3%of flagged company-years in top decile7.9× random concentration
72.7%Critical-tier precisionat 0.4% false-positive rate on clean names
0.4%Low-tier positive ratethe clean floor
Low< 26.3
Moderate26.3–31.8
Elevated31.8–40.3
High40.3–60.1
Critical≥ 60.1
Important: performance figures above are measured against pattern-based income-inflating labels, not enforcement outcomes. Against confirmed SEC enforcement cases, AAAS captures roughly 2× random in the top decile — lower because enforcement includes mechanisms this screen does not measure.
Audience lens
Same evidence. Different first question.
AAAS should not change its score depending on who is looking. What changes is the route through the evidence. Choose the perspective closest to yours.
Board / Audit Committee
Is the reporting pattern important enough to demand management evidence?
The governance route starts with business consequence, then moves to the records that could support or weaken the concern.
The useful question is not whether the score sounds alarming. It is whether management can reconcile the pattern to supported business events before a decision depends on it.
Start with the advisorySee how the signal is translated into a Board-level decision frame.Then challenge itInspect the evidence that could change the interpretation.
Run the screen on a company you oversee↗
Forensic / Internal Audit
Which records would explain the anomaly — and which would make it harder to explain?
The investigation route follows the chain from filed values to calculation, signal, counter-evidence and verification work.
A useful screen should shorten the distance between “this looks unusual” and “here is exactly what I need to obtain, reconcile and test.”
Follow the evidence chainMove from observation to quantification, triangulation and verification.Trace it to the filingSee how each interpretation retains a path back to reported values.
Use AAAS to prioritize the next investigation↗
Credit / Lending
Are reported earnings behaving like cash-generating performance?
The credit route begins with cash conversion and persistence, then uses the broader forensic context to decide where deeper diligence is justified.
The screen does not replace a credit opinion. It can identify the company-years where the earnings story deserves more reconciliation before you rely on it.
Start with cash conversionSee the contradiction directly in the filed statements.Keep the boundary visibleReview what the consolidated screen can and cannot detect.
Screen a borrower before the next review↗
Investor / Research
Which filing deserves more diligence before you accept the headline narrative?
The research route uses AAAS as an attention-ranking instrument: surface the unusual year, inspect the evidence, then decide whether the explanation survives scrutiny.
The advantage is not hindsight. It is having a repeatable way to ask the same forensic questions of a company whose ending you do not already know.
Inspect the derivationFollow the signal back to the filed values and calculations.Look for disconfirming evidenceSee what would weaken the screening thesis.
Choose a company and test the filing yourself↗
One instrument, one score. The audience lens changes navigation — never the underlying evidence.
69.3
You followed one filing from contradiction → evidence → verification.Now choose a company where you do not already know the ending.
Your turn
Give AAAS the statements. Let the numbers argue back.
Choose a company, load the filed data, and see where the financial statements stop behaving normally — then inspect exactly why the screen thinks the year deserves attention.
Six forensic layersShow-the-math evidenceBoard + investigator interpretationNo silent imputation
Launch the live forensic screen
Which company would you test first?
Upload one to five annual filings as CSV and inspect the score, layer verdicts, traceable calculations and investigation checklist.
Try AAAS now↗